flet ui (experimental)

.github/workflows/test_pip.yml

@@ -1,9 +1,7 @@
-name: CI
-run-name: CI
-
+name: pip install
+run-name: pip install
 on: 
   push:
-    branches: [main]
     paths:
       - '**.py'
       - 'pyproject.toml'
@@ -13,41 +11,23 @@ on:
       - 'pyproject.toml'
       - '.github/workflows/**'
   workflow_dispatch:
-
 jobs:
-  test:
+  install-and-run:
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-14, windows-latest]
+        os: [ubuntu-latest, macos-latest, windows-latest]
         python-version: ['3.12']
       fail-fast: false
+    continue-on-error: true
     runs-on: ${{ matrix.os }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v7
-
+        uses: actions/checkout@v4
       - name: Set up Python
-        uses: actions/setup-python@v6
+        uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-
-      - name: Install uv
-        uses: astral-sh/setup-uv@v8.3.1
-        with:
-          enable-cache: true
-          prune-cache: false
-          cache-dependency-glob: pyproject.toml
-
-      - name: Install system dependencies (Ubuntu)
-        if: runner.os == 'Linux'
-        run: sudo apt-get update && sudo apt-get install -y libegl1
-
-      - name: Install dependencies
-        run: uv pip install --system .[dev]
-        env:
-          UV_LINK_MODE: copy
-
-      - name: Run tests
-        env:
-          QT_QPA_PLATFORM: offscreen
-        run: pytest tests/ -v --tb=short
+      - name: Install from repository
+        run: python -m pip install .
+      #- name: Run abogen
+      #  run: abogen

.github/workflows/test_publish.yml

@@ -18,7 +18,7 @@ jobs:
   build:
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v7
+    - uses: actions/checkout@v4
 
     - name: Login to Github Container Registry
       # Only if we need to push an image

abogen/constants.py

@@ -63,6 +63,64 @@ SUPPORTED_INPUT_FORMATS = [
 # 384 if self.lang_code in 'ab':
 SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION = list(LANGUAGE_DESCRIPTIONS.keys())
 
+# Voice and sample text constants
+VOICES_INTERNAL = [
+    "af_alloy",
+    "af_aoede",
+    "af_bella",
+    "af_heart",
+    "af_jessica",
+    "af_kore",
+    "af_nicole",
+    "af_nova",
+    "af_river",
+    "af_sarah",
+    "af_sky",
+    "am_adam",
+    "am_echo",
+    "am_eric",
+    "am_fenrir",
+    "am_liam",
+    "am_michael",
+    "am_onyx",
+    "am_puck",
+    "am_santa",
+    "bf_alice",
+    "bf_emma",
+    "bf_isabella",
+    "bf_lily",
+    "bm_daniel",
+    "bm_fable",
+    "bm_george",
+    "bm_lewis",
+    "ef_dora",
+    "em_alex",
+    "em_santa",
+    "ff_siwis",
+    "hf_alpha",
+    "hf_beta",
+    "hm_omega",
+    "hm_psi",
+    "if_sara",
+    "im_nicola",
+    "jf_alpha",
+    "jf_gongitsune",
+    "jf_nezumi",
+    "jf_tebukuro",
+    "jm_kumo",
+    "pf_dora",
+    "pm_alex",
+    "pm_santa",
+    "zf_xiaobei",
+    "zf_xiaoni",
+    "zf_xiaoxiao",
+    "zf_xiaoyi",
+    "zm_yunjian",
+    "zm_yunxi",
+    "zm_yunxia",
+    "zm_yunyang",
+]
+
 # Voice and sample text mapping
 SAMPLE_VOICE_TEXTS = {
     "a": "This is a sample of the selected voice.",

abogen/frontend/__init__.py

@@ -0,0 +1,8 @@
+"""
+Abogen Flet Frontend Package.
+
+This package provides a unified, dual-target (desktop + web) user interface
+for the Abogen audiobook generation application, built with the Flet framework.
+"""
+
+__all__ = ["main"]

abogen/frontend/components/__init__.py

@@ -0,0 +1,32 @@
+"""Components sub-package."""
+from .widgets import (
+    resolve_icon,
+    build_drop_zone,
+    build_log_terminal,
+    log_entry,
+    build_progress_row,
+    build_primary_button,
+    build_secondary_button,
+    build_card,
+    build_section_header,
+    build_status_badge,
+    labelled_row,
+    show_snack,
+    build_divider,
+)
+
+__all__ = [
+    "build_drop_zone",
+    "resolve_icon",
+    "build_log_terminal",
+    "log_entry",
+    "build_progress_row",
+    "build_primary_button",
+    "build_secondary_button",
+    "build_card",
+    "build_section_header",
+    "build_status_badge",
+    "labelled_row",
+    "show_snack",
+    "build_divider",
+]

abogen/frontend/components/widgets.py

@@ -0,0 +1,630 @@
+"""
+Reusable UI components for the Abogen Flet frontend.
+
+Each function in this module returns a standalone Flet control or small
+widget tree.  Components read the current palette from the page's theme
+mode and should not hold any mutable state themselves – state lives in the
+session's ``AppState`` object.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Callable, List, Optional
+
+import flet as ft
+
+from ..utils.theme import get_palette, RADIUS_MD, RADIUS_SM, SPACE_SM, SPACE_MD, SPACE_LG
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def resolve_icon(icon: Any) -> Any:
+    """Convert a snake_case icon name to Flet IconData when possible."""
+    if isinstance(icon, str):
+        return getattr(ft.Icons, icon.upper(), icon)
+    return icon
+
+
+# ---------------------------------------------------------------------------
+# Drop-zone (file input area)
+# ---------------------------------------------------------------------------
+
+
+def build_drop_zone(
+    *,
+    on_pick: Callable[[], None],
+    label: str = "Drag & drop your file here or click to browse",
+    sub_label: str = "Supports: .txt · .epub · .pdf · .md · .srt · .ass · .vtt",
+    accent: bool = False,
+    error: bool = False,
+    filename: Optional[str] = None,
+    file_size: Optional[str] = None,
+    char_count: Optional[str] = None,
+    page: Optional[ft.Page] = None,
+) -> ft.GestureDetector:
+    """
+    Build an interactive file drop-zone widget.
+
+    The zone shows a dashed border and centred instructions by default,
+    switching to an 'active' green style when a file is loaded and a red
+    style when an error has occurred.
+
+    Args:
+        on_pick: Callback invoked when the user clicks or activates the zone.
+        label: Primary instruction text.
+        sub_label: Secondary hint text shown beneath the label.
+        accent: When True, renders the 'active/success' green style.
+        error: When True, renders the 'error/red' style.
+        filename: When provided, replaces the instruction text with file info.
+        file_size: Human-readable file size to display alongside the filename.
+        char_count: Character count to display alongside file info.
+        page: The current Flet ``Page``; used to derive the active palette.
+
+    Returns:
+        A ``ft.GestureDetector`` wrapping the visual drop-zone container.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    p = get_palette(page) if page else None
+
+    # Colour scheme
+    if error:
+        border_color = "#e84e3c" if dark else "#c0392b"
+        bg_color = "#1a0a08" if dark else "#fff5f5"
+        text_color = "#e84e3c" if dark else "#c0392b"
+        icon_name = "error_outline"
+    elif accent:
+        border_color = "#42ad4a" if dark else "#2e9437"
+        bg_color = "#091810" if dark else "#f0fff1"
+        text_color = "#42ad4a" if dark else "#2e9437"
+        icon_name = "check_circle_outline"
+    else:
+        border_color = "#3a4466" if dark else "#a8b4d0"
+        bg_color = "#151928" if dark else "#f7f8fd"
+        text_color = "#9ba3b8" if dark else "#5a6172"
+        icon_name = "upload_file"
+
+    if filename:
+        # Compact file-info display
+        info_rows: List[ft.Control] = [
+            ft.Row(
+                [
+                    ft.Icon(resolve_icon("insert_drive_file"), color=text_color, size=28),
+                    ft.Column(
+                        [
+                            ft.Text(
+                                filename,
+                                weight=ft.FontWeight.W_600,
+                                size=13,
+                                color=text_color,
+                                no_wrap=False,
+                                max_lines=2,
+                                overflow=ft.TextOverflow.ELLIPSIS,
+                            ),
+                        ],
+                        tight=True,
+                        expand=True,
+                    ),
+                ],
+                alignment=ft.MainAxisAlignment.CENTER,
+                spacing=SPACE_SM,
+            )
+        ]
+        if file_size or char_count:
+            chips: List[ft.Control] = []
+            if file_size:
+                chips.append(
+                    ft.Text(f"📄 {file_size}", size=11, color=text_color, italic=True)
+                )
+            if char_count:
+                chips.append(
+                    ft.Text(f"🔤 {char_count} chars", size=11, color=text_color, italic=True)
+                )
+            info_rows.append(
+                ft.Row(chips, alignment=ft.MainAxisAlignment.CENTER, spacing=SPACE_MD)
+            )
+        content = ft.Column(
+            info_rows,
+            alignment=ft.MainAxisAlignment.CENTER,
+            horizontal_alignment=ft.CrossAxisAlignment.CENTER,
+            spacing=SPACE_SM,
+        )
+    else:
+        content = ft.Column(
+            [
+                ft.Icon(resolve_icon(icon_name), size=48, color=border_color, opacity=0.8),
+                ft.Text(
+                    label,
+                    size=14,
+                    weight=ft.FontWeight.W_500,
+                    color=text_color,
+                    text_align=ft.TextAlign.CENTER,
+                ),
+                ft.Text(
+                    sub_label,
+                    size=11,
+                    color=text_color,
+                    opacity=0.6,
+                    text_align=ft.TextAlign.CENTER,
+                ),
+            ],
+            alignment=ft.MainAxisAlignment.CENTER,
+            horizontal_alignment=ft.CrossAxisAlignment.CENTER,
+            spacing=SPACE_SM,
+        )
+
+    inner = ft.Container(
+        content=content,
+        border=ft.Border.all(2, border_color),
+        border_radius=RADIUS_MD,
+        bgcolor=bg_color,
+        padding=ft.Padding.all(SPACE_LG),
+        height=160,
+        alignment=ft.Alignment.CENTER,
+        expand=True,
+    )
+
+    return ft.GestureDetector(
+        content=ft.Row([inner], spacing=0),
+        on_tap=lambda _: on_pick(),
+        mouse_cursor=ft.MouseCursor.CLICK,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Log terminal
+# ---------------------------------------------------------------------------
+
+
+def build_log_terminal(
+    *,
+    ref: Optional[ft.Ref] = None,
+    max_height: int = 260,
+    page: Optional[ft.Page] = None,
+) -> ft.Container:
+    """
+    Build a scrollable, read-only log terminal widget.
+
+    Args:
+        ref: Optional ``ft.Ref[ft.ListView]`` to bind the inner list-view so
+             callers can append entries programmatically.
+        max_height: Maximum pixel height before vertical scrolling activates.
+        page: Current Flet ``Page`` for palette derivation.
+
+    Returns:
+        A styled ``ft.Container`` wrapping a ``ft.ListView``.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    bg = "#0d1117" if dark else "#f8f9fc"
+    text_color = "#b0b8cc" if dark else "#3d4358"
+    border_color = "#252a38" if dark else "#dce0ea"
+
+    list_view = ft.ListView(
+        expand=True,
+        auto_scroll=True,
+        spacing=1,
+        padding=ft.Padding.all(SPACE_SM),
+    )
+    if ref is not None:
+        ref.current = list_view
+
+    return ft.Container(
+        content=list_view,
+        bgcolor=bg,
+        border=ft.Border.all(1, border_color),
+        border_radius=RADIUS_SM,
+        height=max_height,
+        clip_behavior=ft.ClipBehavior.HARD_EDGE,
+    )
+
+
+def log_entry(message: str, level: str = "info", page: Optional[ft.Page] = None) -> ft.Text:
+    """
+    Create a single log-line ``ft.Text`` widget with appropriate colour coding.
+
+    Args:
+        message: The log message string.
+        level: Severity string: ``'info'``, ``'success'``, ``'error'``,
+               ``'warning'``, ``'debug'``, ``'critical'``.
+        page: Current Flet ``Page`` for dark/light mode detection.
+
+    Returns:
+        A styled ``ft.Text`` control.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    palette: dict[str, str] = {
+        "info": "#9ba3b8" if dark else "#5a6172",
+        "success": "#42ad4a" if dark else "#2e9437",
+        "error": "#e84e3c" if dark else "#c0392b",
+        "warning": "#f5a623" if dark else "#d4870a",
+        "debug": "#5a6172" if dark else "#9ba3b8",
+        "critical": "#ff5722",
+        "trace": "#4e5568" if dark else "#b0b8cc",
+    }
+    color = palette.get(level.lower(), palette["info"])
+    return ft.Text(message, size=12, color=color, selectable=True, no_wrap=False)
+
+
+# ---------------------------------------------------------------------------
+# Progress row
+# ---------------------------------------------------------------------------
+
+
+def build_progress_row(
+    *,
+    progress_value: float = 0.0,
+    etr_text: str = "",
+    page: Optional[ft.Page] = None,
+) -> ft.Column:
+    """
+    Build a progress-bar + ETR-label column.
+
+    Args:
+        progress_value: Float in [0.0, 1.0].
+        etr_text: Pre-formatted estimated-time-remaining string.
+        page: Current ``Page`` for palette derivation.
+
+    Returns:
+        A ``ft.Column`` containing the progress bar and label.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    fill = "#5b8af5" if dark else "#3a5fc4"
+    bg = "#1e2230" if dark else "#e4e8f0"
+
+    bar = ft.ProgressBar(
+        value=progress_value,
+        color=fill,
+        bgcolor=bg,
+        height=8,
+        border_radius=ft.BorderRadius.all(4),
+        expand=True,
+    )
+    label = ft.Text(
+        etr_text,
+        size=11,
+        color="#9ba3b8" if dark else "#5a6172",
+        text_align=ft.TextAlign.CENTER,
+    )
+    return ft.Column(
+        [bar, label],
+        spacing=SPACE_SM,
+        horizontal_alignment=ft.CrossAxisAlignment.CENTER,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Primary action button
+# ---------------------------------------------------------------------------
+
+
+def build_primary_button(
+    text: str,
+    *,
+    icon: Optional[str] = None,
+    on_click: Optional[Callable] = None,
+    disabled: bool = False,
+    width: Optional[int] = None,
+    page: Optional[ft.Page] = None,
+) -> ft.ElevatedButton:
+    """
+    Build a prominent, styled primary action button.
+
+    Args:
+        text: Button label.
+        icon: Optional Flet icon name (e.g. ``'play_arrow'``).
+        on_click: Click callback.
+        disabled: Whether the button is non-interactive.
+        width: Optional fixed pixel width.
+        page: Current ``Page`` for accent colour derivation.
+
+    Returns:
+        A styled ``ft.ElevatedButton``.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    bg = "#5b8af5" if dark else "#3a5fc4"
+    on_bg = "#ffffff"
+
+    style = ft.ButtonStyle(
+        bgcolor={
+            ft.ControlState.DEFAULT: bg,
+            ft.ControlState.HOVERED: "#3a5fc4" if dark else "#2a4fae",
+            ft.ControlState.DISABLED: "#2a2f3f" if dark else "#c0c8d8",
+        },
+        color={
+            ft.ControlState.DEFAULT: on_bg,
+            ft.ControlState.DISABLED: "#4e5568" if dark else "#9ba3b8",
+        },
+        elevation={"default": 2, "hovered": 4},
+        padding=ft.Padding.symmetric(horizontal=SPACE_LG, vertical=SPACE_MD),
+        shape=ft.RoundedRectangleBorder(radius=RADIUS_SM),
+        animation_duration=150,
+    )
+
+    return ft.ElevatedButton(
+        content=text,
+        icon=resolve_icon(icon),
+        on_click=on_click,
+        disabled=disabled,
+        width=width,
+        style=style,
+        height=48,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Secondary / ghost button
+# ---------------------------------------------------------------------------
+
+
+def build_secondary_button(
+    text: str,
+    *,
+    icon: Optional[str] = None,
+    on_click: Optional[Callable] = None,
+    disabled: bool = False,
+    page: Optional[ft.Page] = None,
+) -> ft.OutlinedButton:
+    """
+    Build a secondary outlined button.
+
+    Args:
+        text: Button label.
+        icon: Optional Flet icon name.
+        on_click: Click callback.
+        disabled: Whether the button is non-interactive.
+        page: Current ``Page`` for border colour derivation.
+
+    Returns:
+        A styled ``ft.OutlinedButton``.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    border_clr = "#3a4466" if dark else "#a8b4d0"
+    text_clr = "#e8eaf0" if dark else "#1a1d27"
+
+    style = ft.ButtonStyle(
+        side={
+            ft.ControlState.DEFAULT: ft.BorderSide(1.5, border_clr),
+            ft.ControlState.HOVERED: ft.BorderSide(1.5, "#5b8af5" if dark else "#3a5fc4"),
+        },
+        color={
+            ft.ControlState.DEFAULT: text_clr,
+            ft.ControlState.HOVERED: "#5b8af5" if dark else "#3a5fc4",
+            ft.ControlState.DISABLED: "#4e5568" if dark else "#9ba3b8",
+        },
+        padding=ft.Padding.symmetric(horizontal=SPACE_LG, vertical=SPACE_MD),
+        shape=ft.RoundedRectangleBorder(radius=RADIUS_SM),
+        animation_duration=150,
+    )
+
+    return ft.OutlinedButton(
+        content=text,
+        icon=resolve_icon(icon),
+        on_click=on_click,
+        disabled=disabled,
+        style=style,
+        height=44,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Section card
+# ---------------------------------------------------------------------------
+
+
+def build_card(
+    content: ft.Control,
+    *,
+    padding: int = SPACE_LG,
+    page: Optional[ft.Page] = None,
+) -> ft.Container:
+    """
+    Wrap a control in a styled card container.
+
+    Args:
+        content: The child control to embed.
+        padding: Internal padding in pixels.
+        page: Current ``Page`` for palette derivation.
+
+    Returns:
+        A styled ``ft.Container``.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    bg = "#181b23" if dark else "#ffffff"
+    border_clr = "#2c3147" if dark else "#dce0ea"
+
+    return ft.Container(
+        content=content,
+        bgcolor=bg,
+        border=ft.Border.all(1, border_clr),
+        border_radius=RADIUS_MD,
+        padding=ft.Padding.all(padding),
+        shadow=ft.BoxShadow(
+            spread_radius=0,
+            blur_radius=12,
+            color=ft.Colors.with_opacity(0.12 if dark else 0.06, ft.Colors.BLACK),
+            offset=ft.Offset(0, 2),
+        ),
+    )
+
+
+# ---------------------------------------------------------------------------
+# Section header
+# ---------------------------------------------------------------------------
+
+
+def build_section_header(
+    title: str,
+    *,
+    subtitle: Optional[str] = None,
+    icon: Optional[str] = None,
+    page: Optional[ft.Page] = None,
+) -> ft.Row:
+    """
+    Build a consistent section header row with an optional icon.
+
+    Args:
+        title: Section heading text.
+        subtitle: Optional explanatory sub-text.
+        icon: Optional Flet icon name.
+        page: Current ``Page`` for palette derivation.
+
+    Returns:
+        A ``ft.Row`` containing the icon and text column.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    title_color = "#e8eaf0" if dark else "#1a1d27"
+    sub_color = "#9ba3b8" if dark else "#5a6172"
+    accent = "#5b8af5" if dark else "#3a5fc4"
+
+    children: List[ft.Control] = []
+    if icon:
+        children.append(ft.Icon(resolve_icon(icon), size=20, color=accent))
+
+    text_parts: List[ft.Control] = [
+        ft.Text(title, size=15, weight=ft.FontWeight.W_600, color=title_color)
+    ]
+    if subtitle:
+        text_parts.append(ft.Text(subtitle, size=11, color=sub_color))
+
+    children.append(
+        ft.Column(text_parts, spacing=1, tight=True, expand=True)
+    )
+
+    return ft.Row(children, spacing=SPACE_SM, vertical_alignment=ft.CrossAxisAlignment.START)
+
+
+# ---------------------------------------------------------------------------
+# Status badge
+# ---------------------------------------------------------------------------
+
+
+def build_status_badge(
+    label: str,
+    *,
+    variant: str = "info",
+    page: Optional[ft.Page] = None,
+) -> ft.Container:
+    """
+    Build a small status badge chip.
+
+    Args:
+        label: Badge text.
+        variant: Colour variant: ``'info'``, ``'success'``, ``'error'``,
+                 ``'warning'``, ``'neutral'``.
+        page: Current ``Page`` for theme derivation.
+
+    Returns:
+        A pill-shaped ``ft.Container``.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    palette = {
+        "info":    ("#1a2a5e" if dark else "#dde8ff", "#5b8af5" if dark else "#3a5fc4"),
+        "success": ("#0d2010" if dark else "#d4f4d7", "#42ad4a" if dark else "#2e9437"),
+        "error":   ("#2a0a08" if dark else "#ffe0dc", "#e84e3c" if dark else "#c0392b"),
+        "warning": ("#2a1a00" if dark else "#fff4d8", "#f5a623" if dark else "#d4870a"),
+        "neutral": ("#1e2230" if dark else "#edf0f5", "#9ba3b8" if dark else "#5a6172"),
+    }
+    bg, fg = palette.get(variant, palette["info"])
+
+    return ft.Container(
+        content=ft.Text(label, size=10, weight=ft.FontWeight.W_600, color=fg),
+        bgcolor=bg,
+        border_radius=999,
+        padding=ft.Padding.symmetric(horizontal=8, vertical=3),
+    )
+
+
+# ---------------------------------------------------------------------------
+# Labelled control row
+# ---------------------------------------------------------------------------
+
+
+def labelled_row(
+    label: str,
+    control: ft.Control,
+    *,
+    label_width: int = 200,
+    tooltip: Optional[str] = None,
+    page: Optional[ft.Page] = None,
+) -> ft.Row:
+    """
+    Lay a label and a control side-by-side in a consistent row.
+
+    Args:
+        label: Human-readable label text.
+        control: The UI control placed to the right of the label.
+        label_width: Fixed pixel width of the label column.
+        tooltip: Optional tooltip text on the label.
+        page: Current ``Page`` for palette derivation.
+
+    Returns:
+        A ``ft.Row`` with the label pinned to a fixed width.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    lbl_color = "#9ba3b8" if dark else "#5a6172"
+
+    lbl = ft.Text(label, size=13, color=lbl_color, weight=ft.FontWeight.W_500, width=label_width)
+    if tooltip:
+        lbl.tooltip = tooltip
+
+    return ft.Row(
+        [lbl, ft.Container(content=control, expand=True)],
+        alignment=ft.MainAxisAlignment.START,
+        vertical_alignment=ft.CrossAxisAlignment.CENTER,
+        spacing=SPACE_MD,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Snack-bar helper
+# ---------------------------------------------------------------------------
+
+
+def show_snack(
+    page: ft.Page,
+    message: str,
+    *,
+    error: bool = False,
+    duration: int = 3000,
+) -> None:
+    """
+    Display a brief snack-bar notification.
+
+    Args:
+        page: The Flet ``Page`` instance.
+        message: Text to display.
+        error: When True, colours the bar red instead of the default accent.
+        duration: Visible duration in milliseconds.
+    """
+    dark = page.theme_mode == ft.ThemeMode.DARK
+    bg = "#e84e3c" if error else ("#5b8af5" if dark else "#3a5fc4")
+    page.snack_bar = ft.SnackBar(
+        content=ft.Text(message, color="#ffffff", size=13),
+        bgcolor=bg,
+        duration=duration,
+        show_close_icon=True,
+        close_icon_color="#ffffff",
+    )
+    page.snack_bar.open = True
+    page.update()
+
+
+# ---------------------------------------------------------------------------
+# Divider helper
+# ---------------------------------------------------------------------------
+
+
+def build_divider(page: Optional[ft.Page] = None) -> ft.Divider:
+    """
+    Build a styled horizontal rule divider.
+
+    Args:
+        page: Current ``Page`` for palette derivation.
+
+    Returns:
+        A ``ft.Divider``.
+    """
+    dark = page is not None and page.theme_mode == ft.ThemeMode.DARK
+    return ft.Divider(color="#252a38" if dark else "#e8ebf2", height=1, thickness=1)

abogen/frontend/main.py

@@ -0,0 +1,365 @@
+"""
+Abogen Flet Frontend – main entry point.
+
+Run as desktop app:
+    python -m abogen.frontend.main
+
+Run as web app (binds to port 8080 by default):
+    python -m abogen.frontend.main --web --port 8080
+
+Architecture
+------------
+One ``ft.app()`` call launches the server.  For every new browser tab (or the
+desktop window) Flet invokes ``_app_entry(page)`` in its own coroutine, which
+creates a fresh ``AppState`` and wires together the navigation rail and views.
+This guarantees complete per-session isolation in multi-user web deployments.
+"""
+from __future__ import annotations
+
+import argparse
+import sys
+from pathlib import Path
+from typing import Optional
+
+import flet as ft
+
+from .state import AppState
+from .components import resolve_icon
+from .views.dashboard import DashboardView
+from .views.settings import SettingsView
+from .views.queue_view import QueueView
+from .utils.theme import make_theme, DARK, LIGHT, SPACE_SM, SPACE_MD, SPACE_LG, RADIUS_MD
+from abogen.constants import PROGRAM_NAME as APP_NAME
+
+
+# ---------------------------------------------------------------------------
+# Navigation destinations
+# ---------------------------------------------------------------------------
+
+_NAV_ITEMS = [
+    ("Convert", "swap_horiz", "swap_horiz"),
+    ("Queue",   "list_alt",   "list_alt"),
+    ("Settings", "settings",  "settings"),
+]
+
+_ASSETS_DIR = Path(__file__).resolve().parents[1] / "assets"
+
+
+def _build_sidebar_item(
+    *,
+    label: str,
+    icon: str,
+    selected: bool,
+    palette,
+    on_click,
+) -> ft.Container:
+    accent = palette.accent if selected else palette.text_secondary
+    bg = palette.sidebar_selected_bg if selected else palette.sidebar_bg
+    return ft.Container(
+        content=ft.Row(
+            [
+                ft.Icon(resolve_icon(icon), size=20, color=accent),
+                ft.Text(
+                    label,
+                    size=13,
+                    weight=ft.FontWeight.W_600 if selected else ft.FontWeight.W_500,
+                    color=accent,
+                ),
+            ],
+            spacing=SPACE_MD,
+            vertical_alignment=ft.CrossAxisAlignment.CENTER,
+        ),
+        bgcolor=bg,
+        border_radius=RADIUS_MD,
+        padding=ft.Padding.symmetric(horizontal=SPACE_MD, vertical=10),
+        ink=True,
+        on_click=on_click,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Per-session entry point
+# ---------------------------------------------------------------------------
+
+def _app_entry(page: ft.Page) -> None:
+    try:
+        # ── State ────────────────────────────────────────────────────────────
+        state = AppState()
+        state.load_from_config()
+
+        # ── Page basics ──────────────────────────────────────────────────────
+        page.title = APP_NAME
+        page.padding = 0
+        page.spacing = 0
+        page.bgcolor = DARK.bg_base
+        page.theme_mode = ft.ThemeMode.DARK
+        page.theme = make_theme(dark=True)
+        page.dark_theme = make_theme(dark=True)
+        page.fonts = {}
+        page.window.min_width = 520
+        page.window.min_height = 600
+        page.update()
+
+        # ── Content area ref ─────────────────────────────────────────────────
+        content_area = ft.Column(expand=True, spacing=0)
+        sidebar_body = ft.Column(spacing=SPACE_SM)
+        theme_button_host = ft.Container()
+        brand_title = ft.Text(
+            APP_NAME,
+            size=18,
+            weight=ft.FontWeight.W_700,
+            color=DARK.text_primary,
+        )
+        brand_fallback_icon = ft.Icon(resolve_icon("speaker_notes"), size=32, color=DARK.accent)
+        divider = ft.VerticalDivider(width=1, color=DARK.border)
+
+        # ── Views ────────────────────────────────────────────────────────────
+        dashboard_view = DashboardView(page, state)
+        settings_view = SettingsView(page, state)
+        queue_view = QueueView(page, state)
+
+        views = [
+            dashboard_view.build,
+            queue_view.build,
+            settings_view.build,
+        ]
+        _selected_index = [0]
+
+        def _refresh_sidebar() -> None:
+            dark = page.theme_mode == ft.ThemeMode.DARK
+            pal = DARK if dark else LIGHT
+            sidebar_body.controls = [
+                _build_sidebar_item(
+                    label=label,
+                    icon=icon,
+                    selected=index == _selected_index[0],
+                    palette=pal,
+                    on_click=lambda _, i=index: _navigate(i),
+                )
+                for index, (label, icon, _) in enumerate(_NAV_ITEMS)
+            ]
+            sidebar.bgcolor = pal.sidebar_bg
+            divider.color = pal.border
+            brand_title.color = pal.text_primary
+            brand_fallback_icon.color = pal.accent
+            theme_button_host.content = ft.Container(
+                content=ft.Icon(
+                    resolve_icon("dark_mode" if dark else "light_mode"),
+                    size=20,
+                    color=pal.text_secondary,
+                ),
+                tooltip="Toggle theme",
+                border_radius=RADIUS_MD,
+                padding=8,
+                ink=True,
+                on_click=lambda _: _toggle_theme(page, _refresh_sidebar),
+            )
+
+        def _navigate(index: int) -> None:
+            _selected_index[0] = index
+            content_area.controls.clear()
+            built = views[index]()
+            content_area.controls.append(
+                ft.Container(
+                    content=built,
+                    expand=True,
+                    padding=ft.Padding.symmetric(horizontal=SPACE_LG, vertical=SPACE_LG),
+                )
+            )
+            _refresh_sidebar()
+            page.update()
+
+        # ── Sidebar ──────────────────────────────────────────────────────────
+        pal = DARK
+        sidebar = ft.Container(
+            width=220,
+            bgcolor=pal.sidebar_bg,
+            padding=ft.Padding.all(SPACE_MD),
+            content=ft.Column(
+                [
+                    ft.Container(
+                        content=ft.Row(
+                            [
+                                ft.Image(
+                                    src="icon.png",
+                                    width=36,
+                                    height=36,
+                                    fit=ft.BoxFit.CONTAIN,
+                                    error_content=brand_fallback_icon,
+                                ),
+                                brand_title,
+                            ],
+                            spacing=SPACE_MD,
+                            vertical_alignment=ft.CrossAxisAlignment.CENTER,
+                        ),
+                        padding=ft.Padding.only(top=SPACE_SM, bottom=SPACE_LG),
+                    ),
+                    sidebar_body,
+                    ft.Container(expand=True),
+                    ft.Row([theme_button_host], alignment=ft.MainAxisAlignment.END),
+                ],
+                expand=True,
+                spacing=SPACE_SM,
+            ),
+        )
+        _refresh_sidebar()
+
+        # ── Page handle for pubsub (queue → dashboard) ───────────────────────
+        def _handle_pubsub(topic: str) -> None:
+            if topic == "start_queue":
+                _navigate(0)
+
+        page.pubsub.subscribe(_handle_pubsub)
+
+        # ── Layout ────────────────────────────────────────────────────────────
+        page.add(
+            ft.Row(
+                [
+                    sidebar,
+                    divider,
+                    ft.Container(content=content_area, expand=True),
+                ],
+                expand=True,
+                spacing=0,
+                vertical_alignment=ft.CrossAxisAlignment.START,
+            )
+        )
+
+        # Show dashboard by default
+        _navigate(0)
+        page.update()
+    except Exception as e:
+        import traceback
+        traceback.print_exc()
+        print(f"ERROR IN _app_entry: {e}")
+        raise
+
+
+def _toggle_theme(page: ft.Page, refresh_sidebar) -> None:
+    """Switch between dark and light theme modes."""
+    if page.theme_mode == ft.ThemeMode.DARK:
+        page.theme_mode = ft.ThemeMode.LIGHT
+        page.bgcolor = LIGHT.bg_base
+    else:
+        page.theme_mode = ft.ThemeMode.DARK
+        page.bgcolor = DARK.bg_base
+
+    page.theme = make_theme(page.theme_mode == ft.ThemeMode.DARK)
+    refresh_sidebar()
+    page.update()
+
+
+# ---------------------------------------------------------------------------
+# CLI helpers & entry point
+# ---------------------------------------------------------------------------
+
+def _is_port_free(host: str, port: int) -> bool:
+    import socket
+    try:
+        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+            s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+            s.bind((host, port))
+            return True
+    except OSError:
+        return False
+
+
+def _find_free_port(host: str, start_port: int) -> int:
+    import socket
+    port = start_port
+    while port < 65535:
+        try:
+            with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+                s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+                s.bind((host, port))
+                return port
+        except OSError:
+            port += 1
+    return start_port
+
+
+def main() -> None:
+    """
+    Start the Abogen Flet frontend.
+
+    Parses ``--web`` and ``--port`` CLI arguments to choose desktop vs. web
+    mode, then hands control to ``ft.app()``.
+    """
+    import logging
+    logging.basicConfig(level=logging.INFO)
+    logging.getLogger("flet").setLevel(logging.INFO)
+
+    parser = argparse.ArgumentParser(description=f"{APP_NAME} – Flet frontend")
+    parser.add_argument(
+        "--web", action="store_true",
+        help="Run as a web server instead of a desktop window.",
+    )
+    parser.add_argument(
+        "--port", type=int, default=8080,
+        help="Port for the web server (default: 8080). Ignored in desktop mode.",
+    )
+    parser.add_argument(
+        "--host", default="127.0.0.1",
+        help="Host for the web server (default: 127.0.0.1). Use 0.0.0.0 to expose publicly.",
+    )
+    args = parser.parse_args()
+
+    if args.web:
+        port_specified = "--port" in sys.argv
+        target_port = args.port
+
+        if not port_specified:
+            target_port = _find_free_port(args.host, 8080)
+            if target_port != 8080:
+                print(f"Port 8080 is in use. Automatically routed to free port: {target_port}")
+        else:
+            if not _is_port_free(args.host, target_port):
+                print(f"Error: Port {target_port} is already in use on {args.host}.", file=sys.stderr)
+                print("Please select a different port or omit the --port flag to find one automatically.", file=sys.stderr)
+                sys.exit(1)
+
+        print(f"Starting Abogen WebUI on http://{args.host}:{target_port} ...")
+        ft.app(
+            target=_app_entry,
+            view=ft.AppView.WEB_BROWSER,
+            port=target_port,
+            host=args.host,
+            assets_dir=str(_ASSETS_DIR) if _ASSETS_DIR.exists() else None,
+            no_cdn=True,
+            web_renderer="canvaskit",
+        )
+    else:
+        try:
+            ft.app(
+                target=_app_entry,
+                view=ft.AppView.FLET_APP,
+                assets_dir=str(_ASSETS_DIR) if _ASSETS_DIR.exists() else None,
+            )
+        except Exception as e:
+            print(f"Warning: Failed to launch native desktop window: {e}", file=sys.stderr)
+            print("Falling back to running as a web application in your default browser...", file=sys.stderr)
+            target_port = _find_free_port("127.0.0.1", 8080)
+            print(f"Starting Abogen WebUI on http://127.0.0.1:{target_port} ...")
+            ft.app(
+                target=_app_entry,
+                view=ft.AppView.WEB_BROWSER,
+                port=target_port,
+                host="127.0.0.1",
+                assets_dir=str(_ASSETS_DIR) if _ASSETS_DIR.exists() else None,
+                no_cdn=True,
+                web_renderer="canvaskit",
+            )
+
+
+def main_web() -> None:
+    """
+    Start the Abogen Flet frontend as a web server.
+    """
+    import sys
+    if "--web" not in sys.argv:
+        sys.argv.insert(1, "--web")
+    main()
+
+
+if __name__ == "__main__":
+    main()

abogen/frontend/state/__init__.py

@@ -0,0 +1,4 @@
+"""State sub-package – exports AppState and ConversionJob."""
+from .app_state import AppState, ConversionJob
+
+__all__ = ["AppState", "ConversionJob"]

abogen/frontend/state/app_state.py

@@ -0,0 +1,451 @@
+"""
+Centralized, per-session application state for the Abogen Flet frontend.
+
+Each Flet page (session) gets its own instance of AppState, which guarantees
+complete isolation between simultaneous web-browser clients and the desktop
+window. The class carries every configuration variable, file buffer reference,
+and generation progress field that the rest of the UI reads or writes.
+
+This module intentionally has no Flet imports so it can be unit-tested without
+a running Flet server.
+"""
+
+from __future__ import annotations
+
+import threading
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Callable, Dict, List, Optional
+
+from abogen.utils import load_config, save_config
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _default_config() -> Dict[str, Any]:
+    """Load the persisted user config dict, returning an empty dict on failure."""
+    try:
+        return load_config() or {}
+    except Exception:
+        return {}
+
+
+# ---------------------------------------------------------------------------
+# Per-session state
+# ---------------------------------------------------------------------------
+
+
+@dataclass
+class ConversionJob:
+    """Lightweight descriptor of a single queued conversion job."""
+
+    file_path: str
+    """Absolute path to the text/epub/pdf/txt input file."""
+
+    display_name: str
+    """User-visible filename (may be the original epub/pdf path)."""
+
+    voice: str
+    """Voice formula string (e.g. 'af_heart' or 'af_heart*0.5+am_adam*0.5')."""
+
+    lang_code: str
+    """Single-char language prefix used by Kokoro (e.g. 'a', 'b', 'e')."""
+
+    speed: float = 1.0
+    """Playback speed multiplier, range 0.1 – 2.0."""
+
+    output_format: str = "mp3"
+    """Output audio container format."""
+
+    subtitle_mode: str = "Disabled"
+    """Subtitle generation mode."""
+
+    save_option: str = "Save next to input file"
+    """Save location strategy."""
+
+    output_folder: Optional[str] = None
+    """Absolute path when save_option is 'Choose output folder'."""
+
+    char_count: int = 0
+    """Pre-computed character count for ETR estimation."""
+
+    replace_single_newlines: bool = True
+    save_chapters_separately: Optional[bool] = None
+    merge_chapters_at_end: Optional[bool] = None
+
+
+@dataclass
+class AppState:
+    """
+    Single source of truth for one Flet session.
+
+    Instantiated once per ``ft.app()`` call on desktop, and once per browser
+    tab on web.  All UI components receive a reference to this object and
+    read/write it to keep themselves in sync.
+
+    Thread-safety: mutation from background threads should be done via the
+    provided ``_lock``.  The UI update callbacks (``on_log``,
+    ``on_progress``, etc.) are always invoked on the Flet event loop via
+    ``page.run_task()`` and must be set by the view layer.
+    """
+
+    # -----------------------------------------------------------------------
+    # Runtime identity
+    # -----------------------------------------------------------------------
+    _lock: threading.Lock = field(default_factory=threading.Lock, repr=False, compare=False)
+
+    # -----------------------------------------------------------------------
+    # Persisted user config (loaded once, written on every change)
+    # -----------------------------------------------------------------------
+    config: Dict[str, Any] = field(default_factory=_default_config)
+
+    # -----------------------------------------------------------------------
+    # File / input state
+    # -----------------------------------------------------------------------
+    selected_file: Optional[str] = None
+    """Path to the processed text file (may be a temp cache copy for epub/pdf)."""
+
+    selected_file_type: Optional[str] = None
+    """'txt' | 'epub' | 'pdf' | 'markdown' | None"""
+
+    selected_book_path: Optional[str] = None
+    """Original epub/pdf path before being converted to txt."""
+
+    displayed_file_path: Optional[str] = None
+    """Path shown in the UI drop-zone (original book or txt file)."""
+
+    selected_chapters: List[str] = field(default_factory=list)
+    """Ordered list of selected chapter href tokens (or page numbers for PDFs)."""
+
+    save_chapters_separately: Optional[bool] = None
+    merge_chapters_at_end: Optional[bool] = None
+    save_as_project: bool = False
+    char_count: int = 0
+
+    # -----------------------------------------------------------------------
+    # Voice / language
+    # -----------------------------------------------------------------------
+    selected_voice: str = "af_heart"
+    selected_lang: str = "a"
+    selected_profile_name: Optional[str] = None
+    mixed_voice_state: Optional[List[Any]] = None
+    """List of [voice_id, weight] pairs when the formula mixer is in use."""
+
+    # -----------------------------------------------------------------------
+    # Conversion parameters
+    # -----------------------------------------------------------------------
+    speed: float = 1.0
+    use_gpu: bool = True
+    selected_format: str = "wav"
+    subtitle_mode: str = "Sentence"
+    subtitle_format: str = "ass_centered_narrow"
+    replace_single_newlines: bool = True
+    save_option: str = "Save next to input file"
+    selected_output_folder: Optional[str] = None
+    silence_duration: float = 2.0
+    max_subtitle_words: int = 50
+    separate_chapters_format: str = "wav"
+    use_silent_gaps: bool = True
+    subtitle_speed_method: str = "tts"
+    use_spacy_segmentation: bool = True
+    chunk_level: str = "paragraph"
+    generate_epub3: bool = False
+
+    # TTS provider
+    tts_provider: str = "kokoro"
+    supertonic_total_steps: int = 5
+
+    # Chapter options
+    chapter_intro_delay: float = 0.5
+    read_title_intro: bool = False
+    read_closing_outro: bool = True
+    auto_prefix_chapter_titles: bool = True
+    normalize_chapter_opening_caps: bool = True
+
+    # Speaker analysis
+    speaker_analysis_threshold: int = 3
+
+    # Word substitutions
+    word_substitutions_enabled: bool = False
+    word_substitutions_list: str = ""
+    case_sensitive_substitutions: bool = False
+    replace_all_caps: bool = False
+    replace_numerals: bool = False
+    fix_nonstandard_punctuation: bool = False
+
+    # -----------------------------------------------------------------------
+    # Conversion runtime state
+    # -----------------------------------------------------------------------
+    is_converting: bool = False
+    is_cancelled: bool = False
+    progress: float = 0.0
+    """Fractional progress 0.0 – 1.0."""
+    etr_seconds: Optional[float] = None
+    """Estimated seconds remaining, or None if unknown."""
+    last_output_path: Optional[str] = None
+    log_lines: List[str] = field(default_factory=list)
+    """Buffered log messages, capped at LOG_MAX_LINES."""
+
+    LOG_MAX_LINES: int = 2000
+
+    # -----------------------------------------------------------------------
+    # Queue
+    # -----------------------------------------------------------------------
+    queued_items: List[ConversionJob] = field(default_factory=list)
+    current_queue_index: int = 0
+
+    # -----------------------------------------------------------------------
+    # Callbacks (set by the view layer, not serialised)
+    # -----------------------------------------------------------------------
+    on_log: Optional[Callable[[str, str], None]] = field(default=None, repr=False, compare=False)
+    """Called from any thread: ``on_log(message, level)``."""
+
+    on_progress: Optional[Callable[[float, Optional[float]], None]] = field(
+        default=None, repr=False, compare=False
+    )
+    """Called from any thread: ``on_progress(fraction, etr_seconds)``."""
+
+    on_conversion_finished: Optional[Callable[[str, Optional[str]], None]] = field(
+        default=None, repr=False, compare=False
+    )
+    """Called from any thread: ``on_conversion_finished(message, output_path)``."""
+
+    # -----------------------------------------------------------------------
+    # Integrations
+    # -----------------------------------------------------------------------
+    audiobookshelf_enabled: bool = False
+    audiobookshelf_base_url: str = ""
+    audiobookshelf_api_token: str = ""
+    audiobookshelf_library_id: str = ""
+    audiobookshelf_folder_id: str = ""
+    audiobookshelf_verify_ssl: bool = True
+    audiobookshelf_auto_send: bool = False
+    audiobookshelf_send_cover: bool = True
+    audiobookshelf_send_chapters: bool = True
+    audiobookshelf_send_subtitles: bool = False
+    audiobookshelf_timeout: float = 30.0
+
+    calibre_opds_enabled: bool = False
+    calibre_opds_base_url: str = ""
+    calibre_opds_username: str = ""
+    calibre_opds_password: str = ""
+    calibre_opds_verify_ssl: bool = True
+
+    # -----------------------------------------------------------------------
+    # Public helpers
+    # -----------------------------------------------------------------------
+
+    def load_from_config(self) -> None:
+        """
+        Populate all fields from the persisted JSON config file.
+
+        Called once at startup and whenever the settings page is saved.
+        Thread-safe.
+        """
+        with self._lock:
+            cfg = _default_config()
+            self.config = cfg
+
+            self.selected_voice = cfg.get("selected_voice", "af_heart")
+            self.selected_lang = self.selected_voice[0] if self.selected_voice else "a"
+            self.selected_profile_name = cfg.get("selected_profile_name")
+            self.speed = cfg.get("speed", 1.0)
+            self.use_gpu = cfg.get("use_gpu", True)
+            self.selected_format = cfg.get("selected_format", "wav")
+            self.subtitle_mode = cfg.get("subtitle_mode", "Sentence")
+            self.subtitle_format = cfg.get("subtitle_format", "ass_centered_narrow")
+            self.replace_single_newlines = cfg.get("replace_single_newlines", True)
+            self.save_option = cfg.get("save_option", "Save next to input file")
+            self.selected_output_folder = cfg.get("selected_output_folder")
+            self.silence_duration = cfg.get("silence_duration", 2.0)
+            self.max_subtitle_words = cfg.get("max_subtitle_words", 50)
+            self.separate_chapters_format = cfg.get("separate_chapters_format", "wav")
+            self.use_silent_gaps = cfg.get("use_silent_gaps", True)
+            self.subtitle_speed_method = cfg.get("subtitle_speed_method", "tts")
+            self.use_spacy_segmentation = cfg.get("use_spacy_segmentation", True)
+            self.chunk_level = cfg.get("chunk_level", "paragraph")
+            self.generate_epub3 = cfg.get("generate_epub3", False)
+            self.tts_provider = cfg.get("tts_provider", "kokoro")
+            self.supertonic_total_steps = cfg.get("supertonic_total_steps", 5)
+            self.chapter_intro_delay = cfg.get("chapter_intro_delay", 0.5)
+            self.read_title_intro = cfg.get("read_title_intro", False)
+            self.read_closing_outro = cfg.get("read_closing_outro", True)
+            self.auto_prefix_chapter_titles = cfg.get("auto_prefix_chapter_titles", True)
+            self.normalize_chapter_opening_caps = cfg.get("normalize_chapter_opening_caps", True)
+            self.speaker_analysis_threshold = cfg.get("speaker_analysis_threshold", 3)
+            self.word_substitutions_enabled = cfg.get("word_substitutions_enabled", False)
+            self.word_substitutions_list = cfg.get("word_substitutions_list", "")
+            self.case_sensitive_substitutions = cfg.get("case_sensitive_substitutions", False)
+            self.replace_all_caps = cfg.get("replace_all_caps", False)
+            self.replace_numerals = cfg.get("replace_numerals", False)
+            self.fix_nonstandard_punctuation = cfg.get("fix_nonstandard_punctuation", False)
+
+            # Integrations
+            integrations: Dict[str, Any] = cfg.get("integrations", {})
+            abs_cfg = integrations.get("audiobookshelf", {})
+            self.audiobookshelf_enabled = bool(abs_cfg.get("enabled", False))
+            self.audiobookshelf_base_url = str(abs_cfg.get("base_url", ""))
+            self.audiobookshelf_api_token = str(abs_cfg.get("api_token", ""))
+            self.audiobookshelf_library_id = str(abs_cfg.get("library_id", ""))
+            self.audiobookshelf_folder_id = str(abs_cfg.get("folder_id", ""))
+            self.audiobookshelf_verify_ssl = bool(abs_cfg.get("verify_ssl", True))
+            self.audiobookshelf_auto_send = bool(abs_cfg.get("auto_send", False))
+            self.audiobookshelf_send_cover = bool(abs_cfg.get("send_cover", True))
+            self.audiobookshelf_send_chapters = bool(abs_cfg.get("send_chapters", True))
+            self.audiobookshelf_send_subtitles = bool(abs_cfg.get("send_subtitles", False))
+            self.audiobookshelf_timeout = float(abs_cfg.get("timeout", 30.0))
+
+            cal_cfg = integrations.get("calibre_opds", {})
+            self.calibre_opds_enabled = bool(cal_cfg.get("enabled", False))
+            self.calibre_opds_base_url = str(cal_cfg.get("base_url", ""))
+            self.calibre_opds_username = str(cal_cfg.get("username", ""))
+            self.calibre_opds_password = str(cal_cfg.get("password", ""))
+            self.calibre_opds_verify_ssl = bool(cal_cfg.get("verify_ssl", True))
+
+    def persist_config(self) -> None:
+        """
+        Write the current config snapshot back to disk.
+
+        Only the fields that map to the JSON config are written; runtime state
+        (progress, log_lines, callbacks) is not persisted.
+        Thread-safe.
+        """
+        with self._lock:
+            cfg = self.config.copy()
+            cfg["selected_voice"] = self.selected_voice
+            cfg["selected_profile_name"] = self.selected_profile_name
+            cfg["speed"] = self.speed
+            cfg["use_gpu"] = self.use_gpu
+            cfg["selected_format"] = self.selected_format
+            cfg["subtitle_mode"] = self.subtitle_mode
+            cfg["subtitle_format"] = self.subtitle_format
+            cfg["replace_single_newlines"] = self.replace_single_newlines
+            cfg["save_option"] = self.save_option
+            cfg["selected_output_folder"] = self.selected_output_folder
+            cfg["silence_duration"] = self.silence_duration
+            cfg["max_subtitle_words"] = self.max_subtitle_words
+            cfg["separate_chapters_format"] = self.separate_chapters_format
+            cfg["use_silent_gaps"] = self.use_silent_gaps
+            cfg["subtitle_speed_method"] = self.subtitle_speed_method
+            cfg["use_spacy_segmentation"] = self.use_spacy_segmentation
+            cfg["chunk_level"] = self.chunk_level
+            cfg["generate_epub3"] = self.generate_epub3
+            cfg["tts_provider"] = self.tts_provider
+            cfg["supertonic_total_steps"] = self.supertonic_total_steps
+            cfg["chapter_intro_delay"] = self.chapter_intro_delay
+            cfg["read_title_intro"] = self.read_title_intro
+            cfg["read_closing_outro"] = self.read_closing_outro
+            cfg["auto_prefix_chapter_titles"] = self.auto_prefix_chapter_titles
+            cfg["normalize_chapter_opening_caps"] = self.normalize_chapter_opening_caps
+            cfg["speaker_analysis_threshold"] = self.speaker_analysis_threshold
+            cfg["word_substitutions_enabled"] = self.word_substitutions_enabled
+            cfg["word_substitutions_list"] = self.word_substitutions_list
+            cfg["case_sensitive_substitutions"] = self.case_sensitive_substitutions
+            cfg["replace_all_caps"] = self.replace_all_caps
+            cfg["replace_numerals"] = self.replace_numerals
+            cfg["fix_nonstandard_punctuation"] = self.fix_nonstandard_punctuation
+            # Integrations
+            cfg.setdefault("integrations", {})
+            cfg["integrations"]["audiobookshelf"] = {
+                "enabled": self.audiobookshelf_enabled,
+                "base_url": self.audiobookshelf_base_url,
+                "api_token": self.audiobookshelf_api_token,
+                "library_id": self.audiobookshelf_library_id,
+                "folder_id": self.audiobookshelf_folder_id,
+                "verify_ssl": self.audiobookshelf_verify_ssl,
+                "auto_send": self.audiobookshelf_auto_send,
+                "send_cover": self.audiobookshelf_send_cover,
+                "send_chapters": self.audiobookshelf_send_chapters,
+                "send_subtitles": self.audiobookshelf_send_subtitles,
+                "timeout": self.audiobookshelf_timeout,
+            }
+            cfg["integrations"]["calibre_opds"] = {
+                "enabled": self.calibre_opds_enabled,
+                "base_url": self.calibre_opds_base_url,
+                "username": self.calibre_opds_username,
+                "password": self.calibre_opds_password,
+                "verify_ssl": self.calibre_opds_verify_ssl,
+            }
+            self.config = cfg
+            try:
+                save_config(cfg)
+            except Exception:
+                pass
+
+    def append_log(self, message: str, level: str = "info") -> None:
+        """
+        Thread-safely append a log line and trigger the UI callback.
+
+        Caps the internal buffer at ``LOG_MAX_LINES`` to prevent unbounded
+        memory growth during very long conversion tasks.
+        """
+        with self._lock:
+            self.log_lines.append(f"[{level.upper()}] {message}")
+            if len(self.log_lines) > self.LOG_MAX_LINES:
+                # Trim oldest 10 % to amortise the cost of trimming
+                trim = self.LOG_MAX_LINES // 10
+                self.log_lines = self.log_lines[trim:]
+
+        cb = self.on_log
+        if cb is not None:
+            try:
+                cb(message, level)
+            except Exception:
+                pass
+
+    def update_progress(self, fraction: float, etr: Optional[float] = None) -> None:
+        """
+        Update fractional progress and ETR, then notify the UI callback.
+
+        Args:
+            fraction: Value in [0.0, 1.0].
+            etr: Estimated seconds remaining, or None.
+        """
+        with self._lock:
+            self.progress = max(0.0, min(1.0, fraction))
+            self.etr_seconds = etr
+
+        cb = self.on_progress
+        if cb is not None:
+            try:
+                cb(fraction, etr)
+            except Exception:
+                pass
+
+    def get_voice_formula(self) -> str:
+        """
+        Return the effective voice formula string.
+
+        Uses the mixed_voice_state if the formula mixer is active, otherwise
+        returns the raw selected_voice.
+        """
+        if self.mixed_voice_state:
+            parts = [f"{name}*{weight}" for name, weight in self.mixed_voice_state]
+            return " + ".join(filter(None, parts))
+        return self.selected_voice or "af_heart"
+
+    def reset_file_state(self) -> None:
+        """Clear all file-related fields without touching voice/settings."""
+        with self._lock:
+            self.selected_file = None
+            self.selected_file_type = None
+            self.selected_book_path = None
+            self.displayed_file_path = None
+            self.selected_chapters = []
+            self.save_chapters_separately = None
+            self.merge_chapters_at_end = None
+            self.save_as_project = False
+            self.char_count = 0
+
+    def reset_conversion_state(self) -> None:
+        """Clear all runtime conversion fields to start fresh."""
+        with self._lock:
+            self.is_converting = False
+            self.is_cancelled = False
+            self.progress = 0.0
+            self.etr_seconds = None
+            self.last_output_path = None
+            self.log_lines = []

abogen/frontend/utils/__init__.py

@@ -0,0 +1,38 @@
+"""Utils sub-package."""
+from .helpers import (
+    human_readable_size,
+    format_duration,
+    format_etr,
+    detect_file_type,
+    is_supported_file,
+    is_book_type,
+    voice_lang_code,
+    language_label,
+    grouped_voices,
+    voice_display_name,
+    parse_voice_formula,
+    format_number,
+    safe_basename,
+    output_format_label,
+    subtitle_format_label,
+    SUPPORTED_EXTENSIONS,
+)
+
+__all__ = [
+    "human_readable_size",
+    "format_duration",
+    "format_etr",
+    "detect_file_type",
+    "is_supported_file",
+    "is_book_type",
+    "voice_lang_code",
+    "language_label",
+    "grouped_voices",
+    "voice_display_name",
+    "parse_voice_formula",
+    "format_number",
+    "safe_basename",
+    "output_format_label",
+    "subtitle_format_label",
+    "SUPPORTED_EXTENSIONS",
+]

abogen/frontend/utils/conversion_bridge.py

@@ -0,0 +1,462 @@
+"""
+Background conversion bridge for the Abogen Flet frontend.
+
+This module wraps the existing ``abogen.webui.conversion_runner`` (and its
+``ConversionService`` / ``Job`` machinery) in an async-friendly interface that
+can push real-time progress and log updates back to the Flet event loop without
+blocking the UI thread.
+
+Key design decisions
+--------------------
+* All heavy work is offloaded to daemon threads. The Flet page event loop
+  is never blocked.
+* Progress and log callbacks are scheduled back onto the Flet page via
+  ``page.run_task()`` so Flet's session isolation remains intact.
+* Cancellation is cooperative: the underlying job's ``cancel_requested``
+  flag is set, and the runner checks it at chunk boundaries.
+* The module is a pure adapter – it does NOT duplicate any processing logic
+  from the core pipeline.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import os
+import tempfile
+import threading
+import time
+import traceback
+from pathlib import Path
+from typing import Any, Callable, Dict, List, Optional
+
+import flet as ft
+
+from abogen.utils import (
+    get_gpu_acceleration,
+    get_user_cache_path,
+    get_user_output_path,
+    load_numpy_kpipeline,
+    prevent_sleep_end,
+    prevent_sleep_start,
+)
+from abogen.webui.service import (
+    ConversionService,
+    Job,
+    JobStatus,
+    PendingJob,
+    build_service,
+)
+from abogen.webui.conversion_runner import run_conversion_job
+
+from ..state import AppState
+
+# ---------------------------------------------------------------------------
+# Module-level singleton ConversionService (shared across sessions, as in the
+# web UI – but each job carries its own output folder keyed by session).
+# ---------------------------------------------------------------------------
+
+_SERVICE_LOCK = threading.Lock()
+_SERVICE: Optional[ConversionService] = None
+
+
+def _get_service() -> ConversionService:
+    """
+    Return (creating if necessary) the module-level ConversionService.
+
+    The service manages the background worker thread and persistent job state.
+    Thread-safe via a module-level lock.
+    """
+    global _SERVICE
+    with _SERVICE_LOCK:
+        if _SERVICE is None:
+            output_root = Path(get_user_output_path("frontend"))
+            uploads_root = Path(get_user_cache_path("frontend/uploads"))
+            _SERVICE = build_service(
+                runner=run_conversion_job,
+                output_root=output_root,
+                uploads_root=uploads_root,
+            )
+    return _SERVICE
+
+
+# ---------------------------------------------------------------------------
+# Public conversion bridge
+# ---------------------------------------------------------------------------
+
+
+class ConversionBridge:
+    """
+    Thin adapter between the Flet UI session and the core conversion pipeline.
+
+    One ``ConversionBridge`` instance is created per Flet page (session) and
+    is responsible for:
+    1. Accepting a conversion request from the UI.
+    2. Writing the input text to a temp file if needed.
+    3. Submitting the job to ``ConversionService``.
+    4. Polling the job from a daemon thread and forwarding progress/logs to
+       the Flet page via ``page.run_task()``.
+    5. Providing a ``cancel()`` method that sets the cooperative flag.
+    """
+
+    def __init__(self, page: ft.Page, state: AppState) -> None:
+        """
+        Initialise the bridge.
+
+        Args:
+            page: The Flet ``Page`` for this session. Used to schedule
+                  UI callbacks on the correct event loop.
+            state: The session's ``AppState`` instance.
+        """
+        self._page = page
+        self._state = state
+        self._current_job: Optional[Job] = None
+        self._poll_thread: Optional[threading.Thread] = None
+        self._stop_poll = threading.Event()
+        self._seen_log_count = 0
+
+    # ------------------------------------------------------------------
+    # Public API
+    # ------------------------------------------------------------------
+
+    def start(
+        self,
+        *,
+        input_file: str,
+        voice: str,
+        lang_code: str,
+        speed: float,
+        output_format: str,
+        subtitle_mode: str,
+        subtitle_format: str,
+        use_gpu: bool,
+        save_option: str,
+        output_folder: Optional[str],
+        replace_single_newlines: bool,
+        char_count: int,
+        chapters: Optional[List[Dict[str, Any]]] = None,
+        save_chapters_separately: bool = False,
+        merge_chapters_at_end: bool = True,
+        separate_chapters_format: str = "wav",
+        silence_between_chapters: float = 2.0,
+        max_subtitle_words: int = 50,
+        chapter_intro_delay: float = 0.5,
+        read_title_intro: bool = False,
+        read_closing_outro: bool = True,
+        auto_prefix_chapter_titles: bool = True,
+        normalize_chapter_opening_caps: bool = True,
+        tts_provider: str = "kokoro",
+        supertonic_total_steps: int = 5,
+        chunk_level: str = "paragraph",
+        generate_epub3: bool = False,
+        word_substitutions_enabled: bool = False,
+        word_substitutions_list: str = "",
+        case_sensitive_substitutions: bool = False,
+        replace_all_caps: bool = False,
+        replace_numerals: bool = False,
+        fix_nonstandard_punctuation: bool = False,
+    ) -> None:
+        """
+        Submit a conversion job and begin the progress-polling loop.
+
+        This method returns immediately; all heavy work runs on daemon threads.
+        UI callbacks (``state.on_log``, ``state.on_progress``,
+        ``state.on_conversion_finished``) are scheduled on the Flet event loop.
+
+        Args:
+            input_file: Absolute path to the text/epub/pdf input file.
+            voice: Kokoro voice formula string.
+            lang_code: Single-char language code.
+            speed: Playback speed multiplier (0.1 – 2.0).
+            output_format: Audio container key (``'wav'``, ``'mp3'``, …).
+            subtitle_mode: Subtitle generation mode string.
+            subtitle_format: Subtitle container key (``'srt'``, ``'ass_wide'``, …).
+            use_gpu: Whether to request GPU acceleration.
+            save_option: Save-location strategy string.
+            output_folder: Explicit output folder or None.
+            replace_single_newlines: Pre-processing flag.
+            char_count: Pre-computed character count for ETR estimation.
+            chapters: Optional list of chapter dicts for epub/pdf.
+            save_chapters_separately: Split chapters into separate files.
+            merge_chapters_at_end: Merge chapter files into one after generation.
+            separate_chapters_format: Format for individual chapter files.
+            silence_between_chapters: Silence gap (seconds) between chapters.
+            max_subtitle_words: Maximum words per subtitle block.
+            chapter_intro_delay: Silence before chapter title announcement (s).
+            read_title_intro: Announce book title at the start.
+            read_closing_outro: Announce book title at the end.
+            auto_prefix_chapter_titles: Prepend "Chapter N." to titles.
+            normalize_chapter_opening_caps: Fix ALL-CAPS opening lines.
+            tts_provider: ``'kokoro'`` or ``'supertonic'``.
+            supertonic_total_steps: Quality steps for the Supertonic pipeline.
+            chunk_level: ``'paragraph'`` or ``'sentence'`` chunking granularity.
+            generate_epub3: Also produce an EPUB3 audiobook package.
+            word_substitutions_enabled: Toggle word-substitution pre-processing.
+            word_substitutions_list: Newline-delimited ``word|replacement`` rules.
+            case_sensitive_substitutions: Case-sensitive matching for substitutions.
+            replace_all_caps: Lowercase ALL-CAPS words.
+            replace_numerals: Convert digits to spoken words.
+            fix_nonstandard_punctuation: Normalise curly quotes etc.
+        """
+        if self._state.is_converting:
+            return
+
+        # Resolve the effective output folder
+        resolved_output: Optional[Path] = self._resolve_output_folder(
+            save_option=save_option,
+            output_folder=output_folder,
+            input_file=input_file,
+        )
+
+        # Store the input file as a Path
+        stored_path = Path(input_file)
+        original_filename = stored_path.name
+
+        # Block signals until the job is submitted
+        prevent_sleep_start()
+        self._state.is_converting = True
+        self._state.is_cancelled = False
+        self._state.progress = 0.0
+        self._state.etr_seconds = None
+        self._state.log_lines = []
+        self._seen_log_count = 0
+
+        # Enqueue the job on the service
+        service = _get_service()
+        job = service.enqueue(
+            original_filename=original_filename,
+            stored_path=stored_path,
+            language=lang_code,
+            voice=voice,
+            speed=speed,
+            tts_provider=tts_provider,
+            supertonic_total_steps=supertonic_total_steps,
+            use_gpu=use_gpu,
+            subtitle_mode=subtitle_mode,
+            output_format=output_format,
+            save_mode=self._save_mode_key(save_option),
+            output_folder=resolved_output,
+            replace_single_newlines=replace_single_newlines,
+            subtitle_format=subtitle_format,
+            total_characters=char_count,
+            chapters=chapters or [],
+            save_chapters_separately=save_chapters_separately,
+            merge_chapters_at_end=merge_chapters_at_end,
+            separate_chapters_format=separate_chapters_format,
+            silence_between_chapters=silence_between_chapters,
+            max_subtitle_words=max_subtitle_words,
+            chapter_intro_delay=chapter_intro_delay,
+            read_title_intro=read_title_intro,
+            read_closing_outro=read_closing_outro,
+            auto_prefix_chapter_titles=auto_prefix_chapter_titles,
+            normalize_chapter_opening_caps=normalize_chapter_opening_caps,
+            chunk_level=chunk_level,
+            generate_epub3=generate_epub3,
+        )
+        self._current_job = job
+
+        # Persist word-substitution settings to config so the runner picks them up
+        self._state.word_substitutions_enabled = word_substitutions_enabled
+        self._state.word_substitutions_list = word_substitutions_list
+        self._state.case_sensitive_substitutions = case_sensitive_substitutions
+        self._state.replace_all_caps = replace_all_caps
+        self._state.replace_numerals = replace_numerals
+        self._state.fix_nonstandard_punctuation = fix_nonstandard_punctuation
+        self._state.persist_config()
+
+        # Start the poll thread
+        self._stop_poll.clear()
+        self._poll_thread = threading.Thread(
+            target=self._poll_job_loop, daemon=True, name="abogen-poll"
+        )
+        self._poll_thread.start()
+
+    def cancel(self) -> None:
+        """
+        Request cancellation of the currently running job.
+
+        Sets the cooperative flag on the underlying ``Job`` object; the runner
+        will stop after completing the current text chunk.
+        """
+        if self._current_job is not None:
+            self._state.is_cancelled = True
+            try:
+                _get_service().cancel(self._current_job.id)
+            except Exception:
+                pass
+
+    # ------------------------------------------------------------------
+    # Internal helpers
+    # ------------------------------------------------------------------
+
+    @staticmethod
+    def _save_mode_key(option: str) -> str:
+        """
+        Convert the human-readable save option to the service's internal key.
+
+        Args:
+            option: UI-facing string (``'Save next to input file'``, …).
+
+        Returns:
+            Service key string.
+        """
+        mapping = {
+            "Save next to input file": "save_next_to_input",
+            "Save to Desktop": "save_to_desktop",
+            "Choose output folder": "custom",
+        }
+        return mapping.get(option, "save_next_to_input")
+
+    @staticmethod
+    def _resolve_output_folder(
+        save_option: str,
+        output_folder: Optional[str],
+        input_file: str,
+    ) -> Optional[Path]:
+        """
+        Return the output ``Path`` based on the save option, or None for
+        the "next to input" strategy (the runner handles that internally).
+
+        Args:
+            save_option: UI-facing save strategy string.
+            output_folder: Explicit path when ``save_option`` is ``'Choose output folder'``.
+            input_file: Path to the source file for the ``'Save to Desktop'`` strategy.
+
+        Returns:
+            Resolved ``Path`` or ``None``.
+        """
+        if save_option == "Choose output folder" and output_folder:
+            p = Path(output_folder)
+            p.mkdir(parents=True, exist_ok=True)
+            return p
+        if save_option == "Save to Desktop":
+            desktop = Path.home() / "Desktop"
+            desktop.mkdir(exist_ok=True)
+            return desktop
+        # "Save next to input file" – let the runner decide
+        return None
+
+    def _poll_job_loop(self) -> None:
+        """
+        Background daemon loop that polls the current Job for updates.
+
+        Runs until the job enters a terminal state or until ``_stop_poll``
+        is set.  Uses ``page.run_task()`` to schedule UI updates on the Flet
+        event loop without triggering thread-safety violations.
+        """
+        job = self._current_job
+        if job is None:
+            return
+
+        service = _get_service()
+        POLL_INTERVAL = 0.25  # seconds
+
+        while not self._stop_poll.is_set():
+            # Re-fetch the current job state (it's mutated in-place by the runner)
+            current = service.get_job(job.id)
+            if current is None:
+                break
+
+            # Forward new log lines
+            new_logs = current.logs[self._seen_log_count:]
+            self._seen_log_count += len(new_logs)
+            for log_entry in new_logs:
+                level = getattr(log_entry, "level", "info")
+                message = getattr(log_entry, "message", str(log_entry))
+                self._schedule_log(message, level)
+
+            # Forward progress
+            if current.progress is not None:
+                etr = getattr(current, "estimated_time_remaining", None)
+                self._schedule_progress(float(current.progress), etr)
+
+            # Check for terminal states
+            status = current.status
+            if status in (
+                JobStatus.COMPLETED,
+                JobStatus.FAILED,
+                JobStatus.CANCELLED,
+            ):
+                output_path: Optional[str] = None
+                if current.result and current.result.audio_path:
+                    output_path = str(current.result.audio_path)
+                if status == JobStatus.COMPLETED:
+                    finish_msg = "Conversion completed successfully."
+                elif status == JobStatus.CANCELLED:
+                    finish_msg = "Cancelled"
+                else:
+                    finish_msg = f"Conversion failed: {current.error or 'Unknown error'}"
+
+                self._schedule_finished(finish_msg, output_path)
+                break
+
+            time.sleep(POLL_INTERVAL)
+
+        prevent_sleep_end()
+        self._state.is_converting = False
+
+    def _schedule_log(self, message: str, level: str) -> None:
+        """Schedule a log update on the Flet event loop."""
+        state = self._state
+        page = self._page
+        state.append_log(message, level)
+
+        async def _update() -> None:
+            cb = state.on_log
+            if cb:
+                cb(message, level)
+            try:
+                page.update()
+            except Exception:
+                pass
+
+        try:
+            page.run_task(_update)
+        except Exception:
+            pass
+
+    def _schedule_progress(self, fraction: float, etr: Optional[float]) -> None:
+        """Schedule a progress update on the Flet event loop."""
+        state = self._state
+        page = self._page
+        state.progress = max(0.0, min(1.0, fraction))
+        state.etr_seconds = etr
+
+        async def _update() -> None:
+            cb = state.on_progress
+            if cb:
+                cb(fraction, etr)
+            try:
+                page.update()
+            except Exception:
+                pass
+
+        try:
+            page.run_task(_update)
+        except Exception:
+            pass
+
+    def _schedule_finished(
+        self, message: str, output_path: Optional[str]
+    ) -> None:
+        """Schedule a completion notification on the Flet event loop."""
+        state = self._state
+        page = self._page
+        state.last_output_path = output_path
+        self._stop_poll.set()
+
+        async def _update() -> None:
+            state.is_converting = False
+            state.progress = 1.0
+            state.last_output_path = output_path
+            cb = state.on_conversion_finished
+            if cb:
+                cb(message, output_path)
+            try:
+                page.update()
+            except Exception:
+                pass
+
+        try:
+            page.run_task(_update)
+        except Exception:
+            pass

abogen/frontend/utils/helpers.py

@@ -0,0 +1,313 @@
+"""
+Frontend-specific utilities for the Abogen Flet application.
+
+Contains helpers for:
+- Human-readable size / duration formatting
+- Voice formula parsing and display
+- File-type detection
+- ETR (Estimated Time Remaining) formatting
+- Path resolution that adapts to desktop vs. web context
+"""
+
+from __future__ import annotations
+
+import os
+import re
+from pathlib import Path
+from typing import List, Optional, Tuple
+
+from abogen.constants import (
+    LANGUAGE_DESCRIPTIONS,
+    SUPPORTED_INPUT_FORMATS,
+    SUPPORTED_SOUND_FORMATS,
+    SUBTITLE_FORMATS,
+    VOICES_INTERNAL,
+)
+
+
+# ---------------------------------------------------------------------------
+# Size / duration helpers
+# ---------------------------------------------------------------------------
+
+
+def human_readable_size(size_bytes: int, decimal_places: int = 2) -> str:
+    """
+    Convert a byte count into a human-readable string.
+
+    Args:
+        size_bytes: Number of bytes.
+        decimal_places: Significant decimal digits in the output.
+
+    Returns:
+        A string like ``"3.14 MB"`` or ``"1.00 KB"``.
+    """
+    for unit in ("B", "KB", "MB", "GB", "TB"):
+        if size_bytes < 1024.0:
+            return f"{size_bytes:.{decimal_places}f} {unit}"
+        size_bytes /= 1024.0  # type: ignore[assignment]
+    return f"{size_bytes:.{decimal_places}f} PB"
+
+
+def format_duration(seconds: float) -> str:
+    """
+    Format a duration in seconds as ``HH:MM:SS``.
+
+    Args:
+        seconds: Non-negative floating-point duration.
+
+    Returns:
+        A colon-delimited time string, e.g. ``"00:03:42"``.
+    """
+    total = max(0, int(seconds))
+    h, remainder = divmod(total, 3600)
+    m, s = divmod(remainder, 60)
+    return f"{h:02d}:{m:02d}:{s:02d}"
+
+
+def format_etr(etr_seconds: Optional[float]) -> str:
+    """
+    Format an estimated time remaining value for the UI.
+
+    Args:
+        etr_seconds: Seconds remaining, or None when unknown.
+
+    Returns:
+        Human-readable string such as ``"~3 min 42 sec"`` or ``"Calculating…"``.
+    """
+    if etr_seconds is None:
+        return "Calculating…"
+    total = max(0, int(etr_seconds))
+    if total < 60:
+        return f"~{total} sec"
+    m, s = divmod(total, 60)
+    if m < 60:
+        return f"~{m} min {s} sec"
+    h, m = divmod(m, 60)
+    return f"~{h} h {m} min"
+
+
+# ---------------------------------------------------------------------------
+# File helpers
+# ---------------------------------------------------------------------------
+
+
+SUPPORTED_EXTENSIONS: Tuple[str, ...] = (
+    ".txt",
+    ".epub",
+    ".pdf",
+    ".md",
+    ".markdown",
+    ".srt",
+    ".ass",
+    ".vtt",
+)
+"""All file extensions that the drop-zone accepts."""
+
+
+def detect_file_type(file_path: str) -> str:
+    """
+    Return a normalised file-type token for the given path.
+
+    Args:
+        file_path: Absolute or relative path to the input file.
+
+    Returns:
+        One of ``'txt'``, ``'epub'``, ``'pdf'``, ``'markdown'``,
+        ``'subtitle'``, or ``'unknown'``.
+    """
+    ext = Path(file_path).suffix.lower()
+    if ext == ".epub":
+        return "epub"
+    if ext == ".pdf":
+        return "pdf"
+    if ext in (".md", ".markdown"):
+        return "markdown"
+    if ext in (".srt", ".ass", ".vtt"):
+        return "subtitle"
+    if ext == ".txt":
+        return "txt"
+    return "unknown"
+
+
+def is_supported_file(file_path: str) -> bool:
+    """
+    Return True when the file extension is in the supported set.
+
+    Args:
+        file_path: Path whose extension is inspected.
+    """
+    return Path(file_path).suffix.lower() in SUPPORTED_EXTENSIONS
+
+
+def is_book_type(file_type: str) -> bool:
+    """
+    Return True for file types that contain chapters / pages.
+
+    Args:
+        file_type: Token from ``detect_file_type()``.
+    """
+    return file_type in ("epub", "pdf", "markdown")
+
+
+# ---------------------------------------------------------------------------
+# Voice helpers
+# ---------------------------------------------------------------------------
+
+
+def voice_lang_code(voice: str) -> str:
+    """
+    Extract the language code character from a Kokoro voice name.
+
+    The first character of every internal voice name encodes the language
+    (e.g. ``'a'`` for American English, ``'b'`` for British English).
+
+    Args:
+        voice: Raw voice string like ``'af_heart'`` or a formula.
+
+    Returns:
+        Single lowercase character, defaulting to ``'a'`` on failure.
+    """
+    if not voice:
+        return "a"
+    # For plain voice IDs the first char is the language
+    if voice[0].isalpha() and "_" in voice[:4]:
+        return voice[0].lower()
+    # Formula: extract first alpha char
+    match = re.search(r"\b([a-z])", voice)
+    return match.group(1) if match else "a"
+
+
+def language_label(lang_code: str) -> str:
+    """
+    Return the human-readable label for a language code.
+
+    Args:
+        lang_code: Single-character code (``'a'``, ``'b'``, …).
+
+    Returns:
+        Display string, e.g. ``"American English"``.
+    """
+    return LANGUAGE_DESCRIPTIONS.get(lang_code, lang_code.upper())
+
+
+def grouped_voices() -> List[Tuple[str, List[str]]]:
+    """
+    Return the internal voice list grouped by language for display.
+
+    Returns:
+        List of ``(language_label, [voice_id, …])`` tuples.
+    """
+    groups: dict[str, List[str]] = {}
+    for v in VOICES_INTERNAL:
+        lang = language_label(v[0])
+        groups.setdefault(lang, []).append(v)
+    return sorted(groups.items())
+
+
+def voice_display_name(voice_id: str) -> str:
+    """
+    Convert a raw voice ID like ``'af_heart'`` to a prettier display name.
+
+    Args:
+        voice_id: Raw internal voice identifier.
+
+    Returns:
+        Formatted string, e.g. ``"af_heart"`` (unchanged; may be enhanced later).
+    """
+    return voice_id
+
+
+def parse_voice_formula(formula: str) -> List[Tuple[str, float]]:
+    """
+    Parse a Kokoro voice mix formula into a list of ``(voice_id, weight)`` tuples.
+
+    Example:
+        ``"af_heart*0.7+am_adam*0.3"`` → ``[('af_heart', 0.7), ('am_adam', 0.3)]``
+
+    Args:
+        formula: Space- or ``+``-joined mix formula string.
+
+    Returns:
+        Parsed list; empty if parsing fails.
+    """
+    parts: List[Tuple[str, float]] = []
+    for token in re.split(r"[+\s]+", formula.strip()):
+        token = token.strip()
+        if not token:
+            continue
+        if "*" in token:
+            name, _, weight_str = token.partition("*")
+            try:
+                parts.append((name.strip(), float(weight_str.strip())))
+            except ValueError:
+                pass
+        else:
+            # Bare voice id — assume full weight
+            if token in VOICES_INTERNAL:
+                parts.append((token, 1.0))
+    return parts
+
+
+# ---------------------------------------------------------------------------
+# Number formatting
+# ---------------------------------------------------------------------------
+
+
+def format_number(n: int) -> str:
+    """
+    Format an integer with thousands separators.
+
+    Args:
+        n: Integer value.
+
+    Returns:
+        Formatted string, e.g. ``"1,234,567"``.
+    """
+    return f"{n:,}"
+
+
+# ---------------------------------------------------------------------------
+# Path helpers
+# ---------------------------------------------------------------------------
+
+
+def safe_basename(path: Optional[str]) -> str:
+    """
+    Return the basename of a path, or an empty string when path is None/empty.
+
+    Args:
+        path: Optional file-system path.
+    """
+    if not path:
+        return ""
+    return os.path.basename(path)
+
+
+def output_format_label(fmt: str) -> str:
+    """
+    Return a display label for an audio output format key.
+
+    Args:
+        fmt: Lowercase format key (``'wav'``, ``'mp3'``, …).
+    """
+    labels = {
+        "wav": "WAV (lossless)",
+        "flac": "FLAC (lossless compressed)",
+        "mp3": "MP3",
+        "opus": "Opus (best compression)",
+        "m4b": "M4B (with chapters)",
+    }
+    return labels.get(fmt, fmt.upper())
+
+
+def subtitle_format_label(key: str) -> str:
+    """
+    Return the display label for a subtitle format key.
+
+    Args:
+        key: Internal subtitle format key (e.g. ``'ass_centered_narrow'``).
+    """
+    for k, label in SUBTITLE_FORMATS:
+        if k == key:
+            return label
+    return key

abogen/frontend/utils/theme.py

@@ -0,0 +1,264 @@
+"""
+Design tokens and theme configuration for the Abogen Flet frontend.
+
+This module defines the application's complete colour palette, typography
+scale, spacing constants, and border radii in one canonical place.
+All component modules import from here; changing a value here propagates
+instantly across the entire UI.
+
+Flet's ``ft.Theme`` uses ``ColorScheme``, but for custom widgets we paint
+directly with hex colours drawn from ``LIGHT`` and ``DARK`` palettes.
+"""
+
+from __future__ import annotations
+
+import flet as ft
+from dataclasses import dataclass
+
+
+# ---------------------------------------------------------------------------
+# Colour palettes
+# ---------------------------------------------------------------------------
+
+
+@dataclass(frozen=True)
+class _Palette:
+    """A complete colour palette for one theme mode."""
+
+    # Backgrounds
+    bg_base: str        # Deepest background (window / page)
+    bg_surface: str     # Cards, panels, dialogs
+    bg_elevated: str    # Slightly raised elements (toolbar, sidebar)
+    bg_input: str       # Text-field / dropdown backgrounds
+
+    # Brand accent
+    accent: str         # Primary interactive colour (buttons, links)
+    accent_muted: str   # Hover tint over accents
+    accent_on: str      # Text drawn on top of accent fills
+
+    # Semantic
+    success: str
+    error: str
+    warning: str
+    info: str
+
+    # Text hierarchy
+    text_primary: str
+    text_secondary: str
+    text_disabled: str
+    text_on_accent: str
+
+    # Borders / dividers
+    border: str
+    border_focused: str
+    divider: str
+
+    # Specific UI atoms
+    drop_zone_border: str
+    drop_zone_bg: str
+    drop_zone_active_border: str
+    drop_zone_active_bg: str
+    log_bg: str
+    log_text: str
+    progress_bar_bg: str
+    progress_bar_fill: str
+    sidebar_bg: str
+    sidebar_selected_bg: str
+    sidebar_selected_text: str
+    nav_indicator: str
+
+
+DARK = _Palette(
+    bg_base="#0f1117",
+    bg_surface="#181b23",
+    bg_elevated="#1e2230",
+    bg_input="#252a38",
+
+    accent="#5b8af5",
+    accent_muted="#3a5fc4",
+    accent_on="#ffffff",
+
+    success="#42ad4a",
+    error="#e84e3c",
+    warning="#f5a623",
+    info="#5b8af5",
+
+    text_primary="#e8eaf0",
+    text_secondary="#9ba3b8",
+    text_disabled="#4e5568",
+    text_on_accent="#ffffff",
+
+    border="#2c3147",
+    border_focused="#5b8af5",
+    divider="#252a38",
+
+    drop_zone_border="#3a4466",
+    drop_zone_bg="#151928",
+    drop_zone_active_border="#42ad4a",
+    drop_zone_active_bg="#0d1f10",
+    log_bg="#0d1117",
+    log_text="#b0b8cc",
+    progress_bar_bg="#1e2230",
+    progress_bar_fill="#5b8af5",
+    sidebar_bg="#13161f",
+    sidebar_selected_bg="#252a38",
+    sidebar_selected_text="#5b8af5",
+    nav_indicator="#5b8af5",
+)
+
+LIGHT = _Palette(
+    bg_base="#f4f5f8",
+    bg_surface="#ffffff",
+    bg_elevated="#edf0f5",
+    bg_input="#f0f2f7",
+
+    accent="#3a5fc4",
+    accent_muted="#2a4fae",
+    accent_on="#ffffff",
+
+    success="#2e9437",
+    error="#c0392b",
+    warning="#d4870a",
+    info="#3a5fc4",
+
+    text_primary="#1a1d27",
+    text_secondary="#5a6172",
+    text_disabled="#9ba3b8",
+    text_on_accent="#ffffff",
+
+    border="#dce0ea",
+    border_focused="#3a5fc4",
+    divider="#e8ebf2",
+
+    drop_zone_border="#a8b4d0",
+    drop_zone_bg="#f7f8fd",
+    drop_zone_active_border="#2e9437",
+    drop_zone_active_bg="#f0fff1",
+    log_bg="#f8f9fc",
+    log_text="#3d4358",
+    progress_bar_bg="#e4e8f0",
+    progress_bar_fill="#3a5fc4",
+    sidebar_bg="#eff1f5",
+    sidebar_selected_bg="#dde3f2",
+    sidebar_selected_text="#3a5fc4",
+    nav_indicator="#3a5fc4",
+)
+
+
+# ---------------------------------------------------------------------------
+# Typography
+# ---------------------------------------------------------------------------
+
+FONT_FAMILY = "Inter, Segoe UI, Roboto, system-ui, sans-serif"
+FONT_SIZE_XS = 11
+FONT_SIZE_SM = 12
+FONT_SIZE_BASE = 14
+FONT_SIZE_MD = 16
+FONT_SIZE_LG = 20
+FONT_SIZE_XL = 26
+FONT_SIZE_DISPLAY = 34
+
+
+# ---------------------------------------------------------------------------
+# Spacing scale (pixels)
+# ---------------------------------------------------------------------------
+
+SPACE_XS = 4
+SPACE_SM = 8
+SPACE_MD = 12
+SPACE_LG = 16
+SPACE_XL = 24
+SPACE_2XL = 32
+SPACE_3XL = 48
+
+
+# ---------------------------------------------------------------------------
+# Border radii
+# ---------------------------------------------------------------------------
+
+RADIUS_SM = 6
+RADIUS_MD = 10
+RADIUS_LG = 16
+RADIUS_FULL = 999  # Pill-shaped
+
+
+# ---------------------------------------------------------------------------
+# Flet ColorScheme builders
+# ---------------------------------------------------------------------------
+
+
+def build_color_scheme(palette: _Palette) -> ft.ColorScheme:
+    """
+    Construct a ``ft.ColorScheme`` from a ``_Palette`` object.
+
+    Args:
+        palette: The ``DARK`` or ``LIGHT`` palette.
+
+    Returns:
+        A fully-populated Flet ``ColorScheme``.
+    """
+    return ft.ColorScheme(
+        primary=palette.accent,
+        on_primary=palette.accent_on,
+        primary_container=palette.accent_muted,
+        secondary=palette.accent,
+        on_secondary=palette.text_on_accent,
+        surface=palette.bg_surface,
+        on_surface=palette.text_primary,
+        on_surface_variant=palette.text_secondary,
+        error=palette.error,
+        on_error=palette.text_on_accent,
+        outline=palette.border,
+    )
+
+
+def build_text_theme() -> ft.TextTheme:
+    """
+    Construct a ``ft.TextTheme`` using the application's type scale.
+
+    Returns:
+        A Flet ``TextTheme`` with consistent font-size assignments.
+    """
+    return ft.TextTheme(
+        display_large=ft.TextStyle(size=FONT_SIZE_DISPLAY, weight=ft.FontWeight.W_700),
+        headline_large=ft.TextStyle(size=FONT_SIZE_XL, weight=ft.FontWeight.W_700),
+        headline_medium=ft.TextStyle(size=FONT_SIZE_LG, weight=ft.FontWeight.W_600),
+        title_large=ft.TextStyle(size=FONT_SIZE_MD, weight=ft.FontWeight.W_600),
+        title_medium=ft.TextStyle(size=FONT_SIZE_BASE, weight=ft.FontWeight.W_500),
+        body_large=ft.TextStyle(size=FONT_SIZE_BASE),
+        body_medium=ft.TextStyle(size=FONT_SIZE_SM),
+        label_large=ft.TextStyle(size=FONT_SIZE_SM, weight=ft.FontWeight.W_500),
+        label_medium=ft.TextStyle(size=FONT_SIZE_XS),
+    )
+
+
+def make_theme(dark: bool) -> ft.Theme:
+    """
+    Build a complete Flet ``Theme`` for the requested mode.
+
+    Args:
+        dark: True for dark-mode theme, False for light-mode theme.
+
+    Returns:
+        A configured ``ft.Theme`` instance.
+    """
+    palette = DARK if dark else LIGHT
+    return ft.Theme(
+        color_scheme=build_color_scheme(palette),
+        text_theme=build_text_theme(),
+        color_scheme_seed=palette.accent,
+        use_material3=True,
+    )
+
+
+def get_palette(page: ft.Page) -> _Palette:
+    """
+    Return the active colour palette for the given page.
+
+    Args:
+        page: The Flet ``Page`` instance.
+
+    Returns:
+        ``DARK`` or ``LIGHT`` depending on the page's theme mode.
+    """
+    return DARK if page.theme_mode == ft.ThemeMode.DARK else LIGHT

abogen/frontend/views/__init__.py

@@ -0,0 +1,6 @@
+"""Views sub-package for the Abogen Flet frontend."""
+from .dashboard import DashboardView
+from .settings import SettingsView
+from .queue_view import QueueView
+
+__all__ = ["DashboardView", "SettingsView", "QueueView"]

abogen/frontend/views/dashboard.py

@@ -0,0 +1,587 @@
+"""
+Dashboard view – the primary conversion screen.
+
+Hosts the file drop-zone, voice/speed/format controls, real-time log
+terminal, progress bar, and the Start/Cancel/Finish action row.
+
+All heavy work is delegated to ConversionBridge which runs on daemon
+threads and schedules UI updates back onto the Flet event loop.
+"""
+from __future__ import annotations
+
+import os
+import tempfile
+from pathlib import Path
+from typing import Optional
+
+import flet as ft
+
+from ..state import AppState
+from ..utils.helpers import (
+    detect_file_type, human_readable_size, format_number,
+    format_etr, grouped_voices, output_format_label,
+    subtitle_format_label, is_book_type, voice_lang_code, SUPPORTED_EXTENSIONS
+)
+from ..utils.theme import get_palette, RADIUS_MD, RADIUS_SM, SPACE_SM, SPACE_MD, SPACE_LG, SPACE_XL
+from ..utils.conversion_bridge import ConversionBridge
+from ..components import (
+    build_drop_zone, build_log_terminal, log_entry,
+    build_primary_button, build_secondary_button,
+    build_card, build_section_header, labelled_row, show_snack,
+)
+from abogen.constants import (
+    SUBTITLE_FORMATS, SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
+    LANGUAGE_DESCRIPTIONS, VOICES_INTERNAL,
+)
+from abogen.utils import get_gpu_acceleration, get_user_cache_path, calculate_text_length, clean_text
+
+
+class DashboardView:
+    """
+    The main conversion dashboard.
+
+    Instantiated once per Flet session and mounted as a ``ft.Column``
+    inside the page's content area.
+    """
+
+    def __init__(self, page: ft.Page, state: AppState) -> None:
+        self._page = page
+        self._state = state
+        self._bridge = ConversionBridge(page, state)
+
+        # Internal refs
+        self._log_list: Optional[ft.ListView] = None
+        self._progress_bar: Optional[ft.ProgressBar] = None
+        self._etr_label: Optional[ft.Text] = None
+        self._drop_zone_ref: Optional[ft.GestureDetector] = None
+        self._drop_zone_container: Optional[ft.Container] = None
+        self._file_picker: Optional[ft.FilePicker] = None
+
+        # Wire state callbacks
+        state.on_log = self._on_log
+        state.on_progress = self._on_progress
+        state.on_conversion_finished = self._on_finished
+
+        # Build UI refs
+        self._voice_dd: Optional[ft.Dropdown] = None
+        self._speed_slider: Optional[ft.Slider] = None
+        self._speed_label: Optional[ft.Text] = None
+        self._format_dd: Optional[ft.Dropdown] = None
+        self._subtitle_dd: Optional[ft.Dropdown] = None
+        self._subtitle_fmt_dd: Optional[ft.Dropdown] = None
+        self._gpu_switch: Optional[ft.Switch] = None
+        self._start_btn: Optional[ft.ElevatedButton] = None
+        self._cancel_btn: Optional[ft.OutlinedButton] = None
+        self._finish_col: Optional[ft.Column] = None
+        self._controls_col: Optional[ft.Column] = None
+        self._log_section: Optional[ft.Container] = None
+        self._progress_col: Optional[ft.Column] = None
+
+    # ------------------------------------------------------------------
+    # Build
+    # ------------------------------------------------------------------
+
+    def build(self) -> ft.Column:
+        """Return the complete dashboard column."""
+        p = self._page
+        dark = p.theme_mode == ft.ThemeMode.DARK
+        pal = get_palette(p)
+        if self._file_picker is None:
+            self._file_picker = ft.FilePicker()
+
+        # --- Drop zone ---
+        self._drop_zone_container = ft.Container()
+        self._refresh_drop_zone()
+
+        # --- Voice selector ---
+        voice_items = []
+        for lang_label, voices in grouped_voices():
+            voice_items.append(ft.dropdown.Option(key=f"__hdr_{lang_label}", text=f"── {lang_label} ──", disabled=True))
+            for v in voices:
+                voice_items.append(ft.dropdown.Option(key=v, text=v))
+
+        self._voice_dd = ft.Dropdown(
+            options=voice_items,
+            value=self._state.selected_voice,
+            on_select=self._on_voice_changed,
+            dense=True,
+            expand=True,
+            border_radius=RADIUS_SM,
+        )
+
+        # --- Speed slider ---
+        self._speed_label = ft.Text(f"{self._state.speed:.2f}", size=13, width=40)
+        self._speed_slider = ft.Slider(
+            min=0.1, max=2.0, value=self._state.speed,
+            divisions=190, label="{value}",
+            on_change=self._on_speed_changed,
+            expand=True,
+        )
+
+        # --- Format ---
+        self._format_dd = ft.Dropdown(
+            options=[ft.dropdown.Option(key=k, text=output_format_label(k))
+                     for k in ("wav", "flac", "mp3", "opus", "m4b")],
+            value=self._state.selected_format,
+            on_select=lambda e: self._set_field("selected_format", e.control.value),
+            dense=True, expand=True, border_radius=RADIUS_SM,
+        )
+
+        # --- Subtitle mode ---
+        sub_modes = ["Disabled", "Line", "Sentence", "Sentence + Comma",
+                     "Sentence + Highlighting"] + [f"{i} word{'s' if i > 1 else ''}" for i in range(1, 11)]
+        self._subtitle_dd = ft.Dropdown(
+            options=[ft.dropdown.Option(m) for m in sub_modes],
+            value=self._state.subtitle_mode,
+            on_select=lambda e: self._set_field("subtitle_mode", e.control.value),
+            dense=True, expand=True, border_radius=RADIUS_SM,
+        )
+
+        # --- Subtitle format ---
+        self._subtitle_fmt_dd = ft.Dropdown(
+            options=[ft.dropdown.Option(key=k, text=lbl) for k, lbl in SUBTITLE_FORMATS],
+            value=self._state.subtitle_format,
+            on_select=lambda e: self._set_field("subtitle_format", e.control.value),
+            dense=True, expand=True, border_radius=RADIUS_SM,
+        )
+
+        # --- GPU ---
+        self._gpu_switch = ft.Switch(
+            value=self._state.use_gpu, label="",
+            on_change=lambda e: self._set_field("use_gpu", e.control.value),
+            active_color="#5b8af5" if dark else "#3a5fc4",
+        )
+
+        # --- Log ---
+        log_lv = ft.ListView(expand=True, auto_scroll=True, spacing=1, padding=ft.Padding.all(8))
+        self._log_list = log_lv
+        bg_log = "#0d1117" if dark else "#f8f9fc"
+        bd_log = "#252a38" if dark else "#dce0ea"
+        self._log_section = ft.Container(
+            content=log_lv, bgcolor=bg_log,
+            border=ft.Border.all(1, bd_log),
+            border_radius=RADIUS_SM, height=220,
+            clip_behavior=ft.ClipBehavior.HARD_EDGE,
+            visible=False,
+        )
+
+        # --- Progress ---
+        fill = "#5b8af5" if dark else "#3a5fc4"
+        bg_p = "#1e2230" if dark else "#e4e8f0"
+        self._progress_bar = ft.ProgressBar(
+            value=0, color=fill, bgcolor=bg_p, height=8,
+            border_radius=ft.BorderRadius.all(4), expand=True,
+        )
+        self._etr_label = ft.Text("", size=11, color=pal.text_secondary, text_align=ft.TextAlign.CENTER)
+        self._progress_col = ft.Column([
+            ft.Row([self._progress_bar], spacing=0),
+            self._etr_label,
+        ], spacing=SPACE_SM, horizontal_alignment=ft.CrossAxisAlignment.CENTER, visible=False)
+
+        # --- Buttons ---
+        self._start_btn = build_primary_button(
+            "Start Conversion",
+            icon="play_arrow",
+            on_click=self._on_start,
+            page=p,
+        )
+        self._cancel_btn = build_secondary_button(
+            "Cancel", icon="stop",
+            on_click=self._on_cancel, page=p,
+        )
+        self._cancel_btn.visible = False
+
+        # --- Finish row ---
+        self._finish_col = ft.Column([
+            ft.Row([
+                build_secondary_button("Open File", icon="open_in_new",
+                                       on_click=self._on_open_file, page=p),
+                build_secondary_button("Go to Folder", icon="folder_open",
+                                       on_click=self._on_go_folder, page=p),
+                build_secondary_button("New Conversion", icon="refresh",
+                                       on_click=self._on_reset, page=p),
+            ], wrap=True, spacing=SPACE_SM, run_spacing=SPACE_SM),
+        ], visible=False)
+
+        # --- Controls column ---
+        self._controls_col = ft.Column([
+            build_section_header("Voice & Speed", icon="record_voice_over", page=p),
+            labelled_row("Voice", self._voice_dd, page=p),
+            labelled_row("Speed", ft.Row([self._speed_slider, self._speed_label], expand=True, spacing=SPACE_SM), page=p),
+            ft.Divider(height=1, color=pal.divider),
+            build_section_header("Output", icon="audio_file", page=p),
+            labelled_row("Format", self._format_dd, page=p),
+            labelled_row("Subtitles", self._subtitle_dd, page=p),
+            labelled_row("Subtitle Format", self._subtitle_fmt_dd, page=p),
+            ft.Divider(height=1, color=pal.divider),
+            build_section_header("Processing", icon="memory", page=p),
+            labelled_row("GPU Acceleration", self._gpu_switch, page=p),
+        ], spacing=SPACE_MD)
+
+        outer = ft.Column([
+            self._drop_zone_container,
+            ft.Container(height=SPACE_MD),
+            build_card(self._controls_col, page=p),
+            ft.Container(height=SPACE_SM),
+            self._log_section,
+            self._progress_col,
+            ft.Row([self._start_btn, self._cancel_btn], spacing=SPACE_SM, wrap=True),
+            self._finish_col,
+        ], spacing=SPACE_MD, expand=True, scroll=ft.ScrollMode.AUTO)
+
+        return outer
+
+    # ------------------------------------------------------------------
+    # Drop-zone management
+    # ------------------------------------------------------------------
+
+    def _refresh_drop_zone(self, *, accent: bool = False, error: bool = False, err_msg: str = "") -> None:
+        """Rebuild the drop-zone widget and update its container."""
+        p = self._page
+        s = self._state
+        fname = None; fsize = None; fchars = None
+        if s.selected_file and os.path.exists(s.selected_file):
+            disp = s.displayed_file_path or s.selected_file
+            fname = os.path.basename(disp)
+            try:
+                fsize = human_readable_size(os.path.getsize(s.selected_file))
+            except Exception:
+                fsize = ""
+            if s.char_count:
+                fchars = format_number(s.char_count)
+
+        label = err_msg if error else "Drag & drop your file here or click to browse"
+        sub = "Supports .txt · .epub · .pdf · .md · .srt · .ass · .vtt"
+
+        dz = build_drop_zone(
+            on_pick=self._open_file_picker,
+            label=label, sub_label=sub,
+            accent=accent, error=error,
+            filename=fname, file_size=fsize, char_count=fchars,
+            page=p,
+        )
+        if self._drop_zone_container is not None:
+            self._drop_zone_container.content = dz
+        self._drop_zone_ref = dz
+
+    # ------------------------------------------------------------------
+    # File picking
+    # ------------------------------------------------------------------
+
+    def _open_file_picker(self) -> None:
+        """Open the native file picker dialog."""
+        self._page.run_task(self._pick_files_async)
+
+    async def _pick_files_async(self) -> None:
+        """Run the file picker using Flet's async service API."""
+        picker = self._file_picker
+        if picker is None:
+            picker = ft.FilePicker()
+            self._file_picker = picker
+
+        try:
+            files = await picker.pick_files(
+                dialog_title="Select Input File",
+                file_type=ft.FilePickerFileType.CUSTOM,
+                allowed_extensions=["txt", "epub", "pdf", "md", "markdown", "srt", "ass", "vtt"],
+                allow_multiple=False,
+            )
+        except Exception as ex:
+            self._refresh_drop_zone(error=True, err_msg="Could not open file picker.")
+            show_snack(self._page, f"File picker error: {ex}", error=True)
+            self._page.update()
+            return
+        if not files:
+            return
+        file_path = files[0].path
+        if not file_path or not os.path.exists(file_path):
+            return
+        self._load_file(file_path)
+
+    def _load_file(self, file_path: str) -> None:
+        """Validate and load a file into the session state."""
+        from pathlib import Path as _Path
+        ext = _Path(file_path).suffix.lower()
+        if ext not in SUPPORTED_EXTENSIONS:
+            self._state.reset_file_state()
+            self._refresh_drop_zone(error=True, err_msg=f"Unsupported file type: {ext}")
+            self._page.update()
+            return
+
+        ftype = detect_file_type(file_path)
+        s = self._state
+
+        if ftype in ("epub", "pdf", "markdown"):
+            # For book types: extract text to temp cache
+            self._handle_book_file(file_path, ftype)
+        else:
+            # Plain text / subtitle files
+            s.selected_file = file_path
+            s.selected_file_type = ftype
+            s.displayed_file_path = file_path
+            try:
+                with open(file_path, "r", encoding="utf-8", errors="ignore") as f:
+                    text = f.read()
+                s.char_count = calculate_text_length(clean_text(text))
+            except Exception:
+                s.char_count = 0
+            self._refresh_drop_zone(accent=True)
+            self._update_subtitle_availability()
+            self._page.update()
+
+    def _handle_book_file(self, book_path: str, ftype: str) -> None:
+        """Extract text from epub/pdf/markdown and store as temp txt."""
+        import threading as _t
+        s = self._state
+
+        def _extract():
+            try:
+                from abogen.text_extractor import extract_from_path
+                chapters = extract_from_path(book_path, file_type=ftype)
+                combined = "\n\n".join(ch.text for ch in chapters if ch.text.strip())
+                cache_dir = get_user_cache_path()
+                base = os.path.splitext(os.path.basename(book_path))[0]
+                fd, tmp = tempfile.mkstemp(prefix=f"{base}_", suffix=".txt", dir=cache_dir)
+                os.close(fd)
+                with open(tmp, "w", encoding="utf-8") as f:
+                    f.write(combined)
+
+                s.selected_file = tmp
+                s.selected_file_type = ftype
+                s.selected_book_path = book_path
+                s.displayed_file_path = book_path
+                s.char_count = calculate_text_length(clean_text(combined))
+                s.selected_chapters = [f"ch_{i}" for i in range(len(chapters))]
+
+                self._refresh_drop_zone(accent=True)
+                self._update_subtitle_availability()
+                self._page.update()
+            except Exception as ex:
+                s.reset_file_state()
+                self._refresh_drop_zone(error=True, err_msg=f"Could not parse file: {ex}")
+                self._page.update()
+
+        _t.Thread(target=_extract, daemon=True).start()
+
+    # ------------------------------------------------------------------
+    # Control event handlers
+    # ------------------------------------------------------------------
+
+    def _set_field(self, attr: str, value) -> None:
+        setattr(self._state, attr, value)
+        self._state.persist_config()
+
+    def _on_voice_changed(self, e: ft.ControlEvent) -> None:
+        v = e.control.value or "af_heart"
+        self._state.selected_voice = v
+        self._state.selected_lang = voice_lang_code(v)
+        self._state.persist_config()
+        self._update_subtitle_availability()
+        self._page.update()
+
+    def _on_speed_changed(self, e: ft.ControlEvent) -> None:
+        val = round(float(e.control.value), 2)
+        self._state.speed = val
+        if self._speed_label:
+            self._speed_label.value = f"{val:.2f}"
+        self._state.persist_config()
+        self._page.update()
+
+    def _update_subtitle_availability(self) -> None:
+        """Enable or disable subtitle controls based on selected language."""
+        lang = self._state.selected_lang
+        enabled = lang in SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION
+        if self._subtitle_dd:
+            self._subtitle_dd.disabled = not enabled
+        if self._subtitle_fmt_dd:
+            self._subtitle_fmt_dd.disabled = not enabled
+
+    # ------------------------------------------------------------------
+    # Conversion control
+    # ------------------------------------------------------------------
+
+    def _on_start(self, _: ft.ControlEvent) -> None:
+        """Validate inputs and kick off conversion."""
+        s = self._state
+        if not s.selected_file or not os.path.exists(s.selected_file):
+            self._refresh_drop_zone(error=True, err_msg="Please select an input file first.")
+            self._page.update()
+            return
+
+        # Transition UI to converting state
+        self._set_converting_ui(True)
+
+        self._bridge.start(
+            input_file=s.selected_file,
+            voice=s.get_voice_formula(),
+            lang_code=s.selected_lang,
+            speed=s.speed,
+            output_format=s.selected_format,
+            subtitle_mode=s.subtitle_mode,
+            subtitle_format=s.subtitle_format,
+            use_gpu=s.use_gpu,
+            save_option=s.save_option,
+            output_folder=s.selected_output_folder,
+            replace_single_newlines=s.replace_single_newlines,
+            char_count=s.char_count,
+            save_chapters_separately=s.save_chapters_separately or False,
+            merge_chapters_at_end=True if s.merge_chapters_at_end is None else s.merge_chapters_at_end,
+            separate_chapters_format=s.separate_chapters_format,
+            silence_between_chapters=s.silence_duration,
+            max_subtitle_words=s.max_subtitle_words,
+            chapter_intro_delay=s.chapter_intro_delay,
+            read_title_intro=s.read_title_intro,
+            read_closing_outro=s.read_closing_outro,
+            auto_prefix_chapter_titles=s.auto_prefix_chapter_titles,
+            normalize_chapter_opening_caps=s.normalize_chapter_opening_caps,
+            tts_provider=s.tts_provider,
+            supertonic_total_steps=s.supertonic_total_steps,
+            chunk_level=s.chunk_level,
+            generate_epub3=s.generate_epub3,
+            word_substitutions_enabled=s.word_substitutions_enabled,
+            word_substitutions_list=s.word_substitutions_list,
+            case_sensitive_substitutions=s.case_sensitive_substitutions,
+            replace_all_caps=s.replace_all_caps,
+            replace_numerals=s.replace_numerals,
+            fix_nonstandard_punctuation=s.fix_nonstandard_punctuation,
+        )
+
+    def _on_cancel(self, _: ft.ControlEvent) -> None:
+        self._bridge.cancel()
+
+    def _set_converting_ui(self, converting: bool) -> None:
+        """Toggle UI between idle and converting states."""
+        if self._start_btn:
+            self._start_btn.visible = not converting
+        if self._cancel_btn:
+            self._cancel_btn.visible = converting
+        if self._controls_col:
+            self._controls_col.visible = not converting
+        if self._log_section:
+            self._log_section.visible = converting
+            if self._log_list:
+                self._log_list.controls.clear()
+        if self._progress_col:
+            self._progress_col.visible = converting
+            if self._progress_bar:
+                self._progress_bar.value = 0
+            if self._etr_label:
+                self._etr_label.value = "Estimating…"
+        if self._finish_col:
+            self._finish_col.visible = False
+        self._page.update()
+
+    # ------------------------------------------------------------------
+    # State callbacks (called from background thread via page.run_task)
+    # ------------------------------------------------------------------
+
+    def _on_log(self, message: str, level: str) -> None:
+        if self._log_list is None:
+            return
+        entry = log_entry(message, level, self._page)
+        self._log_list.controls.append(entry)
+        # Cap log lines
+        if len(self._log_list.controls) > 2000:
+            self._log_list.controls = self._log_list.controls[-1800:]
+        try:
+            self._page.update()
+        except Exception:
+            pass
+
+    def _on_progress(self, fraction: float, etr: Optional[float]) -> None:
+        if self._progress_bar:
+            self._progress_bar.value = min(fraction, 0.99)
+        if self._etr_label:
+            self._etr_label.value = format_etr(etr)
+        try:
+            self._page.update()
+        except Exception:
+            pass
+
+    def _on_finished(self, message: str, output_path: Optional[str]) -> None:
+        if self._progress_bar:
+            self._progress_bar.value = 1.0
+        if self._cancel_btn:
+            self._cancel_btn.visible = False
+
+        if message == "Cancelled":
+            # Restore idle state
+            self._set_converting_ui(False)
+            show_snack(self._page, "Conversion cancelled.", error=True)
+            return
+
+        if "failed" in message.lower() or "error" in message.lower():
+            self._log_on_log(message, "error")
+            self._set_converting_ui(False)
+            show_snack(self._page, f"Error: {message}", error=True)
+            return
+
+        # Success
+        if self._log_section:
+            self._log_section.visible = True
+        if self._progress_col:
+            self._progress_col.visible = False
+        if self._controls_col:
+            self._controls_col.visible = False
+        if self._finish_col:
+            self._finish_col.visible = True
+        if self._start_btn:
+            self._start_btn.visible = False
+        show_snack(self._page, "Conversion completed!")
+        try:
+            self._page.update()
+        except Exception:
+            pass
+
+    def _log_on_log(self, message: str, level: str) -> None:
+        self._on_log(message, level)
+
+    # ------------------------------------------------------------------
+    # Finish actions
+    # ------------------------------------------------------------------
+
+    def _on_open_file(self, _: ft.ControlEvent) -> None:
+        path = self._state.last_output_path
+        if path and os.path.exists(path):
+            import subprocess, platform
+            try:
+                if platform.system() == "Darwin":
+                    subprocess.Popen(["open", path])
+                elif platform.system() == "Windows":
+                    os.startfile(path)
+                else:
+                    subprocess.Popen(["xdg-open", path])
+            except Exception as ex:
+                show_snack(self._page, f"Cannot open file: {ex}", error=True)
+        else:
+            show_snack(self._page, "Output file not found.", error=True)
+
+    def _on_go_folder(self, _: ft.ControlEvent) -> None:
+        path = self._state.last_output_path
+        folder = os.path.dirname(path) if path and os.path.isfile(path) else path
+        if folder and os.path.isdir(folder):
+            import subprocess, platform
+            try:
+                if platform.system() == "Darwin":
+                    subprocess.Popen(["open", folder])
+                elif platform.system() == "Windows":
+                    subprocess.Popen(["explorer", folder])
+                else:
+                    subprocess.Popen(["xdg-open", folder])
+            except Exception as ex:
+                show_snack(self._page, f"Cannot open folder: {ex}", error=True)
+        else:
+            show_snack(self._page, "Output folder not found.", error=True)
+
+    def _on_reset(self, _: ft.ControlEvent) -> None:
+        self._state.reset_file_state()
+        self._state.reset_conversion_state()
+        self._refresh_drop_zone()
+        self._set_converting_ui(False)
+        if self._finish_col:
+            self._finish_col.visible = False
+        if self._controls_col:
+            self._controls_col.visible = True
+        if self._start_btn:
+            self._start_btn.visible = True
+        self._page.update()

abogen/frontend/views/queue_view.py

@@ -0,0 +1,154 @@
+"""
+Queue management view.
+
+Displays the current conversion queue, allowing the user to reorder,
+remove, and inspect queued items before starting batch processing.
+"""
+from __future__ import annotations
+from typing import Optional
+
+import flet as ft
+
+from ..state import AppState, ConversionJob
+from ..utils.theme import get_palette, RADIUS_SM, SPACE_SM, SPACE_MD, SPACE_LG
+from ..utils.helpers import safe_basename, output_format_label, format_number
+from ..components import (
+    build_card, build_section_header, build_primary_button,
+    build_secondary_button, show_snack, build_divider,
+    resolve_icon,
+)
+
+
+class QueueView:
+    """Queue manager view."""
+
+    def __init__(self, page: ft.Page, state: AppState) -> None:
+        self._page = page
+        self._state = state
+        self._list_col: Optional[ft.Column] = None
+
+    def build(self) -> ft.Column:
+        p = self._page
+        s = self._state
+        pal = get_palette(p)
+        dark = p.theme_mode == ft.ThemeMode.DARK
+
+        self._list_col = ft.Column(spacing=SPACE_SM)
+        self._refresh_list()
+
+        header = build_section_header("Conversion Queue",
+                                      icon="list_alt", page=p)
+
+        action_row = ft.Row([
+            build_primary_button(
+                "Start Queue",
+                icon="play_arrow",
+                on_click=self._on_start_queue,
+                page=p,
+                disabled=not s.queued_items,
+            ),
+            build_secondary_button(
+                "Clear All",
+                icon="delete_sweep",
+                on_click=self._on_clear_queue,
+                page=p,
+            ),
+        ], spacing=SPACE_SM, wrap=True)
+
+        queue_card = build_card(ft.Column([
+            header,
+            ft.Divider(height=1, color=pal.divider),
+            self._list_col,
+            ft.Container(height=SPACE_SM),
+            action_row,
+        ], spacing=SPACE_MD), page=p)
+
+        return ft.Column([queue_card], scroll=ft.ScrollMode.AUTO, expand=True)
+
+    # ------------------------------------------------------------------
+
+    def _refresh_list(self) -> None:
+        if self._list_col is None:
+            return
+        self._list_col.controls.clear()
+        s = self._state
+        pal = get_palette(self._page)
+        dark = self._page.theme_mode == ft.ThemeMode.DARK
+
+        if not s.queued_items:
+            self._list_col.controls.append(
+                ft.Text("No items in the queue.", size=13,
+                        color=pal.text_secondary,
+                        text_align=ft.TextAlign.CENTER)
+            )
+            return
+
+        for idx, job in enumerate(s.queued_items):
+            tile = self._build_job_tile(idx, job, dark, pal)
+            self._list_col.controls.append(tile)
+
+        try:
+            self._page.update()
+        except Exception:
+            pass
+
+    def _build_job_tile(self, idx: int, job: ConversionJob, dark: bool, pal) -> ft.Container:
+        """Build a single queue-item tile."""
+        bg = pal.bg_elevated
+        border_clr = pal.border
+        accent = "#5b8af5" if dark else "#3a5fc4"
+        text_primary = pal.text_primary
+        text_secondary = pal.text_secondary
+
+        def _remove(_):
+            self._state.queued_items.pop(idx)
+            self._refresh_list()
+
+        name = safe_basename(job.display_name or job.file_path)
+        details = (
+            f"Voice: {job.voice}  ·  Format: {output_format_label(job.output_format)}"
+            f"  ·  Speed: {job.speed:.2f}x  ·  Chars: {format_number(job.char_count)}"
+        )
+
+        return ft.Container(
+            content=ft.Row([
+                ft.Container(
+                    content=ft.Text(str(idx + 1), size=12, weight=ft.FontWeight.W_700,
+                                    color=accent),
+                    width=32,
+                ),
+                ft.Column([
+                    ft.Text(name, size=13, weight=ft.FontWeight.W_600, color=text_primary,
+                            no_wrap=True, overflow=ft.TextOverflow.ELLIPSIS),
+                    ft.Text(details, size=11, color=text_secondary),
+                ], expand=True, tight=True, spacing=2),
+                ft.IconButton(
+                    icon=resolve_icon("delete_outline"),
+                    icon_color=pal.error if hasattr(pal, "error") else "#e84e3c",
+                    icon_size=18,
+                    tooltip="Remove",
+                    on_click=_remove,
+                ),
+            ], vertical_alignment=ft.CrossAxisAlignment.CENTER, spacing=SPACE_SM),
+            bgcolor=bg,
+            border=ft.Border.all(1, border_clr),
+            border_radius=RADIUS_SM,
+            padding=ft.Padding.symmetric(horizontal=SPACE_MD, vertical=SPACE_SM),
+        )
+
+    # ------------------------------------------------------------------
+
+    def _on_start_queue(self, _: ft.ControlEvent) -> None:
+        if not self._state.queued_items:
+            show_snack(self._page, "Queue is empty.", error=True)
+            return
+        # Navigate to dashboard and trigger queue start
+        # This is wired in main.py via the nav controller
+        self._page.pubsub.send_all("start_queue")
+
+    def _on_clear_queue(self, _: ft.ControlEvent) -> None:
+        if not self._state.queued_items:
+            return
+        self._state.queued_items.clear()
+        self._refresh_list()
+        show_snack(self._page, "Queue cleared.")

abogen/frontend/views/settings.py

@@ -0,0 +1,305 @@
+"""
+Settings view – a categorised, scrollable settings page.
+
+Groups settings into collapsible cards:
+  - Output (format, save location, chapters)
+  - Text processing (newlines, caps, substitutions, numerals)
+  - Subtitle options
+  - TTS pipeline (provider, GPU, chunking)
+  - Integrations (Audiobookshelf, Calibre OPDS)
+"""
+from __future__ import annotations
+from typing import Optional
+
+import flet as ft
+
+from ..state import AppState
+from ..utils.theme import get_palette, RADIUS_MD, RADIUS_SM, SPACE_SM, SPACE_MD, SPACE_LG
+from ..utils.helpers import output_format_label, subtitle_format_label, SUPPORTED_EXTENSIONS
+from ..components import (
+    build_card, build_section_header, labelled_row, show_snack, build_divider,
+    build_primary_button,
+)
+from abogen.constants import SUBTITLE_FORMATS
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _dd(options, value, on_change, **kw):
+    """Compact dropdown factory."""
+    return ft.Dropdown(
+        options=[ft.dropdown.Option(key=k, text=v) for k, v in options],
+        value=value, on_select=on_change, dense=True,
+        border_radius=RADIUS_SM, expand=True, **kw
+    )
+
+
+def _sw(value, on_change, label=""):
+    return ft.Switch(value=value, on_change=on_change, label=label)
+
+
+class SettingsView:
+    """The full settings panel."""
+
+    def __init__(self, page: ft.Page, state: AppState) -> None:
+        self._page = page
+        self._state = state
+
+    def build(self) -> ft.Column:
+        p = self._page
+        s = self._state
+        pal = get_palette(p)
+
+        # ── Output card ──────────────────────────────────────────────
+        format_dd = _dd(
+            [(k, output_format_label(k)) for k in ("wav", "flac", "mp3", "opus", "m4b")],
+            s.selected_format,
+            lambda e: self._save("selected_format", e.control.value),
+        )
+        save_dd = _dd(
+            [
+                ("Save next to input file", "Save next to input file"),
+                ("Save to Desktop", "Save to Desktop"),
+                ("Choose output folder", "Choose output folder"),
+            ],
+            s.save_option,
+            lambda e: self._save("save_option", e.control.value),
+        )
+        chapters_sw = _sw(s.save_chapters_separately or False,
+                          lambda e: self._save("save_chapters_separately", e.control.value))
+        merge_sw = _sw(True if s.merge_chapters_at_end is None else s.merge_chapters_at_end,
+                       lambda e: self._save("merge_chapters_at_end", e.control.value))
+        sep_fmt_dd = _dd(
+            [(k, output_format_label(k)) for k in ("wav", "flac", "mp3", "opus")],
+            s.separate_chapters_format,
+            lambda e: self._save("separate_chapters_format", e.control.value),
+        )
+        epub3_sw = _sw(s.generate_epub3, lambda e: self._save("generate_epub3", e.control.value))
+
+        output_card = build_card(ft.Column([
+            build_section_header("Output", icon="audio_file", page=p),
+            labelled_row("Audio Format", format_dd, page=p),
+            labelled_row("Save Location", save_dd, page=p),
+            build_divider(p),
+            labelled_row("Save Chapters Separately", chapters_sw, page=p),
+            labelled_row("Merge at End", merge_sw, page=p),
+            labelled_row("Chapter Format", sep_fmt_dd, page=p),
+            labelled_row("Generate EPUB3", epub3_sw, page=p),
+        ], spacing=SPACE_MD), page=p)
+
+        # ── Text processing card ─────────────────────────────────────
+        newlines_sw = _sw(s.replace_single_newlines,
+                          lambda e: self._save("replace_single_newlines", e.control.value))
+        caps_sw = _sw(s.replace_all_caps, lambda e: self._save("replace_all_caps", e.control.value))
+        norm_sw = _sw(s.normalize_chapter_opening_caps,
+                      lambda e: self._save("normalize_chapter_opening_caps", e.control.value))
+        numerals_sw = _sw(s.replace_numerals, lambda e: self._save("replace_numerals", e.control.value))
+        punct_sw = _sw(s.fix_nonstandard_punctuation,
+                       lambda e: self._save("fix_nonstandard_punctuation", e.control.value))
+        wordsub_sw = _sw(s.word_substitutions_enabled,
+                         lambda e: self._save("word_substitutions_enabled", e.control.value))
+        wordsub_tf = ft.TextField(
+            value=s.word_substitutions_list,
+            multiline=True, min_lines=3, max_lines=6,
+            hint_text="word|replacement  (one per line)",
+            on_change=lambda e: self._save("word_substitutions_list", e.control.value),
+            expand=True, border_radius=RADIUS_SM, text_size=12,
+        )
+        case_sw = _sw(s.case_sensitive_substitutions,
+                      lambda e: self._save("case_sensitive_substitutions", e.control.value))
+        spacy_sw = _sw(s.use_spacy_segmentation,
+                       lambda e: self._save("use_spacy_segmentation", e.control.value))
+        chunk_dd = _dd(
+            [("paragraph", "Paragraph"), ("sentence", "Sentence")],
+            s.chunk_level,
+            lambda e: self._save("chunk_level", e.control.value),
+        )
+        title_intro_sw = _sw(s.read_title_intro, lambda e: self._save("read_title_intro", e.control.value))
+        outro_sw = _sw(s.read_closing_outro, lambda e: self._save("read_closing_outro", e.control.value))
+        prefix_sw = _sw(s.auto_prefix_chapter_titles,
+                        lambda e: self._save("auto_prefix_chapter_titles", e.control.value))
+
+        text_card = build_card(ft.Column([
+            build_section_header("Text Processing", icon="text_fields", page=p),
+            labelled_row("Replace Single Newlines", newlines_sw,
+                         tooltip="Replace single newlines with spaces before processing.", page=p),
+            labelled_row("Replace ALL CAPS Words", caps_sw, page=p),
+            labelled_row("Normalize Opening CAPS", norm_sw, page=p),
+            labelled_row("Replace Numerals (spoken)", numerals_sw, page=p),
+            labelled_row("Fix Non-standard Punctuation", punct_sw, page=p),
+            build_divider(p),
+            labelled_row("Word Substitutions", wordsub_sw, page=p),
+            labelled_row("Case Sensitive", case_sw, page=p),
+            ft.Text("Substitution rules  (word|replacement, one per line):",
+                    size=12, color=pal.text_secondary),
+            wordsub_tf,
+            build_divider(p),
+            build_section_header("Chapter Options", icon="library_books", page=p),
+            labelled_row("Announce Book Title (intro)", title_intro_sw, page=p),
+            labelled_row("Announce Book Title (outro)", outro_sw, page=p),
+            labelled_row("Auto-prefix Chapter Titles", prefix_sw, page=p),
+            labelled_row("Chunk Level", chunk_dd, page=p),
+            labelled_row("Use spaCy Segmentation", spacy_sw, page=p),
+        ], spacing=SPACE_MD), page=p)
+
+        # ── Subtitle card ─────────────────────────────────────────────
+        sub_modes = ["Disabled", "Line", "Sentence", "Sentence + Comma",
+                     "Sentence + Highlighting"] + [f"{i} word{'s' if i > 1 else ''}" for i in range(1, 11)]
+        sub_mode_dd = _dd(
+            [(m, m) for m in sub_modes],
+            s.subtitle_mode,
+            lambda e: self._save("subtitle_mode", e.control.value),
+        )
+        sub_fmt_dd = _dd(
+            [(k, lbl) for k, lbl in SUBTITLE_FORMATS],
+            s.subtitle_format,
+            lambda e: self._save("subtitle_format", e.control.value),
+        )
+
+        def _mk_mw_slider():
+            lbl = ft.Text(str(s.max_subtitle_words), size=12, width=36)
+            sl = ft.Slider(
+                min=1, max=200, value=s.max_subtitle_words, divisions=199, label="{value}",
+                expand=True,
+                on_change=lambda e: (self._save("max_subtitle_words", int(e.control.value)),
+                                     setattr(lbl, "value", str(int(e.control.value))),
+                                     self._page.update()),
+            )
+            return ft.Row([sl, lbl], expand=True, spacing=SPACE_SM)
+
+        sub_speed_dd = _dd(
+            [("tts", "TTS duration"), ("silence", "Silence detection")],
+            s.subtitle_speed_method,
+            lambda e: self._save("subtitle_speed_method", e.control.value),
+        )
+        silent_gaps_sw = _sw(s.use_silent_gaps,
+                             lambda e: self._save("use_silent_gaps", e.control.value))
+
+        subtitle_card = build_card(ft.Column([
+            build_section_header("Subtitles", icon="subtitles", page=p),
+            labelled_row("Mode", sub_mode_dd, page=p),
+            labelled_row("Format", sub_fmt_dd, page=p),
+            labelled_row("Max Words / Block", _mk_mw_slider(), page=p),
+            labelled_row("Speed Method", sub_speed_dd, page=p),
+            labelled_row("Silent Gaps", silent_gaps_sw, page=p),
+        ], spacing=SPACE_MD), page=p)
+
+        # ── Pipeline card ─────────────────────────────────────────────
+        provider_dd = _dd(
+            [("kokoro", "Kokoro (default)"), ("supertonic", "Supertonic")],
+            s.tts_provider,
+            lambda e: self._save("tts_provider", e.control.value),
+        )
+        gpu_sw = _sw(s.use_gpu, lambda e: self._save("use_gpu", e.control.value),
+                     label="GPU acceleration (if available)")
+
+        def _mk_steps_slider():
+            lbl = ft.Text(str(s.supertonic_total_steps), size=12, width=28)
+            sl = ft.Slider(
+                min=2, max=15, value=s.supertonic_total_steps, divisions=13,
+                label="{value}", expand=True,
+                on_change=lambda e: (self._save("supertonic_total_steps", int(e.control.value)),
+                                     setattr(lbl, "value", str(int(e.control.value))),
+                                     self._page.update()),
+            )
+            return ft.Row([sl, lbl], expand=True, spacing=SPACE_SM)
+
+        thresh_tf = ft.TextField(
+            value=str(s.speaker_analysis_threshold), width=80,
+            keyboard_type=ft.KeyboardType.NUMBER, border_radius=RADIUS_SM,
+            on_change=lambda e: self._save_int("speaker_analysis_threshold", e.control.value, 1, 25),
+        )
+        silence_tf = ft.TextField(
+            value=str(s.silence_duration), width=80,
+            keyboard_type=ft.KeyboardType.NUMBER, border_radius=RADIUS_SM,
+            on_change=lambda e: self._save_float("silence_duration", e.control.value, 0.0),
+        )
+        intro_tf = ft.TextField(
+            value=str(s.chapter_intro_delay), width=80,
+            keyboard_type=ft.KeyboardType.NUMBER, border_radius=RADIUS_SM,
+            on_change=lambda e: self._save_float("chapter_intro_delay", e.control.value, 0.0),
+        )
+
+        pipeline_card = build_card(ft.Column([
+            build_section_header("TTS Pipeline", icon="settings", page=p),
+            labelled_row("Provider", provider_dd, page=p),
+            labelled_row("GPU Acceleration", gpu_sw, page=p),
+            labelled_row("Supertonic Steps", _mk_steps_slider(), page=p),
+            build_divider(p),
+            labelled_row("Speaker Analysis Threshold", thresh_tf, page=p),
+            labelled_row("Silence Between Chapters (s)", silence_tf, page=p),
+            labelled_row("Chapter Intro Delay (s)", intro_tf, page=p),
+        ], spacing=SPACE_MD), page=p)
+
+        # ── Integration card (Audiobookshelf) ─────────────────────────
+        abs_enabled_sw = _sw(s.audiobookshelf_enabled,
+                             lambda e: self._save("audiobookshelf_enabled", e.control.value))
+        abs_url_tf = ft.TextField(value=s.audiobookshelf_base_url, hint_text="http://abs-server:13378",
+                                  expand=True, border_radius=RADIUS_SM, text_size=12,
+                                  on_change=lambda e: self._save("audiobookshelf_base_url", e.control.value))
+        abs_token_tf = ft.TextField(value=s.audiobookshelf_api_token, password=True,
+                                    can_reveal_password=True, expand=True,
+                                    border_radius=RADIUS_SM, text_size=12,
+                                    on_change=lambda e: self._save("audiobookshelf_api_token", e.control.value))
+        abs_lib_tf = ft.TextField(value=s.audiobookshelf_library_id, hint_text="Library ID",
+                                  expand=True, border_radius=RADIUS_SM, text_size=12,
+                                  on_change=lambda e: self._save("audiobookshelf_library_id", e.control.value))
+        abs_auto_sw = _sw(s.audiobookshelf_auto_send,
+                          lambda e: self._save("audiobookshelf_auto_send", e.control.value))
+
+        integ_card = build_card(ft.Column([
+            build_section_header("Audiobookshelf Integration",
+                                 icon="cloud_upload", page=p),
+            labelled_row("Enabled", abs_enabled_sw, page=p),
+            labelled_row("Server URL", abs_url_tf, page=p),
+            labelled_row("API Token", abs_token_tf, page=p),
+            labelled_row("Library ID", abs_lib_tf, page=p),
+            labelled_row("Auto-upload on finish", abs_auto_sw, page=p),
+        ], spacing=SPACE_MD), page=p)
+
+        save_btn = build_primary_button(
+            "Save Settings", icon="save",
+            on_click=self._on_save, page=p,
+        )
+
+        return ft.Column([
+            output_card,
+            ft.Container(height=SPACE_MD),
+            text_card,
+            ft.Container(height=SPACE_MD),
+            subtitle_card,
+            ft.Container(height=SPACE_MD),
+            pipeline_card,
+            ft.Container(height=SPACE_MD),
+            integ_card,
+            ft.Container(height=SPACE_LG),
+            save_btn,
+            ft.Container(height=SPACE_LG),
+        ], spacing=0, scroll=ft.ScrollMode.AUTO, expand=True)
+
+    # ------------------------------------------------------------------
+    # Helpers
+    # ------------------------------------------------------------------
+
+    def _save(self, attr: str, value) -> None:
+        setattr(self._state, attr, value)
+
+    def _save_int(self, attr: str, raw: str, lo: int, hi: int) -> None:
+        try:
+            v = max(lo, min(hi, int(raw)))
+            setattr(self._state, attr, v)
+        except ValueError:
+            pass
+
+    def _save_float(self, attr: str, raw: str, lo: float) -> None:
+        try:
+            v = max(lo, float(raw))
+            setattr(self._state, attr, v)
+        except ValueError:
+            pass
+
+    def _on_save(self, _: ft.ControlEvent) -> None:
+        self._state.persist_config()
+        show_snack(self._page, "Settings saved.")

abogen/predownload_gui.py

@@ -21,8 +21,7 @@ from PyQt6.QtWidgets import (
 )
 from PyQt6.QtCore import QThread, pyqtSignal
 
-from abogen.constants import COLORS
-from abogen.tts_backend_registry import get_metadata
+from abogen.constants import COLORS, VOICES_INTERNAL
 from abogen.spacy_utils import SPACY_MODELS
 import abogen.hf_tracker
 
@@ -115,7 +114,7 @@ class PreDownloadWorker(QThread):
             self._voices_success = False
             return
 
-        voice_list = get_metadata("kokoro").voices
+        voice_list = VOICES_INTERNAL
         for idx, voice in enumerate(voice_list, start=1):
             if self._cancelled:
                 self._voices_success = False
@@ -463,14 +462,14 @@ class PreDownloadDialog(QDialog):
         try:
             from huggingface_hub import try_to_load_from_cache
 
-            for voice in get_metadata("kokoro").voices:
+            for voice in VOICES_INTERNAL:
                 if not try_to_load_from_cache(
                     repo_id="hexgrad/Kokoro-82M", filename=f"voices/{voice}.pt"
                 ):
                     missing.append(voice)
         except Exception:
             # If HF missing, report all as missing
-            return False, list(get_metadata("kokoro").voices)
+            return False, list(VOICES_INTERNAL)
         return (len(missing) == 0), missing
 
     def _check_kokoro_model(self) -> bool:

abogen/pyqt/conversion.py

@@ -5,7 +5,6 @@ import hashlib  # For generating unique cache filenames
 from platformdirs import user_desktop_dir
 from PyQt6.QtCore import QThread, pyqtSignal, Qt, QTimer
 from PyQt6.QtWidgets import QCheckBox, QVBoxLayout, QDialog, QLabel, QDialogButtonBox
-import numpy as np
 import soundfile as sf
 from abogen.utils import (
     create_process,
@@ -260,7 +259,8 @@ class ConversionThread(QThread):
         output_folder,
         subtitle_mode,
         output_format,
-        backend,
+        np_module,
+        kpipeline_class,
         start_time,
         total_char_count,
         use_gpu=True,
@@ -270,7 +270,8 @@ class ConversionThread(QThread):
         super().__init__()
         self._chapter_options_event = threading.Event()
         self._timestamp_response_event = threading.Event()
-        self.backend = backend
+        self.np = np_module
+        self.KPipeline = kpipeline_class
         self.file_name = file_name
         self.lang_code = lang_code
         self.speed = speed
@@ -489,6 +490,19 @@ class ConversionThread(QThread):
 
             self.log_updated.emit(("\nInitializing TTS pipeline...", "grey"))
 
+            # Set device based on use_gpu setting and platform
+            if self.use_gpu:
+                if platform.system() == "Darwin" and platform.processor() == "arm":
+                    device = "mps"  # Use MPS for Apple Silicon
+                else:
+                    device = "cuda"  # Use CUDA for other platforms
+            else:
+                device = "cpu"
+
+            tts = self.KPipeline(
+                lang_code=self.lang_code, repo_id="hexgrad/Kokoro-82M", device=device
+            )
+
             # Check if the input is a subtitle file or timestamp text file
             is_subtitle_file = False
             is_timestamp_text = False
@@ -524,7 +538,7 @@ class ConversionThread(QThread):
 
             # Process subtitle files separately
             if is_subtitle_file or is_timestamp_text:
-                self._process_subtitle_file(self.backend, base_path, is_timestamp_text)
+                self._process_subtitle_file(tts, base_path, is_timestamp_text)
                 return
 
             if self.is_direct_text:
@@ -1057,7 +1071,7 @@ class ConversionThread(QThread):
                 for segment_idx, (voice_name, segment_text) in enumerate(voice_segments):
                     # Load voice for this segment (with caching)
                     try:
-                        loaded_voice = self.load_voice_cached(voice_name, self.backend)
+                        loaded_voice = self.load_voice_cached(voice_name, tts)
                         if segment_idx > 0:
                             voice_display = voice_name if len(voice_name) < 50 else voice_name[:47] + "..."
                             self.log_updated.emit((f"  → Voice: {voice_display}", "grey"))
@@ -1066,7 +1080,7 @@ class ConversionThread(QThread):
                             (f"⚠ Voice loading error for '{voice_name}', continuing with previous", "orange")
                         )
                         if segment_idx == 0:
-                            loaded_voice = self.load_voice_cached(self.voice, self.backend)
+                            loaded_voice = self.load_voice_cached(self.voice, tts)
 
                     # Determine if spaCy segmentation should be used for PRE-TTS segmentation
                     # Only non-English languages use spaCy for pre-segmentation
@@ -1152,7 +1166,7 @@ class ConversionThread(QThread):
                         print("Using split pattern: (unprintable)")
 
                     for text_segment in text_segments:
-                        for result in self.backend(
+                        for result in tts(
                             text_segment,
                             voice=loaded_voice,
                             speed=self.speed,
@@ -1354,7 +1368,7 @@ class ConversionThread(QThread):
                     silence_samples = int(
                         self.silence_duration * 24000
                     )  # Silence duration at 24,000 Hz
-                    silence_audio = np.zeros(silence_samples, dtype="float32")
+                    silence_audio = self.np.zeros(silence_samples, dtype="float32")
                     silence_bytes = silence_audio.tobytes()
 
                     if merged_out_file:
@@ -1693,7 +1707,7 @@ class ConversionThread(QThread):
             max_end_time = max(
                 (end for _, end, _ in subtitles if end is not None), default=0
             )
-            audio_buffer = np.zeros(
+            audio_buffer = self.np.zeros(
                 int(max_end_time * rate) + rate, dtype="float32"
             )
 
@@ -1757,7 +1771,7 @@ class ConversionThread(QThread):
                 # Generate TTS audio
                 tts_results = [
                     r
-                    for r in self.backend(
+                    for r in tts(
                         processed_text,
                         voice=loaded_voice,
                         speed=self.speed,
@@ -1775,11 +1789,11 @@ class ConversionThread(QThread):
 
                 # Concatenate audio and determine duration
                 full_audio = (
-                    np.concatenate(
+                    self.np.concatenate(
                         [a.numpy() if hasattr(a, "numpy") else a for a in audio_chunks]
                     )
                     if audio_chunks
-                    else np.zeros(
+                    else self.np.zeros(
                         int((subtitle_duration or 0) * rate), dtype="float32"
                     )
                 )
@@ -1813,8 +1827,8 @@ class ConversionThread(QThread):
                         num_stages = max(
                             1,
                             int(
-                                np.ceil(
-                                    np.log(speed_factor) / np.log(2.0)
+                                self.np.ceil(
+                                    self.np.log(speed_factor) / self.np.log(2.0)
                                 )
                             ),
                         )
@@ -1847,7 +1861,7 @@ class ConversionThread(QThread):
                             stdout=subprocess.PIPE,
                             stderr=subprocess.PIPE,
                         )
-                        full_audio = np.frombuffer(
+                        full_audio = self.np.frombuffer(
                             speed_proc.communicate(input=full_audio.tobytes())[0],
                             dtype="float32",
                         )
@@ -1861,7 +1875,7 @@ class ConversionThread(QThread):
 
                         tts_results = [
                             r
-                            for r in self.backend(
+                            for r in tts(
                                 processed_text,
                                 voice=loaded_voice,
                                 speed=new_speed,
@@ -1872,14 +1886,14 @@ class ConversionThread(QThread):
                         audio_chunks = [r.audio for r in tts_results]
 
                         full_audio = (
-                            np.concatenate(
+                            self.np.concatenate(
                                 [
                                     a.numpy() if hasattr(a, "numpy") else a
                                     for a in audio_chunks
                                 ]
                             )
                             if audio_chunks
-                            else np.zeros(
+                            else self.np.zeros(
                                 int(subtitle_duration * rate), dtype="float32"
                             )
                         )
@@ -1896,10 +1910,10 @@ class ConversionThread(QThread):
                 # Pad or trim to subtitle duration
                 target_samples = int(subtitle_duration * rate)
                 if len(full_audio) < target_samples:
-                    full_audio = np.concatenate(
+                    full_audio = self.np.concatenate(
                         [
                             full_audio,
-                            np.zeros(
+                            self.np.zeros(
                                 target_samples - len(full_audio), dtype="float32"
                             ),
                         ]
@@ -1912,10 +1926,10 @@ class ConversionThread(QThread):
                 end_sample = start_sample + len(full_audio)
                 if end_sample > len(audio_buffer):
                     # Extend buffer if needed
-                    audio_buffer = np.concatenate(
+                    audio_buffer = self.np.concatenate(
                         [
                             audio_buffer,
-                            np.zeros(
+                            self.np.zeros(
                                 end_sample - len(audio_buffer), dtype="float32"
                             ),
                         ]
@@ -1957,7 +1971,7 @@ class ConversionThread(QThread):
                 self.progress_updated.emit(percent, etr_str)
 
             # Normalize audio buffer to prevent clipping from mixed overlaps
-            max_amplitude = np.abs(audio_buffer).max()
+            max_amplitude = self.np.abs(audio_buffer).max()
             if max_amplitude > 1.0:
                 self.log_updated.emit(
                     f"\n  -> Normalizing audio (peak: {max_amplitude:.2f})"
@@ -2426,7 +2440,8 @@ class VoicePreviewThread(QThread):
 
     def __init__(
         self,
-        backend,
+        np_module,
+        kpipeline_class,
         lang_code,
         voice,
         speed,
@@ -2434,7 +2449,8 @@ class VoicePreviewThread(QThread):
         parent=None,
     ):
         super().__init__(parent)
-        self.backend = backend
+        self.np_module = np_module
+        self.kpipeline_class = kpipeline_class
         self.lang_code = lang_code
         self.voice = voice
         self.speed = speed
@@ -2468,19 +2484,31 @@ class VoicePreviewThread(QThread):
         # Generate the preview and save to cache
         try:
 
+            # Set device based on use_gpu setting and platform
+            if self.use_gpu:
+                if platform.system() == "Darwin" and platform.processor() == "arm":
+                    device = "mps"  # Use MPS for Apple Silicon
+                else:
+                    device = "cuda"  # Use CUDA for other platforms
+            else:
+                device = "cpu"
+
+            tts = self.kpipeline_class(
+                lang_code=self.lang_code, repo_id="hexgrad/Kokoro-82M", device=device
+            )
             # Enable voice formula support for preview
             if "*" in self.voice:
-                loaded_voice = get_new_voice(self.backend, self.voice, self.use_gpu)
+                loaded_voice = get_new_voice(tts, self.voice, self.use_gpu)
             else:
                 loaded_voice = self.voice
             sample_text = get_sample_voice_text(self.lang_code)
             audio_segments = []
-            for result in self.backend(
+            for result in tts(
                 sample_text, voice=loaded_voice, speed=self.speed, split_pattern=None
             ):
                 audio_segments.append(result.audio)
             if audio_segments:
-                audio = np.concatenate(audio_segments)
+                audio = self.np_module.concatenate(audio_segments)
                 # Save directly to the cache path
                 sf.write(self.cache_path, audio, 24000)
                 self.temp_wav = self.cache_path

abogen/pyqt/gui.py

@@ -82,11 +82,11 @@ from abogen.constants import (
     GITHUB_URL,
     PROGRAM_DESCRIPTION,
     LANGUAGE_DESCRIPTIONS,
+    VOICES_INTERNAL,
     SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
     COLORS,
     SUBTITLE_FORMATS,
 )
-from abogen.tts_backend_registry import get_metadata
 import threading
 from abogen.pyqt.voice_formula_gui import VoiceFormulaDialog
 from abogen.voice_profiles import load_profiles
@@ -1873,7 +1873,7 @@ class abogen(QWidget):
         for pname in load_profiles().keys():
             self.voice_combo.addItem(profile_icon, pname, f"profile:{pname}")
         # re-add voices
-        for v in get_metadata("kokoro").voices:
+        for v in VOICES_INTERNAL:
             icon = QIcon()
             flag_path = get_resource_path("abogen.assets.flags", f"{v[0]}.png")
             if flag_path and os.path.exists(flag_path):
@@ -2316,9 +2316,9 @@ class abogen(QWidget):
             file_size_str = "Unknown"
 
         # pipeline_loaded_callback remains unchanged
-        def pipeline_loaded_callback(backend, error):
+        def pipeline_loaded_callback(np_module, kpipeline_class, error):
             if error:
-                self.update_log((f"Error loading TTS backend: {error}", "red"))
+                self.update_log((f"Error loading numpy or KPipeline: {error}", "red"))
                 prevent_sleep_end()
                 return
 
@@ -2341,7 +2341,8 @@ class abogen(QWidget):
                 self.selected_output_folder,
                 subtitle_mode=actual_subtitle_mode,
                 output_format=self.selected_format,
-                backend=backend,
+                np_module=np_module,
+                kpipeline_class=kpipeline_class,
                 start_time=self.start_time,
                 total_char_count=self.char_count,
                 use_gpu=self.gpu_ok,
@@ -2425,20 +2426,7 @@ class abogen(QWidget):
             self.gpu_ok = gpu_ok
             self.update_log((gpu_msg, gpu_ok))
             self.update_log("Loading modules...")
-
-            # Determine device based on GPU availability
-            if gpu_ok:
-                if platform.system() == "Darwin" and platform.processor() == "arm":
-                    device = "mps"
-                else:
-                    device = "cuda"
-            else:
-                device = "cpu"
-
-            lang_code = self.selected_lang or "a"
-            load_thread = LoadPipelineThread(
-                pipeline_loaded_callback, lang_code=lang_code, device=device
-            )
+            load_thread = LoadPipelineThread(pipeline_loaded_callback)
             load_thread.start()
 
         threading.Thread(target=gpu_and_load, daemon=True).start()
@@ -2875,27 +2863,18 @@ class abogen(QWidget):
             )
             self.loading_movie.start()
 
-        # Determine device based on GPU availability
-        if self.gpu_ok:
-            if platform.system() == "Darwin" and platform.processor() == "arm":
-                device = "mps"
-            else:
-                device = "cuda"
-        else:
-            device = "cpu"
+        def pipeline_loaded_callback(np_module, kpipeline_class, error):
+            self._on_pipeline_loaded_for_preview(np_module, kpipeline_class, error)
 
-        lang = self.selected_lang or "a"
-        load_thread = LoadPipelineThread(
-            self._on_pipeline_loaded_for_preview, lang_code=lang, device=device
-        )
+        load_thread = LoadPipelineThread(pipeline_loaded_callback)
         load_thread.start()
 
-    def _on_pipeline_loaded_for_preview(self, backend, error):
+    def _on_pipeline_loaded_for_preview(self, np_module, kpipeline_class, error):
         # stop loading animation and restore icon on error
         if error:
             self.loading_movie.stop()
             self._show_error_message_box(
-                "Loading Error", f"Error loading TTS backend: {error}"
+                "Loading Error", f"Error loading numpy or KPipeline: {error}"
             )
             self.btn_preview.setIcon(self.play_icon)
             self.btn_preview.setEnabled(True)
@@ -2933,7 +2912,7 @@ class abogen(QWidget):
         gpu_msg, gpu_ok = get_gpu_acceleration(self.use_gpu)
 
         self.preview_thread = VoicePreviewThread(
-            backend, lang, voice, speed, gpu_ok
+            np_module, kpipeline_class, lang, voice, speed, gpu_ok
         )
         self.preview_thread.finished.connect(self._play_preview_audio)
         self.preview_thread.error.connect(self._preview_error)

abogen/pyqt/predownload_gui.py

@@ -21,8 +21,7 @@ from PyQt6.QtWidgets import (
 )
 from PyQt6.QtCore import QThread, pyqtSignal
 
-from abogen.constants import COLORS
-from abogen.tts_backend_registry import get_metadata
+from abogen.constants import COLORS, VOICES_INTERNAL
 from abogen.spacy_utils import SPACY_MODELS
 import abogen.hf_tracker
 
@@ -115,7 +114,7 @@ class PreDownloadWorker(QThread):
             self._voices_success = False
             return
 
-        voice_list = get_metadata("kokoro").voices
+        voice_list = VOICES_INTERNAL
         for idx, voice in enumerate(voice_list, start=1):
             if self._cancelled:
                 self._voices_success = False
@@ -463,14 +462,14 @@ class PreDownloadDialog(QDialog):
         try:
             from huggingface_hub import try_to_load_from_cache
 
-            for voice in get_metadata("kokoro").voices:
+            for voice in VOICES_INTERNAL:
                 if not try_to_load_from_cache(
                     repo_id="hexgrad/Kokoro-82M", filename=f"voices/{voice}.pt"
                 ):
                     missing.append(voice)
         except Exception:
             # If HF missing, report all as missing
-            return False, list(get_metadata("kokoro").voices)
+            return False, list(VOICES_INTERNAL)
         return (len(missing) == 0), missing
 
     def _check_kokoro_model(self) -> bool:

abogen/pyqt/voice_formula_gui.py

@@ -28,11 +28,11 @@ from PyQt6.QtWidgets import (
 from PyQt6.QtCore import Qt, QTimer, QPoint, QRect, QSize
 from PyQt6.QtGui import QPixmap, QIcon, QAction
 from abogen.constants import (
+    VOICES_INTERNAL,
     SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
     LANGUAGE_DESCRIPTIONS,
     COLORS,
 )
-from abogen.tts_backend_registry import get_metadata
 import re
 import platform
 from abogen.utils import get_resource_path
@@ -179,7 +179,7 @@ class VoiceMixer(QWidget):
         layout.addWidget(QLabel(name), alignment=Qt.AlignmentFlag.AlignCenter)
 
         # Voice name label with gender icon
-        is_female = self.voice_name in get_metadata("kokoro").voices and self.voice_name[1] == "f"
+        is_female = self.voice_name in VOICES_INTERNAL and self.voice_name[1] == "f"
 
         # Icons layout (flag and gender)
         icons_layout = QHBoxLayout()
@@ -772,7 +772,7 @@ class VoiceFormulaDialog(QDialog):
 
     def add_voices(self, initial_state):
         first_enabled_voice = None
-        for voice in get_metadata("kokoro").voices:
+        for voice in VOICES_INTERNAL:
             language_code = voice[0]  # First character is the language code
             matching_voice = next(
                 (item for item in initial_state if item[0] == voice), None

abogen/subtitle_utils.py

@@ -466,7 +466,7 @@ def sanitize_name_for_os(name, is_folder=True):
 
 
 def validate_voice_name(voice_name):
-    """Validate voice name against available voices (case-insensitive).
+    """Validate voice name against VOICES_INTERNAL list (case-insensitive).
     Handles both single voices and formulas like 'af_heart*0.5 + am_echo*0.5'.
 
     Args:
@@ -477,10 +477,10 @@ def validate_voice_name(voice_name):
             - is_valid: True if all voices in the name/formula are valid
             - invalid_voice_name: The first invalid voice found, or None if all valid
     """
-    from abogen.tts_backend_registry import get_metadata
+    from abogen.constants import VOICES_INTERNAL
 
     # Create case-insensitive lookup set (done once per call)
-    voice_lookup_lower = {v.lower() for v in get_metadata("kokoro").voices}
+    voice_lookup_lower = {v.lower() for v in VOICES_INTERNAL}
     voice_name = voice_name.strip()
 
     # Check if it's a formula (contains *)
@@ -505,7 +505,7 @@ def split_text_by_voice_markers(text, default_voice):
     """Split text by voice markers, returning list of (voice, text) tuples.
 
     IMPORTANT: Returns the last voice used so it can persist across chapters.
-    Voice names are normalized to lowercase to match canonical voice names.
+    Voice names are normalized to lowercase to match VOICES_INTERNAL.
 
     Args:
         text: Text potentially containing <<VOICE:name>> markers
@@ -518,7 +518,7 @@ def split_text_by_voice_markers(text, default_voice):
             - valid_count: Number of valid voice markers processed
             - invalid_count: Number of invalid voice markers skipped
     """
-    from abogen.tts_backend_registry import get_metadata
+    from abogen.constants import VOICES_INTERNAL
 
     voice_splits = list(_VOICE_MARKER_SEARCH_PATTERN.finditer(text))
 
@@ -560,7 +560,7 @@ def split_text_by_voice_markers(text, default_voice):
                         # Find the canonical (lowercase) voice name
                         voice_part_lower = voice_part.strip().lower()
                         canonical_voice = next(
-                            (v for v in get_metadata("kokoro").voices if v.lower() == voice_part_lower),
+                            (v for v in VOICES_INTERNAL if v.lower() == voice_part_lower),
                             voice_part.strip()
                         )
                         normalized_parts.append(f"{canonical_voice}*{weight.strip()}")
@@ -569,7 +569,7 @@ def split_text_by_voice_markers(text, default_voice):
                 # Find the canonical (lowercase) voice name
                 voice_name_lower = voice_name.lower()
                 current_voice = next(
-                    (v for v in get_metadata("kokoro").voices if v.lower() == voice_name_lower),
+                    (v for v in VOICES_INTERNAL if v.lower() == voice_name_lower),
                     voice_name
                 )
             valid_markers += 1

abogen/tts_backend.py

@@ -1,89 +0,0 @@
-"""
-TTS Backend Interface
-
-This module defines the protocol for TTS backends and the
-metadata model that describes a backend implementation.
-"""
-
-from dataclasses import dataclass
-from typing import Protocol, List, Dict, Any
-
-
-@dataclass(frozen=True)
-class TTSBackendMetadata:
-    """
-    Immutable metadata describing a TTS backend implementation.
-
-    Attributes:
-        id: Unique backend identifier (e.g. ``"kokoro"``, ``"supertonic"``).
-        name: Human-readable display name.
-        description: Short description of the backend.
-        voices: Tuple of supported voice identifiers.
-    """
-
-    id: str
-    name: str
-    description: str
-    voices: tuple[str, ...] = ()
-
-
-class TTSBackend(Protocol):
-    """
-    Protocol for TTS backends.
-
-    All TTS backends must implement this interface to be compatible
-    with the application.
-    """
-
-    @property
-    def metadata(self) -> TTSBackendMetadata:
-        ...
-
-    def __init__(self, **kwargs) -> None:
-        """
-        Initialize the TTS backend.
-
-        Args:
-            **kwargs: Backend-specific configuration parameters
-        """
-        ...
-
-    def synthesize(self, text: str, **kwargs) -> bytes:
-        """
-        Synthesize speech from text.
-
-        Args:
-            text: Text to synthesize
-            **kwargs: Additional parameters for synthesis
-
-        Returns:
-            Audio data as bytes
-        """
-        ...
-
-    def get_available_voices(self) -> List[str]:
-        """
-        Get list of available voices.
-
-        Returns:
-            List of voice identifiers
-        """
-        ...
-
-    def get_supported_formats(self) -> List[str]:
-        """
-        Get list of supported audio formats.
-
-        Returns:
-            List of supported audio formats
-        """
-        ...
-
-    def get_info(self) -> Dict[str, Any]:
-        """
-        Get backend information.
-
-        Returns:
-            Dictionary with backend information
-        """
-        ...

abogen/tts_backend_registry.py

@@ -1,146 +0,0 @@
-"""
-TTS Backend Registry
-
-Provides a global registry for TTS backend factories.
-Backends register themselves with metadata and a factory callable.
-The registry is universal and does not know about backend constructors.
-"""
-
-from typing import Callable, Any
-
-from abogen.tts_backend import TTSBackend, TTSBackendMetadata
-
-
-class TTSBackendRegistry:
-    """Registry of TTS backend factories.
-
-    Stores metadata and factory callables for registered backends.
-    """
-
-    def __init__(self) -> None:
-        self._backends: dict[str, TTSBackendMetadata] = {}
-        self._factories: dict[str, Callable[..., TTSBackend]] = {}
-
-    def register(
-        self,
-        metadata: TTSBackendMetadata,
-        factory: Callable[..., TTSBackend],
-    ) -> None:
-        """Register a backend with its metadata and factory callable."""
-        self._backends[metadata.id] = metadata
-        self._factories[metadata.id] = factory
-
-    def is_registered(self, backend_id: str) -> bool:
-        """Return True if a backend with the given id is registered."""
-        return backend_id in self._backends
-
-    def list_backends(self) -> list[TTSBackendMetadata]:
-        """Return metadata for all registered backends."""
-        return list(self._backends.values())
-
-    def get_metadata(self, backend_id: str) -> TTSBackendMetadata:
-        """Get metadata for a specific backend.
-
-        Raises:
-            KeyError: If backend with given id is not registered.
-        """
-        if backend_id not in self._backends:
-            raise KeyError(f"Unknown backend: {backend_id}")
-        return self._backends[backend_id]
-
-    def create_backend(self, backend_id: str, **kwargs: Any) -> TTSBackend:
-        """Create a backend instance by id.
-
-        Raises:
-            KeyError: If backend with given id is not registered.
-        """
-        if backend_id not in self._factories:
-            raise KeyError(f"Unknown backend: {backend_id}")
-        return self._factories[backend_id](**kwargs)
-
-    def resolve_backend_for_voice(
-        self,
-        spec: str,
-        fallback: str = "kokoro",
-    ) -> str:
-        """Determine which backend owns the given voice specification.
-
-        Resolution rules:
-        1. Empty spec -> fallback
-        2. Kokoro formula (contains '*' or '+') -> "kokoro"
-        3. Exact voice ID match against registered backends -> backend id
-        4. Unknown voice -> fallback
-        """
-        raw = str(spec or "").strip()
-        if not raw:
-            return fallback
-
-        if "*" in raw or "+" in raw:
-            return "kokoro"
-
-        upper = raw.upper()
-        for metadata in self._backends.values():
-            if upper in metadata.voices:
-                return metadata.id
-
-        return fallback
-
-
-_registry = TTSBackendRegistry()
-
-
-def register_backend(
-    metadata: TTSBackendMetadata,
-    factory: Callable[..., TTSBackend],
-) -> None:
-    """Register a TTS backend in the global registry."""
-    _registry.register(metadata, factory)
-
-
-def get_metadata(backend_id: str) -> TTSBackendMetadata:
-    """Get metadata for a specific backend by id.
-
-    Ensures all backends are registered by importing the tts_backends
-    package on first access.
-
-    Raises:
-        KeyError: If backend with given id is not registered.
-    """
-    import abogen.tts_backends  # noqa: F401  — triggers backend registration
-    return _registry.get_metadata(backend_id)
-
-
-def get_default_voice(backend_id: str, fallback: str = "") -> str:
-    """Return the first voice of a backend, or *fallback* if none."""
-    voices = get_metadata(backend_id).voices
-    return voices[0] if voices else fallback
-
-
-def create_backend(backend_id: str, **kwargs: Any) -> TTSBackend:
-    """Create a TTS backend instance by provider id."""
-    return _registry.create_backend(backend_id, **kwargs)
-
-
-def is_registered_backend(backend_id: str) -> bool:
-    """Return True if *backend_id* is a registered TTS backend."""
-    import abogen.tts_backends  # noqa: F401  — triggers backend registration
-    return _registry.is_registered(backend_id)
-
-
-def resolve_backend_for_voice(
-    spec: str,
-    fallback: str = "kokoro",
-) -> str:
-    """Determine which backend owns the given voice specification.
-
-    Ensures all backends are registered by importing the tts_backends
-    package on first access.
-
-    Resolution rules:
-    1. Empty spec -> fallback
-    2. Kokoro formula (contains '*' or '+') -> "kokoro"
-    3. Exact voice ID match against registered backends -> backend id
-    4. Unknown voice -> fallback
-    """
-    import abogen.tts_backends  # noqa: F401  — triggers backend registration
-    return _registry.resolve_backend_for_voice(spec, fallback=fallback)

abogen/tts_backends/__init__.py

@@ -1,20 +0,0 @@
-"""TTS backends package.
-
-Backend modules are auto-discovered and imported here.
-Each backend module registers itself with the global registry
-when imported.
-"""
-
-import importlib
-import pkgutil
-
-
-def _discover_backends():
-    """Import all modules in this package to trigger their registration."""
-    package = __name__
-    for _importer, modname, _ispkg in pkgutil.iter_modules(path=__path__):
-        importlib.import_module(f"{package}.{modname}")
-
-
-_discover_backends()
-

abogen/tts_backends/kokoro.py

@@ -1,179 +0,0 @@
-"""
-Kokoro TTS Backend
-
-Encapsulates the Kokoro KPipeline as a TTSBackend implementation.
-"""
-
-from __future__ import annotations
-
-from typing import Any, Dict, Iterator, List, Optional
-
-import numpy as np
-
-from abogen.tts_backend import TTSBackendMetadata
-
-# Internal voice list — source of truth for Kokoro voices.
-# The rest of the project accesses voices via get_metadata("kokoro").voices.
-_VOICES_INTERNAL = [
-    "af_alloy",
-    "af_aoede",
-    "af_bella",
-    "af_heart",
-    "af_jessica",
-    "af_kore",
-    "af_nicole",
-    "af_nova",
-    "af_river",
-    "af_sarah",
-    "af_sky",
-    "am_adam",
-    "am_echo",
-    "am_eric",
-    "am_fenrir",
-    "am_liam",
-    "am_michael",
-    "am_onyx",
-    "am_puck",
-    "am_santa",
-    "bf_alice",
-    "bf_emma",
-    "bf_isabella",
-    "bf_lily",
-    "bm_daniel",
-    "bm_fable",
-    "bm_george",
-    "bm_lewis",
-    "ef_dora",
-    "em_alex",
-    "em_santa",
-    "ff_siwis",
-    "hf_alpha",
-    "hf_beta",
-    "hm_omega",
-    "hm_psi",
-    "if_sara",
-    "im_nicola",
-    "jf_alpha",
-    "jf_gongitsune",
-    "jf_nezumi",
-    "jf_tebukuro",
-    "jm_kumo",
-    "pf_dora",
-    "pm_alex",
-    "pm_santa",
-    "zf_xiaobei",
-    "zf_xiaoni",
-    "zf_xiaoxiao",
-    "zf_xiaoyi",
-    "zm_yunjian",
-    "zm_yunxi",
-    "zm_yunxia",
-    "zm_yunyang",
-]
-
-_KOKORO_METADATA = TTSBackendMetadata(
-    id="kokoro",
-    name="Kokoro",
-    description="Kokoro TTS engine",
-    voices=tuple(_VOICES_INTERNAL),
-)
-
-
-def _load_kpipeline():
-    """Lazy-load Kokoro dependencies."""
-    from kokoro import KPipeline  # type: ignore[import-not-found]
-
-    return KPipeline
-
-
-class KokoroBackend:
-    """TTSBackend implementation wrapping the Kokoro KPipeline.
-
-    All interaction with KPipeline is encapsulated here.
-    The rest of the project depends only on this class.
-    """
-
-    def __init__(self, **kwargs: Any) -> None:
-        lang_code = kwargs["lang_code"]
-        repo_id = kwargs.get("repo_id", "hexgrad/Kokoro-82M")
-        device = kwargs.get("device", "cpu")
-
-        KPipeline = _load_kpipeline()
-        self._pipeline = KPipeline(
-            lang_code=lang_code,
-            repo_id=repo_id,
-            device=device,
-        )
-        self._lang_code = lang_code
-
-    @property
-    def metadata(self) -> TTSBackendMetadata:
-        return _KOKORO_METADATA
-
-    def __call__(
-        self,
-        text: str,
-        *,
-        voice: Any,
-        speed: float = 1.0,
-        split_pattern: Optional[str] = None,
-    ) -> Iterator[Any]:
-        """Delegate to KPipeline's __call__."""
-        return self._pipeline(
-            text,
-            voice=voice,
-            speed=speed,
-            split_pattern=split_pattern,
-        )
-
-    def load_single_voice(self, voice_name: str) -> Any:
-        """Load a single voice tensor. Used by voice formula system."""
-        return self._pipeline.load_single_voice(voice_name)
-
-    def synthesize(self, text: str, **kwargs: Any) -> bytes:
-        """Synthesize speech from text. Returns raw audio bytes."""
-        voice = kwargs.get("voice", "")
-        speed = kwargs.get("speed", 1.0)
-        split_pattern = kwargs.get("split_pattern", None)
-
-        audio_parts: list[np.ndarray] = []
-        for segment in self(text, voice=voice, speed=speed, split_pattern=split_pattern):
-            audio = segment.audio
-            if hasattr(audio, "numpy"):
-                audio = audio.numpy()
-            audio_parts.append(np.asarray(audio, dtype="float32"))
-
-        if not audio_parts:
-            return b""
-
-        combined = np.concatenate(audio_parts).astype("float32", copy=False)
-        return combined.tobytes()
-
-    def get_available_voices(self) -> List[str]:
-        """Return known Kokoro voice identifiers."""
-        return list(self.metadata.voices)
-
-    def get_supported_formats(self) -> List[str]:
-        """Kokoro outputs raw PCM float32 audio."""
-        return ["pcm_float32"]
-
-    def get_info(self) -> Dict[str, Any]:
-        return {
-            "id": "kokoro",
-            "name": "Kokoro",
-            "lang_code": self._lang_code,
-        }
-
-
-def create_kokoro_backend(**kwargs: Any) -> KokoroBackend:
-    """Factory callable registered with TTSBackendRegistry."""
-    return KokoroBackend(**kwargs)
-
-
-# --- Registration ---
-from abogen.tts_backend_registry import register_backend  # noqa: E402
-
-register_backend(
-    metadata=_KOKORO_METADATA,
-    factory=create_kokoro_backend,
-)

abogen/tts_supertonic.py

@@ -5,7 +5,7 @@ from dataclasses import dataclass
 import logging
 import math
 import re
-from typing import Any, Dict, Iterable, Iterator, List, Optional
+from typing import Any, Iterable, Iterator, Optional
 
 import numpy as np
 
@@ -15,15 +15,6 @@ logger = logging.getLogger(__name__)
 
 DEFAULT_SUPERTONIC_VOICES = ("M1", "M2", "M3", "M4", "M5", "F1", "F2", "F3", "F4", "F5")
 
-from abogen.tts_backend import TTSBackendMetadata
-
-_SUPERTONIC_METADATA = TTSBackendMetadata(
-    id="supertonic",
-    name="SuperTonic",
-    description="SuperTonic TTS engine",
-    voices=DEFAULT_SUPERTONIC_VOICES,
-)
-
 
 @dataclass
 class SupertonicSegment:
@@ -282,111 +273,3 @@ class SupertonicPipeline:
                 audio = _resample_linear(audio, src_rate, self.sample_rate)
 
             yield SupertonicSegment(graphemes=chunk_to_speak, audio=audio)
-
-
-class SupertonicBackend:
-    """Supertonic TTS backend implementing the TTSBackend protocol.
-
-    Encapsulates ``SupertonicPipeline`` as an internal implementation detail.
-    """
-
-    @property
-    def metadata(self) -> TTSBackendMetadata:
-        return _SUPERTONIC_METADATA
-
-    def __init__(self, **kwargs: Any) -> None:
-        self._pipeline = SupertonicPipeline(
-            sample_rate=kwargs.get("sample_rate", 24000),
-            auto_download=kwargs.get("auto_download", True),
-            total_steps=kwargs.get("total_steps", 5),
-        )
-
-    def synthesize(self, text: str, **kwargs: Any) -> bytes:
-        """Synthesize speech and return raw audio bytes (WAV).
-
-        Delegates to the internal :class:`SupertonicPipeline` and concatenates
-        all produced segments into a single byte buffer.
-        """
-        import io
-
-        import soundfile as sf
-
-        voice = kwargs.get("voice", "M1")
-        speed = float(kwargs.get("speed", 1.0))
-        split_pattern = kwargs.get("split_pattern")
-        total_steps = kwargs.get("total_steps")
-
-        segments = self._pipeline(
-            text,
-            voice=voice,
-            speed=speed,
-            split_pattern=split_pattern,
-            total_steps=total_steps,
-        )
-
-        audio_parts: list[np.ndarray] = []
-        for seg in segments:
-            audio_parts.append(seg.audio)
-
-        if not audio_parts:
-            return b""
-
-        combined = np.concatenate(audio_parts)
-        buf = io.BytesIO()
-        sf.write(buf, combined, self._pipeline.sample_rate, format="WAV")
-        return buf.getvalue()
-
-    def get_available_voices(self) -> List[str]:
-        """Return the list of built-in SuperTonic voice identifiers."""
-        return list(self.metadata.voices)
-
-    def get_supported_formats(self) -> List[str]:
-        return ["wav"]
-
-    def get_info(self) -> Dict[str, Any]:
-        return {
-            "sample_rate": self._pipeline.sample_rate,
-            "total_steps": self._pipeline.total_steps,
-            "max_chunk_length": self._pipeline.max_chunk_length,
-            "voices": list(DEFAULT_SUPERTONIC_VOICES),
-        }
-
-    def __call__(
-        self,
-        text: str,
-        *,
-        voice: str,
-        speed: float,
-        split_pattern: Optional[str] = None,
-        total_steps: Optional[int] = None,
-    ) -> Iterator[SupertonicSegment]:
-        """Backward-compatible call interface, delegates to the pipeline."""
-        return self._pipeline(
-            text,
-            voice=voice,
-            speed=speed,
-            split_pattern=split_pattern,
-            total_steps=total_steps,
-        )
-
-
-def create_supertonic_backend(**kwargs: Any) -> SupertonicBackend:
-    """Create a SuperTonic TTS backend instance.
-
-    Args:
-        sample_rate: Audio sample rate. Defaults to 24000.
-        auto_download: Auto-download models. Defaults to True.
-        total_steps: Inference steps. Defaults to 5.
-
-    Returns:
-        SupertonicBackend instance.
-    """
-    return SupertonicBackend(**kwargs)
-
-
-from abogen.tts_backend_registry import register_backend  # noqa: E402
-
-register_backend(
-    metadata=_SUPERTONIC_METADATA,
-    factory=create_supertonic_backend,
-)

abogen/utils.py

@@ -529,20 +529,21 @@ def prevent_sleep_end():
                 _sleep_procs[system] = None
 
 
+def load_numpy_kpipeline():
+    import numpy as np
+    from kokoro import KPipeline  # type: ignore[import-not-found]
+
+    return np, KPipeline
+
+
 class LoadPipelineThread(Thread):
-    def __init__(self, callback, lang_code="a", device="cpu"):
+    def __init__(self, callback):
         super().__init__()
         self.callback = callback
-        self.lang_code = lang_code
-        self.device = device
 
     def run(self):
         try:
-            from abogen.tts_backend_registry import create_backend
-
-            backend = create_backend(
-                "kokoro", lang_code=self.lang_code, device=self.device
-            )
-            self.callback(backend, None)
+            np_module, kpipeline_class = load_numpy_kpipeline()
+            self.callback(np_module, kpipeline_class, None)
         except Exception as e:
-            self.callback(None, str(e))
+            self.callback(None, None, str(e))

abogen/voice_cache.py

@@ -17,7 +17,7 @@ if LocalEntryNotFoundError is None:  # pragma: no cover - fallback for tests
         pass
 
 
-from abogen.tts_backend_registry import get_metadata
+from abogen.constants import VOICES_INTERNAL
 
 _CACHE_LOCK = threading.Lock()
 _CACHED_VOICES: Set[str] = set()
@@ -26,9 +26,8 @@ _BOOTSTRAPPED = False
 
 
 def _normalize_targets(voices: Optional[Iterable[str]]) -> Set[str]:
-    kokoro_voices = get_metadata("kokoro").voices
     if not voices:
-        return set(kokoro_voices)
+        return set(VOICES_INTERNAL)
     normalized: Set[str] = set()
     for voice in voices:
         if not voice:
@@ -36,7 +35,7 @@ def _normalize_targets(voices: Optional[Iterable[str]]) -> Set[str]:
         voice_id = str(voice).strip()
         if not voice_id:
             continue
-        if voice_id in kokoro_voices:
+        if voice_id in VOICES_INTERNAL:
             normalized.add(voice_id)
     return normalized
 

abogen/voice_formulas.py

@@ -1,7 +1,7 @@
 import re
 from typing import List, Tuple
 
-from abogen.tts_backend_registry import get_metadata
+from abogen.constants import VOICES_INTERNAL
 
 
 # Calls parsing and loads the voice to gpu or cpu
@@ -22,7 +22,6 @@ def parse_formula_terms(formula: str) -> List[Tuple[str, float]]:
         raise ValueError("Empty voice formula")
 
     terms: List[Tuple[str, float]] = []
-    kokoro_voices = get_metadata("kokoro").voices
     for segment in formula.split("+"):
         part = segment.strip()
         if not part:
@@ -31,7 +30,7 @@ def parse_formula_terms(formula: str) -> List[Tuple[str, float]]:
             raise ValueError("Each component must be in the form voice*weight")
         voice_name, raw_weight = part.split("*", 1)
         voice_name = voice_name.strip()
-        if voice_name not in kokoro_voices:
+        if voice_name not in VOICES_INTERNAL:
             raise ValueError(f"Unknown voice: {voice_name}")
         try:
             weight = float(raw_weight.strip())

abogen/voice_metadata.py

@@ -1,33 +0,0 @@
-from dataclasses import dataclass
-
-
-@dataclass(frozen=True)
-class VoiceMetadata:
-    """
-    Immutable metadata describing a voice from a TTS backend.
-
-    This model describes a voice independently of any backend implementation.
-    Backends populate these objects; the application consumes them.
-
-    The ``backend_id`` field is set by the backend itself (via
-    ``self.metadata.id``) — the application never hardcodes it.
-    This ensures renaming a backend does not require touching voice definitions.
-    """
-
-    id: str
-    """Unique voice identifier within the backend (e.g. ``"af_alloy"``, ``"M1"``)."""
-
-    display_name: str
-    """Human-readable display name (e.g. ``"Alloy"``, ``"Male 1"``)."""
-
-    language: str
-    """Language code — backend-specific format is acceptable (e.g. ``"a"``, ``"en"``)."""
-
-    gender: str
-    """Gender category: ``"female"``, ``"male"``, or ``"unknown"``."""
-
-    backend_id: str
-    """Identifier of the backend that owns this voice (e.g. ``"kokoro"``).
-
-    Set automatically by the backend — never hardcoded in voice definitions.
-    """

abogen/voice_profiles.py

@@ -2,7 +2,8 @@ import json
 import os
 from typing import Any, Dict, Iterable, List, Tuple
 
-from abogen.tts_backend_registry import get_metadata, is_registered_backend
+from abogen.constants import VOICES_INTERNAL
+from abogen.tts_supertonic import DEFAULT_SUPERTONIC_VOICES
 from abogen.utils import get_user_config_path
 
 
@@ -69,8 +70,7 @@ def serialize_profiles() -> Dict[str, Dict[str, Iterable[Tuple[str, float]]]]:
 
 def _normalize_supertonic_voice(value: Any) -> str:
     raw = str(value or "").strip().upper()
-    supertonic_voices = get_metadata("supertonic").voices
-    return raw if raw in supertonic_voices else "M1"
+    return raw if raw in DEFAULT_SUPERTONIC_VOICES else "M1"
 
 
 def _coerce_supertonic_steps(value: Any) -> int:
@@ -101,7 +101,7 @@ def normalize_profile_entry(entry: Any) -> Dict[str, Any]:
         return {}
 
     provider = str(entry.get("provider") or "kokoro").strip().lower()
-    if not is_registered_backend(provider):
+    if provider not in {"kokoro", "supertonic"}:
         provider = "kokoro"
 
     language = str(entry.get("language") or "a").strip().lower() or "a"
@@ -135,7 +135,6 @@ def normalize_profile_entry(entry: Any) -> Dict[str, Any]:
 
 def _normalize_voice_entries(entries: Iterable) -> List[Tuple[str, float]]:
     normalized: List[Tuple[str, float]] = []
-    kokoro_voices = get_metadata("kokoro").voices
     for item in entries or []:
         if isinstance(item, dict):
             voice = item.get("id") or item.get("voice")
@@ -144,7 +143,7 @@ def _normalize_voice_entries(entries: Iterable) -> List[Tuple[str, float]]:
             voice, weight = item[0], item[1]
         else:
             continue
-        if voice not in kokoro_voices:
+        if voice not in VOICES_INTERNAL:
             continue
         if weight is None:
             continue

abogen/webui/Dockerfile

@@ -2,6 +2,7 @@ FROM nvidia/cuda:12.6.3-cudnn-runtime-ubuntu22.04
 
 ENV PYTHONDONTWRITEBYTECODE=1 \
   PYTHONUNBUFFERED=1 \
+  PIP_NO_CACHE_DIR=1 \
   VIRTUAL_ENV=/opt/venv \
   PATH=/opt/venv/bin:$PATH
 
@@ -26,22 +27,22 @@ RUN python3 -m venv "$VIRTUAL_ENV"
 WORKDIR /app
 
 COPY pyproject.toml README.md ./
-RUN pip install uv \
-  && if [ -n "$TORCH_VERSION" ]; then \
-       uv pip install --system torch=="$TORCH_VERSION" torchvision=="$TORCH_VERSION" torchaudio=="$TORCH_VERSION" --index-url "$TORCH_INDEX_URL"; \
-     else \
-       uv pip install --system torch torchvision torchaudio --index-url "$TORCH_INDEX_URL"; \
-     fi \
-  && uv pip install --system . \
-    https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl \
-  && uv pip install --system "mutagen>=1.47.0"
-
 COPY abogen ./abogen
 
+RUN pip install --upgrade pip \
+  && if [ -n "$TORCH_VERSION" ]; then \
+       pip install torch=="$TORCH_VERSION" torchvision=="$TORCH_VERSION" torchaudio=="$TORCH_VERSION" --index-url "$TORCH_INDEX_URL"; \
+     else \
+       pip install torch torchvision torchaudio --index-url "$TORCH_INDEX_URL"; \
+     fi \
+  && pip install --no-cache-dir . \
+    https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl \
+  && pip install --no-cache-dir "mutagen>=1.47.0"
+
 # Install onnxruntime-gpu for CUDA acceleration (supertonic uses ONNX Runtime)
 # Set USE_GPU=false to skip this for CPU-only deployments
 RUN if [ "$USE_GPU" = "true" ]; then \
-      uv pip install --system onnxruntime-gpu; \
+      pip install --no-cache-dir onnxruntime-gpu; \
     fi
 
 ENV ABOGEN_HOST=0.0.0.0 \

abogen/webui/conversion_runner.py

@@ -20,7 +20,7 @@ import numpy as np
 import soundfile as sf
 import static_ffmpeg
 
-from abogen.tts_backend_registry import get_metadata, is_registered_backend, resolve_backend_for_voice
+from abogen.constants import VOICES_INTERNAL
 from abogen.epub3.exporter import build_epub3_package
 from abogen.kokoro_text_normalization import ApostropheConfig, normalize_for_pipeline, HAS_NUM2WORDS
 from abogen.normalization_settings import (
@@ -39,15 +39,14 @@ from abogen.utils import (
     get_user_cache_path,
     get_user_output_path,
     load_config,
+    load_numpy_kpipeline,
 )
-from abogen.tts_backend_registry import create_backend
-from abogen.tts_backend import TTSBackend
 from abogen.voice_cache import ensure_voice_assets
 from abogen.voice_formulas import extract_voice_ids, get_new_voice
 from abogen.voice_profiles import load_profiles, normalize_profile_entry
 from abogen.pronunciation_store import increment_usage
 from abogen.llm_client import LLMClientError
-
+from abogen.tts_supertonic import DEFAULT_SUPERTONIC_VOICES, SupertonicPipeline
 
 from .service import Job, JobStatus
 
@@ -57,27 +56,26 @@ SAMPLE_RATE = 24000
 
 
 def _supertonic_voice_from_spec(spec: Any, fallback: str) -> str:
-    """Normalize a voice specification for Supertonic.
-
-    This function only performs Supertonic-specific normalization (uppercase conversion
-    and fallback handling). Backend resolution is handled by the registry.
-    """
     raw = str(spec or "").strip()
     fallback_raw = str(fallback or "").strip()
 
-    # Normalize to uppercase for Supertonic voice IDs
-    upper = raw.upper() if raw else ""
-
-    # If empty or contains formula characters, use fallback
-    if not upper or "*" in upper or "+" in upper:
-        upper = fallback_raw.upper() if fallback_raw else ""
-
-    # If still empty, use default Supertonic voice
-    if not upper or "*" in upper or "+" in upper:
-        upper = "M1"
+    # SuperTonic voices are discrete IDs (M1/F3/...). If we see a Kokoro mix
+    # formula (contains '*' or '+'), ignore it and fall back to a safe voice.
+    if not raw or "*" in raw or "+" in raw:
+        raw = fallback_raw
+    if not raw or "*" in raw or "+" in raw:
+        raw = "M1"
 
+    upper = raw.upper()
+    if upper in DEFAULT_SUPERTONIC_VOICES:
         return upper
 
+    fallback_upper = fallback_raw.upper() if fallback_raw else ""
+    if fallback_upper in DEFAULT_SUPERTONIC_VOICES:
+        return fallback_upper
+
+    return "M1"
+
 
 def _split_speaker_reference(value: Any) -> tuple[Optional[str], str]:
     raw = str(value or "").strip()
@@ -120,7 +118,15 @@ def _formula_from_kokoro_entry(entry: Mapping[str, Any]) -> str:
 
 
 def _infer_provider_from_spec(value: Any, fallback: str = "kokoro") -> str:
-    return resolve_backend_for_voice(str(value or ""), fallback=fallback)
+    raw = str(value or "").strip()
+    if not raw:
+        return fallback
+    upper = raw.upper()
+    if upper in DEFAULT_SUPERTONIC_VOICES:
+        return "supertonic"
+    if "*" in raw or "+" in raw:
+        return "kokoro"
+    return fallback
 
 
 class _JobCancelled(Exception):
@@ -569,7 +575,7 @@ def _spec_to_voice_ids(spec: Any) -> Set[str]:
             return set(extract_voice_ids(text))
         except ValueError:
             return set()
-    if text in get_metadata("kokoro").voices:
+    if text in VOICES_INTERNAL:
         return {text}
     return set()
 
@@ -633,7 +639,7 @@ def _collect_required_voice_ids(job: Job) -> Set[str]:
             for key in ("resolved_voice", "voice_formula", "voice"):
                 voices.update(_spec_to_voice_ids(payload.get(key)))
 
-    voices.update(get_metadata("kokoro").voices)
+    voices.update(VOICES_INTERNAL)
     return voices
 
 
@@ -1567,7 +1573,7 @@ def run_conversion_job(job: Job) -> None:
         def get_pipeline(provider: str) -> Any:
             nonlocal kokoro_cache_ready
             provider_norm = str(provider or "kokoro").strip().lower() or "kokoro"
-            if not is_registered_backend(provider_norm):
+            if provider_norm not in {"kokoro", "supertonic"}:
                 provider_norm = "kokoro"
 
             existing = pipelines.get(provider_norm)
@@ -1575,8 +1581,7 @@ def run_conversion_job(job: Job) -> None:
                 return existing
 
             if provider_norm == "supertonic":
-                pipelines[provider_norm] = create_backend(
-                    "supertonic",
+                pipelines[provider_norm] = SupertonicPipeline(
                     sample_rate=SAMPLE_RATE,
                     auto_download=True,
                     total_steps=int(getattr(job, "supertonic_total_steps", 5) or 5),
@@ -1589,12 +1594,16 @@ def run_conversion_job(job: Job) -> None:
             device = "cpu"
             if not disable_gpu:
                 device = _select_device()
-            # Create KPipeline instance directly (conforms to TTSBackend protocol)
-            pipelines[provider_norm] = create_backend(
-                "kokoro",
-                lang_code=job.language,
-                device=device
-            )
+            _np, KPipeline = load_numpy_kpipeline()
+            # Try to initialize with the selected device; fall back to CPU if CUDA fails
+            try:
+                pipelines[provider_norm] = KPipeline(lang_code=job.language, repo_id="hexgrad/Kokoro-82M", device=device)
+            except RuntimeError as e:
+                if "CUDA" in str(e) and device != "cpu":
+                    job.add_log(f"CUDA initialization failed, falling back to CPU: {e}", level="warning")
+                    pipelines[provider_norm] = KPipeline(lang_code=job.language, repo_id="hexgrad/Kokoro-82M", device="cpu")
+                else:
+                    raise
             if not kokoro_cache_ready:
                 _initialize_voice_cache(job)
                 kokoro_cache_ready = True
@@ -1635,8 +1644,8 @@ def run_conversion_job(job: Job) -> None:
                 return provider, resolved, cached, speed, steps
 
             if provider == "kokoro":
-                kokoro_backend = get_pipeline("kokoro")
-                choice = _resolve_voice(kokoro_backend, resolved, job.use_gpu)
+                kokoro_pipeline = get_pipeline("kokoro")
+                choice = _resolve_voice(kokoro_pipeline, resolved, job.use_gpu)
             else:
                 choice = resolved
 
@@ -1765,8 +1774,8 @@ def run_conversion_job(job: Job) -> None:
         voice_cache: Dict[str, Any] = {}
         base_provider, base_voice_resolved, _, _ = resolve_voice_target(base_voice_spec)
         if base_provider == "kokoro" and base_voice_resolved and "*" not in base_voice_resolved:
-            kokoro_backend = get_pipeline("kokoro")
-            voice_cache[f"kokoro:{base_voice_resolved}"] = _resolve_voice(kokoro_backend, base_voice_resolved, job.use_gpu)
+            kokoro_pipeline = get_pipeline("kokoro")
+            voice_cache[f"kokoro:{base_voice_resolved}"] = _resolve_voice(kokoro_pipeline, base_voice_resolved, job.use_gpu)
         processed_chars = 0
         subtitle_index = 1
         current_time = 0.0
@@ -1796,8 +1805,8 @@ def run_conversion_job(job: Job) -> None:
                     fallback_key = next(iter(voice_cache.keys()), "")
                     if fallback_key and fallback_key != "__custom_mix":
                         intro_voice_spec = fallback_key.split(":", 1)[-1]
-                if not intro_voice_spec:
-                    intro_voice_spec = get_default_voice("kokoro")
+                if not intro_voice_spec and VOICES_INTERNAL:
+                    intro_voice_spec = VOICES_INTERNAL[0]
 
                 if intro_voice_spec:
                     intro_provider, _, intro_voice_choice, intro_speed, intro_steps = resolve_voice_choice(
@@ -1851,15 +1860,14 @@ def run_conversion_job(job: Job) -> None:
                     total_steps=int(supertonic_steps_override if supertonic_steps_override is not None else getattr(job, "supertonic_total_steps", 5)),
                 )
             else:
-                kokoro_backend = get_pipeline("kokoro")
-                segment_iter = kokoro_backend(
+                kokoro_pipeline = get_pipeline("kokoro")
+                segment_iter = kokoro_pipeline(
                     normalized,
                     voice=voice_choice,
                     speed=float(speed_override if speed_override is not None else job.speed),
                     split_pattern=split_pattern,
                 )
 
-            try:
             for segment in segment_iter:
                 canceller()
                 graphemes_raw = getattr(segment, "graphemes", "") or ""
@@ -1899,11 +1907,6 @@ def run_conversion_job(job: Job) -> None:
                 if audio_sink:
                     current_time += duration
 
-            except OverflowError as exc:
-                job.add_log(
-                    f"Skipped chunk — number too large for TTS conversion: {exc}",
-                    level="warning",
-                )
             return local_segments
 
         def append_silence(
@@ -1947,8 +1950,8 @@ def run_conversion_job(job: Job) -> None:
             if chapter_provider == "kokoro":
                 voice_choice = voice_cache.get(chapter_cache_key)
                 if voice_choice is None:
-                    kokoro_backend = get_pipeline("kokoro")
-                    voice_choice = _resolve_voice(kokoro_backend, chapter_voice_resolved, job.use_gpu)
+                    kokoro_pipeline = get_pipeline("kokoro")
+                    voice_choice = _resolve_voice(kokoro_pipeline, chapter_voice_resolved, job.use_gpu)
                     voice_cache[chapter_cache_key] = voice_choice
             else:
                 voice_choice = chapter_voice_resolved
@@ -2092,9 +2095,9 @@ def run_conversion_job(job: Job) -> None:
                         if chunk_provider == "kokoro":
                             chunk_voice_choice = voice_cache.get(chunk_cache_key)
                             if chunk_voice_choice is None:
-                                kokoro_backend = get_pipeline("kokoro")
+                                kokoro_pipeline = get_pipeline("kokoro")
                                 chunk_voice_choice = _resolve_voice(
-                                    kokoro_backend,
+                                    kokoro_pipeline,
                                     chunk_voice_resolved,
                                     job.use_gpu,
                                 )
@@ -2236,8 +2239,8 @@ def run_conversion_job(job: Job) -> None:
                 if fallback_key and fallback_key != "__custom_mix":
                     # `voice_cache` keys are internal and include provider prefixes.
                     outro_voice_spec = fallback_key.split(":", 1)[-1]
-            if not outro_voice_spec:
-                outro_voice_spec = get_default_voice("kokoro")
+            if not outro_voice_spec and VOICES_INTERNAL:
+                outro_voice_spec = VOICES_INTERNAL[0]
 
             if outro_text and outro_voice_spec:
                 outro_start_time = current_time
@@ -2442,8 +2445,7 @@ def _load_pipeline(job: Job):
     disable_gpu = not job.use_gpu or not cfg.get("use_gpu", True)
     provider = str(getattr(job, "tts_provider", "kokoro") or "kokoro").strip().lower()
     if provider == "supertonic":
-        return create_backend(
-            "supertonic",
+        return SupertonicPipeline(
             sample_rate=SAMPLE_RATE,
             auto_download=True,
             total_steps=int(getattr(job, "supertonic_total_steps", 5) or 5),
@@ -2452,7 +2454,8 @@ def _load_pipeline(job: Job):
     device = "cpu"
     if not disable_gpu:
         device = _select_device()
-    return create_backend("kokoro", lang_code=job.language, device=device)
+    _np, KPipeline = load_numpy_kpipeline()
+    return KPipeline(lang_code=job.language, repo_id="hexgrad/Kokoro-82M", device=device)
 
 
 def _select_device() -> str:

abogen/webui/debug_tts_runner.py

@@ -15,7 +15,7 @@ from abogen.normalization_settings import build_apostrophe_config
 from abogen.text_extractor import extract_from_path
 from abogen.voice_cache import ensure_voice_assets
 from abogen.webui.conversion_runner import SAMPLE_RATE, SPLIT_PATTERN, _select_device, _to_float32, _resolve_voice, _spec_to_voice_ids
-from abogen.tts_backend_registry import create_backend
+from abogen.utils import load_numpy_kpipeline
 
 
 _MARKER_RE = re.compile(re.escape(MARKER_PREFIX) + r"(?P<code>[A-Z0-9_]+)" + re.escape(MARKER_SUFFIX))
@@ -45,7 +45,8 @@ def _load_pipeline(language: str, use_gpu: bool) -> Any:
     device = "cpu"
     if use_gpu:
         device = _select_device()
-    return create_backend("kokoro", lang_code=language, device=device)
+    _np, KPipeline = load_numpy_kpipeline()
+    return KPipeline(lang_code=language, repo_id="hexgrad/Kokoro-82M", device=device)
 
 
 def _extract_cases_from_text(text: str) -> List[Tuple[str, str]]:

abogen/webui/routes/api.py

@@ -34,7 +34,6 @@ from abogen.normalization_settings import (
 )
 from abogen.llm_client import list_models, LLMClientError
 from abogen.kokoro_text_normalization import normalize_for_pipeline
-from abogen.tts_backend_registry import is_registered_backend
 from abogen.integrations.audiobookshelf import AudiobookshelfClient, AudiobookshelfConfig
 from abogen.integrations.calibre_opds import (
     CalibreOPDSClient,
@@ -64,7 +63,7 @@ def api_save_voice_profile() -> ResponseReturnValue:
     if profile is None:
         # Speaker Studio payload format
         provider = str(payload.get("provider") or "kokoro").strip().lower()
-        if not is_registered_backend(provider):
+        if provider not in {"kokoro", "supertonic"}:
             provider = "kokoro"
         if provider == "supertonic":
             profile = {
@@ -231,7 +230,7 @@ def api_speaker_preview() -> ResponseReturnValue:
     use_gpu = settings.get("use_gpu", False)
 
     base_spec, speaker_name = split_profile_spec(voice)
-    resolved_provider = tts_provider if is_registered_backend(tts_provider) else ""
+    resolved_provider = tts_provider if tts_provider in {"kokoro", "supertonic"} else ""
 
     if speaker_name:
         entry = normalize_profile_entry(load_profiles().get(speaker_name))

abogen/webui/routes/utils/form.py

@@ -7,7 +7,6 @@ from flask.typing import ResponseReturnValue
 
 from abogen.webui.service import PendingJob, JobStatus
 from abogen.webui.routes.utils.service import get_service
-from abogen.tts_backend_registry import is_registered_backend
 from abogen.webui.routes.utils.settings import (
     load_settings,
     coerce_bool,
@@ -33,7 +32,7 @@ from abogen.webui.routes.utils.common import split_profile_spec
 from abogen.utils import calculate_text_length
 from abogen.voice_profiles import serialize_profiles, normalize_profile_entry
 from abogen.chunking import ChunkLevel, build_chunks_for_chapters
-from abogen.tts_backend_registry import get_default_voice
+from abogen.constants import VOICES_INTERNAL
 from abogen.speaker_configs import get_config
 from abogen.kokoro_text_normalization import normalize_roman_numeral_titles
 from dataclasses import dataclass
@@ -580,7 +579,7 @@ def apply_book_step_form(
     # spec (e.g. "speaker:Name" for saved speakers, or a Kokoro mix formula).
     # This enables mixed-provider conversions (e.g. narrator=SuperTonic, characters=Kokoro).
     provider_value = str(form.get("tts_provider") or "").strip().lower()
-    if is_registered_backend(provider_value):
+    if provider_value in {"kokoro", "supertonic"}:
         pending.tts_provider = provider_value
 
     # Determine the base speaker selection (saved speaker ref or raw voice).
@@ -617,8 +616,8 @@ def apply_book_step_form(
         custom_formula = ""
 
     base_voice_spec = resolved_default_voice or narrator_voice_raw
-    if not base_voice_spec:
-        base_voice_spec = get_default_voice("kokoro")
+    if not base_voice_spec and VOICES_INTERNAL:
+        base_voice_spec = VOICES_INTERNAL[0]
 
     voice_choice, resolved_language, selected_profile = resolve_voice_choice(
         pending.language,
@@ -797,8 +796,8 @@ def build_pending_job_from_extraction(
         profile_selection = inferred_profile
 
     base_voice = base_voice_input or resolved_default_voice or str(default_voice_setting).strip()
-    if not base_voice:
-        base_voice = get_default_voice("kokoro")
+    if not base_voice and VOICES_INTERNAL:
+        base_voice = VOICES_INTERNAL[0]
     selected_speaker_config = (form.get("speaker_config") or "").strip()
     speaker_config_payload = get_config(selected_speaker_config) if selected_speaker_config else None
 

abogen/webui/routes/utils/preview.py

@@ -78,9 +78,10 @@ def get_preview_pipeline(language: str, device: str) -> Any:
         pipeline = _preview_pipelines.get(key)
         if pipeline is not None:
             return pipeline
-        from abogen.tts_backend_registry import create_backend
+        from abogen.utils import load_numpy_kpipeline
 
-        pipeline = create_backend("kokoro", lang_code=language, device=device)
+        _, KPipeline = load_numpy_kpipeline()
+        pipeline = KPipeline(lang_code=language, repo_id="hexgrad/Kokoro-82M", device=device)
         _preview_pipelines[key] = pipeline
         return pipeline
 
@@ -136,9 +137,9 @@ def generate_preview_audio(
             normalized_text = source_text
 
     if provider == "supertonic":
-        from abogen.tts_backend_registry import create_backend
+        from abogen.tts_supertonic import SupertonicPipeline
 
-        pipeline = create_backend("supertonic", sample_rate=SAMPLE_RATE, auto_download=True, total_steps=supertonic_total_steps)
+        pipeline = SupertonicPipeline(sample_rate=SAMPLE_RATE, auto_download=True, total_steps=supertonic_total_steps)
         segments = pipeline(
             normalized_text,
             voice=voice_spec,

abogen/webui/routes/utils/settings.py

@@ -6,8 +6,8 @@ from abogen.constants import (
     LANGUAGE_DESCRIPTIONS,
     SUBTITLE_FORMATS,
     SUPPORTED_SOUND_FORMATS,
+    VOICES_INTERNAL,
 )
-from abogen.tts_backend_registry import get_default_voice
 from abogen.normalization_settings import (
     DEFAULT_LLM_PROMPT,
     environment_llm_defaults,
@@ -174,7 +174,7 @@ def settings_defaults() -> Dict[str, Any]:
         "subtitle_format": "srt",
         "save_mode": "default_output" if has_output_override() else "save_next_to_input",
         "default_speaker": "",
-        "default_voice": get_default_voice("kokoro"),
+        "default_voice": VOICES_INTERNAL[0] if VOICES_INTERNAL else "",
         "supertonic_total_steps": 5,
         "supertonic_speed": 1.0,
         "replace_single_newlines": False,

abogen/webui/routes/utils/voice.py

@@ -17,10 +17,10 @@ from abogen.constants import (
     SUPPORTED_SOUND_FORMATS,
     SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
     SAMPLE_VOICE_TEXTS,
+    VOICES_INTERNAL,
 )
-from abogen.tts_backend_registry import get_metadata
 from abogen.speaker_configs import list_configs
-from abogen.tts_backend_registry import create_backend
+from abogen.utils import load_numpy_kpipeline
 from abogen.webui.conversion_runner import _select_device, _to_float32, SAMPLE_RATE, SPLIT_PATTERN
 
 _preview_pipeline_lock = threading.RLock()
@@ -285,7 +285,7 @@ def filter_voice_catalog(
 def build_voice_catalog() -> List[Dict[str, str]]:
     catalog: List[Dict[str, str]] = []
     gender_map = {"f": "Female", "m": "Male"}
-    for voice_id in get_metadata("kokoro").voices:
+    for voice_id in VOICES_INTERNAL:
         prefix, _, rest = voice_id.partition("_")
         language_code = prefix[0] if prefix else "a"
         gender_code = prefix[1] if len(prefix) > 1 else ""
@@ -590,7 +590,7 @@ def template_options() -> Dict[str, Any]:
     voice_catalog = build_voice_catalog()
     return {
         "languages": LANGUAGE_DESCRIPTIONS,
-        "voices": get_metadata("kokoro").voices,
+        "voices": VOICES_INTERNAL,
         "subtitle_formats": SUBTITLE_FORMATS,
         "supported_langs_for_subs": SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
         "output_formats": SUPPORTED_SOUND_FORMATS,
@@ -741,7 +741,8 @@ def get_preview_pipeline(language: str, device: str):
         pipeline = _preview_pipelines.get(key)
         if pipeline is not None:
             return pipeline
-        pipeline = create_backend("kokoro", lang_code=language, device=device)
+        _, KPipeline = load_numpy_kpipeline()
+        pipeline = KPipeline(lang_code=language, repo_id="hexgrad/Kokoro-82M", device=device)
         _preview_pipelines[key] = pipeline
         return pipeline
 

abogen/webui/routes/voices.py

@@ -17,7 +17,7 @@ from abogen.speaker_configs import (
     save_configs,
     delete_config,
 )
-
+from abogen.constants import VOICES_INTERNAL
 
 voices_bp = Blueprint("voices", __name__)
 

pyproject.toml

@@ -50,6 +50,8 @@ dependencies = [
   "num2words>=0.5.13",
   "httpx>=0.27.0",
   "PyQt6>=6.5.0",
+  "flet>=0.85.1",
+  "msgpack>=1.0.0",
 ]
 
 classifiers = [
@@ -77,11 +79,12 @@ allow-direct-references = true
 
 
 [project.gui-scripts]
-abogen = "abogen.pyqt.main:main"
+abogen = "abogen.frontend.main:main"
 
 [project.scripts]
+abogen-ui = "abogen.frontend.main:main"
+abogen-web = "abogen.frontend.main:main_web"
 abogen-cli = "abogen.webui.app:main"
-abogen-web = "abogen.webui.app:main"
 abogen-pyqt = "abogen.pyqt.main:main"
 
 [tool.hatch.build.targets.sdist]

tests/test_conversion_voice_resolution.py

@@ -1,7 +1,7 @@
 from types import SimpleNamespace
 from typing import cast
 
-from abogen.tts_backend_registry import get_metadata
+from abogen.constants import VOICES_INTERNAL
 from abogen.webui.conversion_runner import (
     _chapter_voice_spec,
     _chunk_voice_spec,
@@ -49,4 +49,4 @@ def test_voice_collection_includes_formula_components():
     voices = _collect_required_voice_ids(job)
 
     assert {"af_nova", "am_liam"}.issubset(voices)
-    assert voices.issuperset(get_metadata("kokoro").voices)
+    assert voices.issuperset(VOICES_INTERNAL)

tests/test_epub_exporter.py

@@ -197,7 +197,7 @@ def test_epub3_preserves_original_whitespace(tmp_path) -> None:
     )
     assert match is not None
     original_text = html.unescape(match.group(1))
-    assert "Second line\n\nThird paragraph." in original_text.replace("\r\n", "\n")
+    assert "Second line\n\nThird paragraph." in original_text
 
 
 def test_epub3_sentence_chunks_render_as_paragraphs(tmp_path) -> None:

tests/test_kokoro_backend.py

@@ -1,216 +0,0 @@
-"""Tests for KokoroBackend class."""
-
-from __future__ import annotations
-
-from dataclasses import dataclass
-from typing import Any, Iterator, List
-from unittest.mock import MagicMock, patch
-
-import numpy as np
-import pytest
-
-from abogen.tts_backend import TTSBackendMetadata
-
-
-# ---------------------------------------------------------------------------
-# Helpers
-# ---------------------------------------------------------------------------
-
-@dataclass
-class _FakeSegment:
-    graphemes: str
-    audio: Any  # np.ndarray or torch-like tensor
-
-
-class _FakePipeline:
-    """Minimal mock for kokoro.KPipeline."""
-
-    def __init__(self, *, lang_code: str, repo_id: str, device: str):
-        self.lang_code = lang_code
-        self.repo_id = repo_id
-        self.device = device
-        self._voices: dict[str, np.ndarray] = {}
-
-    def __call__(
-        self,
-        text: str,
-        *,
-        voice: Any = "",
-        speed: float = 1.0,
-        split_pattern: str | None = None,
-    ) -> Iterator[_FakeSegment]:
-        yield _FakeSegment(graphemes=text, audio=np.zeros(100, dtype="float32"))
-
-    def load_single_voice(self, name: str) -> np.ndarray:
-        if name not in self._voices:
-            self._voices[name] = np.ones((1, 256), dtype="float32") * 0.5
-        return self._voices[name]
-
-
-def _make_backend(**kwargs: Any):
-    """Create KokoroBackend with mocked KPipeline."""
-    with patch("abogen.tts_backends.kokoro._load_kpipeline") as load:
-        load.return_value = _FakePipeline
-        from abogen.tts_backends.kokoro import KokoroBackend
-
-        return KokoroBackend(**kwargs)
-
-
-# ---------------------------------------------------------------------------
-# Tests
-# ---------------------------------------------------------------------------
-
-class TestKokoroBackendMetadata:
-    def test_metadata_returns_tts_backend_metadata(self):
-        backend = _make_backend(lang_code="a")
-        meta = backend.metadata
-        assert isinstance(meta, TTSBackendMetadata)
-
-    def test_metadata_fields(self):
-        backend = _make_backend(lang_code="a")
-        meta = backend.metadata
-        assert meta.id == "kokoro"
-        assert meta.name == "Kokoro"
-        assert "Kokoro" in meta.description
-
-
-class TestKokoroBackendInit:
-    def test_stores_lang_code(self):
-        backend = _make_backend(lang_code="b")
-        assert backend._lang_code == "b"
-
-    def test_default_repo_id(self):
-        with patch("abogen.tts_backends.kokoro._load_kpipeline") as load:
-            load.return_value = _FakePipeline
-            from abogen.tts_backends.kokoro import KokoroBackend
-
-            b = KokoroBackend(lang_code="a")
-            assert b._pipeline.repo_id == "hexgrad/Kokoro-82M"
-
-    def test_custom_repo_id(self):
-        backend = _make_backend(lang_code="a", repo_id="custom/repo")
-        assert backend._pipeline.repo_id == "custom/repo"
-
-    def test_default_device(self):
-        backend = _make_backend(lang_code="a")
-        assert backend._pipeline.device == "cpu"
-
-    def test_custom_device(self):
-        backend = _make_backend(lang_code="a", device="cuda")
-        assert backend._pipeline.device == "cuda"
-
-
-class TestKokoroBackendCall:
-    def test_call_delegates_to_pipeline(self):
-        backend = _make_backend(lang_code="a")
-        results = list(backend("hello", voice="af_heart", speed=1.2, split_pattern=r"\n"))
-        assert len(results) == 1
-        assert results[0].graphemes == "hello"
-
-    def test_call_returns_iterator(self):
-        backend = _make_backend(lang_code="a")
-        result = backend("test", voice="af_heart")
-        assert hasattr(result, "__iter__")
-
-    def test_call_with_voice_tensor(self):
-        backend = _make_backend(lang_code="a")
-        voice_tensor = np.ones((1, 256), dtype="float32")
-        results = list(backend("test", voice=voice_tensor))
-        assert len(results) == 1
-
-    def test_call_default_speed(self):
-        backend = _make_backend(lang_code="a")
-        # Should not raise with default speed
-        list(backend("text", voice="af_heart"))
-
-    def test_call_default_split_pattern_is_none(self):
-        backend = _make_backend(lang_code="a")
-        # split_pattern defaults to None
-        list(backend("text", voice="af_heart"))
-
-
-class TestLoadSingleVoice:
-    def test_load_single_voice_delegates(self):
-        backend = _make_backend(lang_code="a")
-        tensor = backend.load_single_voice("af_heart")
-        assert isinstance(tensor, np.ndarray)
-        assert tensor.shape == (1, 256)
-
-    def test_load_single_voice_caches(self):
-        backend = _make_backend(lang_code="a")
-        t1 = backend.load_single_voice("af_heart")
-        t2 = backend.load_single_voice("af_heart")
-        assert t1 is t2  # same object
-
-
-class TestSynthesize:
-    def test_synthesize_returns_bytes(self):
-        backend = _make_backend(lang_code="a")
-        result = backend.synthesize("hello", voice="af_heart")
-        assert isinstance(result, bytes)
-
-    def test_synthesize_nonempty(self):
-        backend = _make_backend(lang_code="a")
-        result = backend.synthesize("hello", voice="af_heart")
-        assert len(result) > 0
-
-    def test_synthesize_with_speed(self):
-        backend = _make_backend(lang_code="a")
-        result = backend.synthesize("hello", voice="af_heart", speed=1.5)
-        assert isinstance(result, bytes)
-
-    def test_synthesize_empty_text(self):
-        backend = _make_backend(lang_code="a")
-        # Empty text produces no segments
-        result = backend.synthesize("", voice="af_heart")
-        assert isinstance(result, bytes)
-
-
-class TestProtocolMethods:
-    def test_get_available_voices(self):
-        backend = _make_backend(lang_code="a")
-        voices = backend.get_available_voices()
-        assert isinstance(voices, list)
-        assert len(voices) > 0
-        assert all(isinstance(v, str) for v in voices)
-
-    def test_get_supported_formats(self):
-        backend = _make_backend(lang_code="a")
-        formats = backend.get_supported_formats()
-        assert "pcm_float32" in formats
-
-    def test_get_info(self):
-        backend = _make_backend(lang_code="a")
-        info = backend.get_info()
-        assert info["id"] == "kokoro"
-        assert info["name"] == "Kokoro"
-        assert info["lang_code"] == "a"
-
-
-class TestRegistration:
-    def test_factory_creates_kokoro_backend(self):
-        from abogen.tts_backends.kokoro import create_kokoro_backend, KokoroBackend
-
-        with patch("abogen.tts_backends.kokoro._load_kpipeline") as load:
-            load.return_value = _FakePipeline
-            backend = create_kokoro_backend(lang_code="a")
-            assert isinstance(backend, KokoroBackend)
-
-    def test_registry_has_kokoro(self):
-        import abogen.tts_backends  # noqa: F401
-        from abogen.tts_backend_registry import _registry
-
-        meta = _registry.get_metadata("kokoro")
-        assert meta.id == "kokoro"
-        assert meta.name == "Kokoro"
-
-    def test_registry_factory_returns_kokoro_backend(self):
-        import abogen.tts_backends  # noqa: F401
-        from abogen.tts_backend_registry import _registry
-        from abogen.tts_backends.kokoro import KokoroBackend
-
-        factory = _registry._factories["kokoro"]
-        with patch("abogen.tts_backends.kokoro._load_kpipeline") as load:
-            load.return_value = _FakePipeline
-            backend = factory(lang_code="a")
-            assert isinstance(backend, KokoroBackend)

tests/test_preview_applies_manual_overrides.py

@@ -19,7 +19,7 @@ def test_preview_applies_manual_override_before_normalization(monkeypatch):
 
     # And stub the kokoro pipeline path so generate_preview_audio won't proceed.
     # We'll instead validate by calling the override logic through generate_preview_audio
-    # with provider=supertonic and stub create_backend to capture input.
+    # with provider=supertonic and stub SupertonicPipeline to capture input.
     captured = {}
 
     class DummyPipeline:
@@ -30,16 +30,11 @@ def test_preview_applies_manual_override_before_normalization(monkeypatch):
             captured["text"] = text
             return iter(())
 
-    from abogen import tts_backend_registry
-
-    original_create_backend = tts_backend_registry.create_backend
-
-    def _mock_create_backend(backend_id, **kwargs):
-        if backend_id == "supertonic":
-            return DummyPipeline(**kwargs)
-        return original_create_backend(backend_id, **kwargs)
-
-    monkeypatch.setattr(tts_backend_registry, "create_backend", _mock_create_backend)
+    monkeypatch.setitem(
+        __import__("sys").modules,
+        "abogen.tts_supertonic",
+        type("M", (), {"SupertonicPipeline": DummyPipeline}),
+    )
 
     try:
         preview.generate_preview_audio(

tests/test_tts_backend.py

@@ -1,314 +0,0 @@
-from dataclasses import dataclass
-
-from abogen.tts_backend import TTSBackendMetadata
-from abogen.tts_backend_registry import TTSBackendRegistry
-
-
-class TestTTSBackendMetadata:
-    def test_is_frozen_dataclass(self):
-        assert dataclass(TTSBackendMetadata)
-
-    def test_fields_are_present(self):
-        meta = TTSBackendMetadata(
-            id="test",
-            name="Test Backend",
-            description="A test backend",
-        )
-        assert meta.id == "test"
-        assert meta.name == "Test Backend"
-        assert meta.description == "A test backend"
-
-    def test_voices_field_default_empty(self):
-        meta = TTSBackendMetadata(
-            id="test",
-            name="Test",
-            description="Test backend",
-        )
-        assert meta.voices == ()
-
-    def test_voices_field_stored(self):
-        meta = TTSBackendMetadata(
-            id="test",
-            name="Test",
-            description="Test backend",
-            voices=("v1", "v2"),
-        )
-        assert meta.voices == ("v1", "v2")
-
-    def test_is_immutable(self):
-        import pytest
-
-        meta = TTSBackendMetadata(
-            id="kokoro",
-            name="Kokoro",
-            description="Test",
-        )
-        with pytest.raises(Exception):
-            meta.id = "changed"
-
-
-class TestTTSBackendRegistry:
-    def test_register_and_list(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(id="a", name="A", description="Backend A")
-        registry.register(metadata=meta, factory=lambda: None)
-
-        backends = registry.list_backends()
-        assert len(backends) == 1
-        assert backends[0].id == "a"
-
-    def test_list_multiple(self):
-        registry = TTSBackendRegistry()
-        meta_a = TTSBackendMetadata(id="a", name="A", description="A")
-        meta_b = TTSBackendMetadata(id="b", name="B", description="B")
-        registry.register(metadata=meta_a, factory=lambda: None)
-        registry.register(metadata=meta_b, factory=lambda: None)
-
-        backends = registry.list_backends()
-        ids = [b.id for b in backends]
-        assert "a" in ids
-        assert "b" in ids
-
-    def test_get_metadata(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(id="x", name="X", description="X backend")
-        registry.register(metadata=meta, factory=lambda: None)
-
-        result = registry.get_metadata("x")
-        assert result.id == "x"
-        assert result.name == "X"
-
-    def test_get_metadata_unknown_raises(self):
-        import pytest
-
-        registry = TTSBackendRegistry()
-        with pytest.raises(KeyError, match="Unknown backend: nope"):
-            registry.get_metadata("nope")
-
-    def test_create_backend(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(id="test", name="Test", description="Test backend")
-
-        def factory(**kwargs):
-            return {"created": True, "kwargs": kwargs}
-
-        registry.register(metadata=meta, factory=factory)
-        result = registry.create_backend("test", foo="bar")
-
-        assert result == {"created": True, "kwargs": {"foo": "bar"}}
-
-    def test_create_backend_unknown_raises(self):
-        import pytest
-
-        registry = TTSBackendRegistry()
-        with pytest.raises(KeyError, match="Unknown backend: missing"):
-            registry.create_backend("missing")
-
-    def test_register_overwrites(self):
-        registry = TTSBackendRegistry()
-        meta1 = TTSBackendMetadata(id="x", name="V1", description="First")
-        meta2 = TTSBackendMetadata(id="x", name="V2", description="Second")
-        registry.register(metadata=meta1, factory=lambda: "v1")
-        registry.register(metadata=meta2, factory=lambda: "v2")
-
-        result = registry.get_metadata("x")
-        assert result.name == "V2"
-        assert registry.create_backend("x") == "v2"
-
-
-class TestBackendRegistration:
-    """Tests that existing backends are auto-registered."""
-
-    def test_import_triggers_registration(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        backends = _registry.list_backends()
-        ids = [b.id for b in backends]
-        assert "kokoro" in ids
-        assert "supertonic" in ids
-
-    def test_kokoro_metadata(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        meta = _registry.get_metadata("kokoro")
-        assert meta.id == "kokoro"
-        assert meta.name == "Kokoro"
-        assert "Kokoro" in meta.description
-
-    def test_supertonic_metadata(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        meta = _registry.get_metadata("supertonic")
-        assert meta.id == "supertonic"
-        assert meta.name == "SuperTonic"
-        assert "SuperTonic" in meta.description
-
-    def test_kokoro_metadata_has_voices(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        meta = _registry.get_metadata("kokoro")
-        assert isinstance(meta.voices, tuple)
-        assert len(meta.voices) > 0
-        assert all(isinstance(v, str) for v in meta.voices)
-
-    def test_supertonic_metadata_has_voices(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        meta = _registry.get_metadata("supertonic")
-        assert isinstance(meta.voices, tuple)
-        assert len(meta.voices) == 10
-        assert meta.voices == ("M1", "M2", "M3", "M4", "M5", "F1", "F2", "F3", "F4", "F5")
-
-    def test_kokoro_factory_callable(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        factory = _registry._factories["kokoro"]
-        assert callable(factory)
-
-    def test_supertonic_factory_callable(self):
-        import abogen.tts_backends  # noqa: F401
-
-        from abogen.tts_backend_registry import _registry
-
-        factory = _registry._factories["supertonic"]
-        assert callable(factory)
-
-    def test_kokoro_metadata_voices_match_registry(self):
-        """Ensure the metadata property on the instance shares voices with registry."""
-        from abogen.tts_backends.kokoro import _KOKORO_METADATA
-        from abogen.tts_backend_registry import _registry
-
-        registry_meta = _registry.get_metadata("kokoro")
-        assert _KOKORO_METADATA is registry_meta
-        assert _KOKORO_METADATA.voices == registry_meta.voices
-
-    def test_supertonic_metadata_voices_match_registry(self):
-        """Ensure the metadata property on the instance shares voices with registry."""
-        from abogen.tts_backends.supertonic import _SUPERTONIC_METADATA
-        from abogen.tts_backend_registry import _registry
-
-        registry_meta = _registry.get_metadata("supertonic")
-        assert _SUPERTONIC_METADATA is registry_meta
-        assert _SUPERTONIC_METADATA.voices == registry_meta.voices
-
-
-class TestResolveBackendForVoice:
-    """Tests for the resolve_backend_for_voice method."""
-
-    def test_empty_spec_returns_fallback(self):
-        registry = TTSBackendRegistry()
-        assert registry.resolve_backend_for_voice("", fallback="kokoro") == "kokoro"
-        assert registry.resolve_backend_for_voice("", fallback="supertonic") == "supertonic"
-
-    def test_none_spec_returns_fallback(self):
-        registry = TTSBackendRegistry()
-        assert registry.resolve_backend_for_voice(None, fallback="kokoro") == "kokoro"
-
-    def test_kokoro_formula_with_star_returns_kokoro(self):
-        registry = TTSBackendRegistry()
-        assert registry.resolve_backend_for_voice("af_nova*0.7") == "kokoro"
-
-    def test_kokoro_formula_with_plus_returns_kokoro(self):
-        registry = TTSBackendRegistry()
-        assert registry.resolve_backend_for_voice("af_nova*0.7+am_liam*0.3") == "kokoro"
-
-    def test_kokoro_voice_id_resolves_to_kokoro(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(
-            id="kokoro",
-            name="Kokoro",
-            description="Kokoro TTS",
-            voices=("af_nova", "am_liam"),
-        )
-        registry.register(metadata=meta, factory=lambda: None)
-
-        assert registry.resolve_backend_for_voice("af_nova") == "kokoro"
-        assert registry.resolve_backend_for_voice("am_liam") == "kokoro"
-
-    def test_supertonic_voice_id_resolves_to_supertonic(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(
-            id="supertonic",
-            name="SuperTonic",
-            description="SuperTonic TTS",
-            voices=("M1", "M2", "F1", "F2"),
-        )
-        registry.register(metadata=meta, factory=lambda: None)
-
-        assert registry.resolve_backend_for_voice("M1") == "supertonic"
-        assert registry.resolve_backend_for_voice("F2") == "supertonic"
-
-    def test_unknown_voice_returns_fallback(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(
-            id="kokoro",
-            name="Kokoro",
-            description="Kokoro TTS",
-            voices=("af_nova",),
-        )
-        registry.register(metadata=meta, factory=lambda: None)
-
-        assert registry.resolve_backend_for_voice("unknown_voice") == "kokoro"
-        assert registry.resolve_backend_for_voice("unknown_voice", fallback="supertonic") == "supertonic"
-
-    def test_case_insensitive_matching(self):
-        registry = TTSBackendRegistry()
-        meta = TTSBackendMetadata(
-            id="supertonic",
-            name="SuperTonic",
-            description="SuperTonic TTS",
-            voices=("M1", "F1"),
-        )
-        registry.register(metadata=meta, factory=lambda: None)
-
-        assert registry.resolve_backend_for_voice("m1") == "supertonic"
-        assert registry.resolve_backend_for_voice("f1") == "supertonic"
-
-    def test_default_fallback_is_kokoro(self):
-        registry = TTSBackendRegistry()
-        assert registry.resolve_backend_for_voice("unknown") == "kokoro"
-
-    def test_multiple_backends_resolution(self):
-        registry = TTSBackendRegistry()
-        kokoro_meta = TTSBackendMetadata(
-            id="kokoro",
-            name="Kokoro",
-            description="Kokoro TTS",
-            voices=("af_nova",),
-        )
-        supertonic_meta = TTSBackendMetadata(
-            id="supertonic",
-            name="SuperTonic",
-            description="SuperTonic TTS",
-            voices=("M1",),
-        )
-        registry.register(metadata=kokoro_meta, factory=lambda: None)
-        registry.register(metadata=supertonic_meta, factory=lambda: None)
-
-        assert registry.resolve_backend_for_voice("af_nova") == "kokoro"
-        assert registry.resolve_backend_for_voice("M1") == "supertonic"
-
-    def test_global_wrapper_resolve_backend_for_voice(self):
-        from abogen.tts_backend_registry import resolve_backend_for_voice
-
-        # Test with empty spec
-        assert resolve_backend_for_voice("") == "kokoro"
-
-        # Test with formula
-        assert resolve_backend_for_voice("af_nova*0.7") == "kokoro"
-
-        # Test with a registered voice
-        assert resolve_backend_for_voice("af_nova") == "kokoro"
-        assert resolve_backend_for_voice("M1") == "supertonic"

tests/test_tts_supertonic_unsupported_chars.py

@@ -1,6 +1,6 @@
 import numpy as np
 
-from abogen.tts_backends.supertonic import SupertonicBackend, SupertonicPipeline
+from abogen.tts_supertonic import SupertonicPipeline
 
 
 class _DummyTTS:
@@ -26,23 +26,13 @@ class _DummyTTS:
         return audio, 0.05
 
 
-def _make_pipeline() -> SupertonicPipeline:
+def test_supertonic_pipeline_strips_unsupported_characters_and_retries():
+    # Avoid importing/initializing real supertonic by manually constructing the pipeline.
     pipeline = SupertonicPipeline.__new__(SupertonicPipeline)
     pipeline.sample_rate = 24000
     pipeline.total_steps = 5
     pipeline.max_chunk_length = 1000
     pipeline._tts = _DummyTTS()
-    return pipeline
-
-
-def _make_backend() -> SupertonicBackend:
-    backend = SupertonicBackend.__new__(SupertonicBackend)
-    backend._pipeline = _make_pipeline()
-    return backend
-
-
-def test_supertonic_pipeline_strips_unsupported_characters_and_retries():
-    pipeline = _make_pipeline()
 
     segs = list(pipeline("Hello • world", voice="M1", speed=1.0))
     assert len(segs) == 1
@@ -53,56 +43,11 @@ def test_supertonic_pipeline_strips_unsupported_characters_and_retries():
 
 
 def test_supertonic_pipeline_drops_chunk_if_only_unsupported_characters():
-    pipeline = _make_pipeline()
+    pipeline = SupertonicPipeline.__new__(SupertonicPipeline)
+    pipeline.sample_rate = 24000
+    pipeline.total_steps = 5
+    pipeline.max_chunk_length = 1000
+    pipeline._tts = _DummyTTS()
 
     segs = list(pipeline("•", voice="M1", speed=1.0))
     assert segs == []
-
-
-# --- SupertonicBackend tests ---
-
-
-def test_backend_metadata():
-    backend = _make_backend()
-    meta = backend.metadata
-    assert meta.id == "supertonic"
-    assert meta.name == "SuperTonic"
-    assert "SuperTonic" in meta.description
-
-
-def test_backend_get_available_voices():
-    backend = _make_backend()
-    voices = backend.get_available_voices()
-    assert isinstance(voices, list)
-    assert "M1" in voices
-    assert "F1" in voices
-
-
-def test_backend_get_supported_formats():
-    backend = _make_backend()
-    formats = backend.get_supported_formats()
-    assert "wav" in formats
-
-
-def test_backend_get_info():
-    backend = _make_backend()
-    info = backend.get_info()
-    assert info["sample_rate"] == 24000
-    assert info["total_steps"] == 5
-    assert isinstance(info["voices"], list)
-
-
-def test_backend_call_delegates_to_pipeline():
-    backend = _make_backend()
-    segs = list(backend("Hello • world", voice="M1", speed=1.0))
-    assert len(segs) == 1
-    assert segs[0].audio.size > 0
-
-
-def test_backend_synthesize_returns_wav_bytes():
-    backend = _make_backend()
-    wav_bytes = backend.synthesize("Hello world", voice="M1", speed=1.0)
-    assert isinstance(wav_bytes, bytes)
-    assert len(wav_bytes) > 0
-    # WAV magic number
-    assert wav_bytes[:4] == b"RIFF"

tests/test_voice_cache.py

@@ -3,7 +3,7 @@ from typing import cast
 
 import pytest
 
-from abogen.tts_backend_registry import get_metadata
+from abogen.constants import VOICES_INTERNAL
 from abogen.voice_cache import (
     LocalEntryNotFoundError,
     _CACHED_VOICES,
@@ -66,4 +66,4 @@ def test_collect_required_voice_ids_includes_all():
     voices = _collect_required_voice_ids(cast(Job, job))
 
     assert {"af_nova", "am_liam", "am_michael"}.issubset(voices)
-    assert voices.issuperset(get_metadata("kokoro").voices)
+    assert voices.issuperset(VOICES_INTERNAL)

tests/test_voice_formula_resolution.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 from abogen.webui.conversion_runner import _resolve_voice, _supertonic_voice_from_spec
-from abogen.tts_backends.supertonic import DEFAULT_SUPERTONIC_VOICES
+from abogen.tts_supertonic import DEFAULT_SUPERTONIC_VOICES
 
 
 def test_resolve_voice_formula_without_pipeline_does_not_crash() -> None:

tests/test_voice_metadata.py

@@ -1,233 +0,0 @@
-import pytest
-
-from abogen.voice_metadata import VoiceMetadata
-
-
-class TestVoiceMetadataCreation:
-    def test_create_with_all_fields(self):
-        voice = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        assert voice.id == "af_alloy"
-        assert voice.display_name == "Alloy"
-        assert voice.language == "a"
-        assert voice.gender == "female"
-        assert voice.backend_id == "kokoro"
-
-    def test_create_supertonic_voice(self):
-        voice = VoiceMetadata(
-            id="M1",
-            display_name="Male 1",
-            language="en",
-            gender="male",
-            backend_id="supertonic",
-        )
-        assert voice.id == "M1"
-        assert voice.backend_id == "supertonic"
-
-    def test_create_with_unknown_gender(self):
-        voice = VoiceMetadata(
-            id="custom_voice",
-            display_name="Custom",
-            language="en",
-            gender="unknown",
-            backend_id="custom_backend",
-        )
-        assert voice.gender == "unknown"
-
-
-class TestVoiceMetadataImmutability:
-    def test_frozen_dataclass(self):
-        voice = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        with pytest.raises(AttributeError):
-            voice.id = "new_id"
-
-    def test_cannot_modify_display_name(self):
-        voice = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        with pytest.raises(AttributeError):
-            voice.display_name = "New Name"
-
-    def test_cannot_modify_backend_id(self):
-        voice = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        with pytest.raises(AttributeError):
-            voice.backend_id = "new_backend"
-
-
-class TestVoiceMetadataEquality:
-    def test_equal_voices_are_equal(self):
-        voice1 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        voice2 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        assert voice1 == voice2
-
-    def test_different_voices_are_not_equal(self):
-        voice1 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        voice2 = VoiceMetadata(
-            id="am_adam",
-            display_name="Adam",
-            language="a",
-            gender="male",
-            backend_id="kokoro",
-        )
-        assert voice1 != voice2
-
-    def test_different_backend_id_not_equal(self):
-        voice1 = VoiceMetadata(
-            id="custom",
-            display_name="Custom",
-            language="en",
-            gender="unknown",
-            backend_id="backend_a",
-        )
-        voice2 = VoiceMetadata(
-            id="custom",
-            display_name="Custom",
-            language="en",
-            gender="unknown",
-            backend_id="backend_b",
-        )
-        assert voice1 != voice2
-
-
-class TestVoiceMetadataHashing:
-    def test_hashable(self):
-        voice = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        assert hash(voice) is not None
-
-    def test_equal_voices_same_hash(self):
-        voice1 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        voice2 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        assert hash(voice1) == hash(voice2)
-
-    def test_usable_in_set(self):
-        voice1 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        voice2 = VoiceMetadata(
-            id="af_alloy",
-            display_name="Alloy",
-            language="a",
-            gender="female",
-            backend_id="kokoro",
-        )
-        voice3 = VoiceMetadata(
-            id="am_adam",
-            display_name="Adam",
-            language="a",
-            gender="male",
-            backend_id="kokoro",
-        )
-        voice_set = {voice1, voice2, voice3}
-        assert len(voice_set) == 2
-
-
-class TestVoiceMetadataUseCases:
-    def test_backend_populates_backend_id(self):
-        """Simulate how a backend would populate backend_id automatically."""
-
-        class MockBackend:
-            def __init__(self):
-                self._backend_id = "kokoro"
-
-            def get_voices(self):
-                return [
-                    VoiceMetadata(
-                        id="af_alloy",
-                        display_name="Alloy",
-                        language="a",
-                        gender="female",
-                        backend_id=self._backend_id,
-                    ),
-                ]
-
-        backend = MockBackend()
-        voices = backend.get_voices()
-        assert voices[0].backend_id == "kokoro"
-
-    def test_filter_by_language(self):
-        voices = [
-            VoiceMetadata(id="af_alloy", display_name="Alloy", language="a", gender="female", backend_id="kokoro"),
-            VoiceMetadata(id="jf_alpha", display_name="Alpha", language="j", gender="female", backend_id="kokoro"),
-            VoiceMetadata(id="am_adam", display_name="Adam", language="a", gender="male", backend_id="kokoro"),
-        ]
-        english_voices = [v for v in voices if v.language == "a"]
-        assert len(english_voices) == 2
-
-    def test_filter_by_gender(self):
-        voices = [
-            VoiceMetadata(id="af_alloy", display_name="Alloy", language="a", gender="female", backend_id="kokoro"),
-            VoiceMetadata(id="am_adam", display_name="Adam", language="a", gender="male", backend_id="kokoro"),
-            VoiceMetadata(id="am_puck", display_name="Puck", language="a", gender="male", backend_id="kokoro"),
-        ]
-        male_voices = [v for v in voices if v.gender == "male"]
-        assert len(male_voices) == 2
-
-    def test_filter_by_backend(self):
-        voices = [
-            VoiceMetadata(id="af_alloy", display_name="Alloy", language="a", gender="female", backend_id="kokoro"),
-            VoiceMetadata(id="M1", display_name="Male 1", language="en", gender="male", backend_id="supertonic"),
-        ]
-        kokoro_voices = [v for v in voices if v.backend_id == "kokoro"]
-        assert len(kokoro_voices) == 1
-        assert kokoro_voices[0].id == "af_alloy"