extract normalize_text_for_pipeline to domain/normalization.py

fix(webui): allow large chapter forms

.gitattributes

@@ -0,0 +1,15 @@
+*.py text eol=lf
+*.md text eol=lf
+*.yml text eol=lf
+*.yaml text eol=lf
+*.toml text eol=lf
+*.json text eol=lf
+*.txt text eol=lf
+*.html text eol=lf
+*.css text eol=lf
+*.js text eol=lf
+*.sh text eol=lf
+*.cfg text eol=lf
+*.ini text eol=lf
+*.svg text eol=lf
+*.j2 text eol=lf

.github/workflows/test_pip.yml

@@ -1,7 +1,9 @@
-name: pip install
-run-name: pip install
+name: CI
+run-name: CI
+
 on:
   push:
+    branches: [main]
     paths:
       - '**.py'
       - 'pyproject.toml'
@@ -11,23 +13,41 @@ on:
       - 'pyproject.toml'
       - '.github/workflows/**'
   workflow_dispatch:
+
 jobs:
-  install-and-run:
+  test:
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        os: [ubuntu-latest, macos-14, windows-latest]
         python-version: ['3.12']
       fail-fast: false
-    continue-on-error: true
     runs-on: ${{ matrix.os }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v7
+
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
-      - name: Install from repository
-        run: python -m pip install .
-      #- name: Run abogen
-      #  run: abogen
+
+      - 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

.github/workflows/test_publish.yml

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

README.md

@@ -38,8 +38,6 @@ This method handles everything automatically - installing all dependencies inclu
 #### <b>OPTION 2: Install using uv</b>
 First, [install uv](https://docs.astral.sh/uv/getting-started/installation/) if you haven't already.
 
-The CUDA extras install both GPU-accelerated Kokoro (via PyTorch) and Supertonic (via onnxruntime-gpu).
-
 ```bash
 # For NVIDIA GPUs (CUDA 12.8) - Recommended
 uv tool install --python 3.12 abogen[cuda] --extra-index-url https://download.pytorch.org/whl/cu128 --index-strategy unsafe-best-match
@@ -67,9 +65,6 @@ venv\Scripts\activate
 # We need to use an older version of PyTorch (2.8.0) until this issue is fixed: https://github.com/pytorch/pytorch/issues/166628
 pip install torch==2.8.0+cu128 torchvision==0.23.0+cu128 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu128
 
-# Also install onnxruntime-gpu for Supertonic GPU acceleration:
-pip install onnxruntime-gpu
-
 # For AMD GPUs:
 # Not supported yet, because ROCm is not available on Windows. Use Linux if you have AMD GPU.
 
@@ -178,7 +173,7 @@ Abogen offers **two interfaces**, but currently they have different feature sets
 
 | Command | Interface | Features |
 |---------|-----------|----------|
-| `abogen` | PyQt6 Desktop GUI | Stable core features + **Supertonic TTS**|
+| `abogen` | PyQt6 Desktop GUI | Stable core features |
 | `abogen-web` | Flask Web UI | Core features + **Supertonic TTS**, **LLM Normalization**, **Audiobookshelf Integration** and more! |
 
 > **Note:** The Web UI is under active development. We are working to integrate these new features into the PyQt desktop app. until then, the Web UI provides the most feature-rich experience.
@@ -412,17 +407,17 @@ When Audiobookshelf sits behind Nginx Proxy Manager (NPM), make sure the API pat
 1. Create a **Proxy Host** that points to your ABS container or host (default forward port `13378`).
 2. Under the **SSL** tab, enable your certificate and tick **Force SSL** if you want HTTPS only.
 3. In the **Advanced** tab, append the snippet below so bearer tokens, client IPs, and large uploads survive the proxy hop:
-```nginx
-proxy_set_header Host $host;
-proxy_set_header X-Real-IP $remote_addr;
-proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
-proxy_set_header X-Forwarded-Proto $scheme;
-proxy_set_header X-Forwarded-Host $host;
-proxy_set_header X-Forwarded-Port $server_port;
-proxy_set_header Authorization $http_authorization;
-client_max_body_size 5g;
-proxy_read_timeout 300s;
-proxy_connect_timeout 300s;
+   ```nginx
+  proxy_set_header Host $host;
+  proxy_set_header X-Real-IP $remote_addr;
+  proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+  proxy_set_header X-Forwarded-Proto $scheme;
+  proxy_set_header X-Forwarded-Host $host;
+  proxy_set_header X-Forwarded-Port $server_port;
+  proxy_set_header Authorization $http_authorization;
+  client_max_body_size 5g;
+  proxy_read_timeout 300s;
+  proxy_connect_timeout 300s;
    ```
 4. Disable **Block Common Exploits** (it strips Authorization headers in some NPM builds).
 5. Enable **Websockets Support** on the main proxy screen (Audiobookshelf uses it for the web UI, and it keeps the reverse proxy configuration consistent).

WINDOWS_INSTALL.bat

@@ -323,13 +323,6 @@ if /I "%IS_NVIDIA%"=="true" (
             pause
             exit /b
         )
-        echo Installing onnxruntime-gpu for Supertonic GPU acceleration...
-        %PYTHON_CONSOLE_PATH% -m uv pip install --system onnxruntime-gpu
-        if errorlevel 1 (
-            echo Failed to install onnxruntime-gpu.
-            pause
-            exit /b
-        )
     ) else (
         echo CUDA is available on NVIDIA GPU.
     )
@@ -355,13 +348,6 @@ if /I "%IS_NVIDIA%"=="true" (
             pause
             exit /b
         )
-        echo Installing onnxruntime-gpu for Supertonic GPU acceleration...
-        %PYTHON_CONSOLE_PATH% -m uv pip install --system onnxruntime-gpu
-        if errorlevel 1 (
-            echo Failed to install onnxruntime-gpu.
-            pause
-            exit /b
-        )
     )
 )
 

abogen/constants.py

@@ -29,57 +29,6 @@ LANGUAGE_DESCRIPTIONS = {
     "z": "Mandarin Chinese",
 }
 
-# Mapping from Kokoro single-letter language codes to ISO 3166-1 alpha-2 country codes
-# Used for loading flag icons
-KOKORO_LANG_TO_COUNTRY = {
-    "a": "us",  # American English -> United States
-    "b": "gb",  # British English -> United Kingdom
-    "e": "es",  # Spanish -> Spain
-    "f": "fr",  # French -> France
-    "h": "in",  # Hindi -> India
-    "i": "it",  # Italian -> Italy
-    "j": "jp",  # Japanese -> Japan
-    "p": "br",  # Brazilian Portuguese -> Brazil
-    "z": "cn",  # Mandarin Chinese -> China
-}
-
-# Mapping from Supertonic ISO 639-1 language codes to ISO 3166-1 alpha-2 country codes
-# Used for loading flag icons in the Supertonic language picker
-SUPERTONIC_LANG_TO_COUNTRY = {
-    "en": "gb",
-    "ko": "kr",
-    "ja": "jp",
-    "ar": "ae",
-    "bg": "bg",
-    "cs": "cz",
-    "da": "dk",
-    "de": "de",
-    "el": "gr",
-    "es": "es",
-    "et": "ee",
-    "fi": "fi",
-    "fr": "fr",
-    "hi": "in",
-    "hr": "hr",
-    "hu": "hu",
-    "id": "id",
-    "it": "it",
-    "lt": "lt",
-    "lv": "lv",
-    "nl": "nl",
-    "pl": "pl",
-    "pt": "pt",
-    "ro": "ro",
-    "ru": "ru",
-    "sk": "sk",
-    "sl": "si",
-    "sv": "se",
-    "tr": "tr",
-    "uk": "ua",
-    "vi": "vn",
-    "na": "na",
-}
-
 # Supported sound formats
 SUPPORTED_SOUND_FORMATS = [
     "wav",
@@ -114,64 +63,6 @@ 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/domain/audio_helpers.py

@@ -0,0 +1,118 @@
+"""Audio helper utilities.
+
+Functions for building ffmpeg commands, converting audio formats,
+and applying chapter metadata to MP4 files.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+import numpy as np
+
+
+SAMPLE_RATE = 24000
+
+
+def build_ffmpeg_command(path: Path, fmt: str, metadata: Optional[Dict[str, str]] = None) -> list[str]:
+    from abogen.infrastructure.exporters import ExportService
+
+    base = [
+        "ffmpeg",
+        "-y",
+        "-f",
+        "f32le",
+        "-ar",
+        str(SAMPLE_RATE),
+        "-ac",
+        "1",
+        "-i",
+        "pipe:0",
+    ]
+    if fmt == "mp3":
+        base += ["-c:a", "libmp3lame", "-qscale:a", "2"]
+    elif fmt == "opus":
+        base += ["-c:a", "libopus", "-b:a", "24000"]
+    elif fmt == "m4b":
+        base += ["-c:a", "aac", "-q:a", "2", "-movflags", "+faststart+use_metadata_tags"]
+    else:
+        base += ["-c:a", "copy"]
+
+    if metadata:
+        svc = ExportService()
+        base.extend(svc._metadata_to_ffmpeg_args(metadata))
+    base.append(str(path))
+    return base
+
+
+def to_float32(audio_segment) -> np.ndarray:
+    if audio_segment is None:
+        return np.zeros(0, dtype="float32")
+
+    tensor = audio_segment
+    if hasattr(tensor, "detach"):
+        tensor = tensor.detach()
+    if hasattr(tensor, "cpu"):
+        try:
+            tensor = tensor.cpu()
+        except Exception:
+            pass
+    if hasattr(tensor, "numpy"):
+        return np.asarray(tensor.numpy(), dtype="float32").reshape(-1)
+    return np.asarray(tensor, dtype="float32").reshape(-1)
+
+
+def apply_m4b_chapters_with_mutagen(
+    audio_path: Path,
+    chapters: List[Dict[str, Any]],
+) -> bool:
+    """Apply chapter atoms to an MP4/M4B file using mutagen.
+
+    Returns True if chapters were written, False otherwise.
+    Raises ImportError if mutagen is not installed.
+    """
+    if not chapters:
+        return False
+
+    from fractions import Fraction
+    from mutagen.mp4 import MP4, MP4Chapter  # type: ignore[import]
+
+    mp4 = MP4(str(audio_path))
+
+    chapter_objects: List[MP4Chapter] = []
+    for index, entry in enumerate(sorted(chapters, key=lambda item: float(item.get("start") or 0.0))):
+        start_raw = entry.get("start")
+        if start_raw is None:
+            continue
+        try:
+            start_seconds = max(0.0, float(start_raw))
+        except (TypeError, ValueError):
+            continue
+
+        title_value = entry.get("title")
+        title_text = str(title_value) if title_value else f"Chapter {index + 1}"
+
+        start_fraction = Fraction(int(round(start_seconds * 1000)), 1000)
+        chapter_atom = MP4Chapter(start_fraction, title_text)
+
+        end_raw = entry.get("end")
+        if end_raw is not None:
+            try:
+                end_seconds = float(end_raw)
+            except (TypeError, ValueError):
+                end_seconds = None
+            if end_seconds is not None and end_seconds > start_seconds:
+                chapter_atom.end = Fraction(int(round(end_seconds * 1000)), 1000)
+
+        chapter_objects.append(chapter_atom)
+
+    if not chapter_objects:
+        return False
+
+    from typing import cast
+
+    mp4.chapters = cast(Any, chapter_objects)
+    mp4.save()
+
+    return True

abogen/domain/chapter_overrides.py

@@ -0,0 +1,92 @@
+from __future__ import annotations
+
+from typing import Any, Dict, List, Optional, Tuple
+
+from abogen.text_extractor import ExtractedChapter
+from abogen.domain.voice_utils import coerce_truthy
+
+
+def apply_chapter_overrides(
+    extracted: List[ExtractedChapter],
+    overrides: List[Dict[str, Any]],
+) -> Tuple[List[ExtractedChapter], Dict[str, str], List[str]]:
+    if not overrides:
+        return [], {}, []
+
+    selected: List[ExtractedChapter] = []
+    metadata_updates: Dict[str, str] = {}
+    diagnostics: List[str] = []
+
+    for position, payload in enumerate(overrides):
+        if not isinstance(payload, dict):
+            diagnostics.append(
+                f"Skipped chapter override at position {position + 1}: unsupported payload type {type(payload).__name__}."
+            )
+            continue
+
+        enabled = coerce_truthy(payload.get("enabled", True))
+        payload["enabled"] = enabled
+        if not enabled:
+            continue
+
+        metadata_payload = payload.get("metadata") or {}
+        if isinstance(metadata_payload, dict):
+            for key, value in metadata_payload.items():
+                if value is None:
+                    continue
+                metadata_updates[str(key)] = str(value)
+
+        base: Optional[ExtractedChapter] = None
+        idx_candidate = payload.get("index")
+        idx_normalized: Optional[int] = None
+        if isinstance(idx_candidate, int):
+            idx_normalized = idx_candidate
+        elif isinstance(idx_candidate, str):
+            try:
+                idx_normalized = int(idx_candidate)
+            except ValueError:
+                idx_normalized = None
+        if idx_normalized is not None and 0 <= idx_normalized < len(extracted):
+            base = extracted[idx_normalized]
+            payload["index"] = idx_normalized
+
+        if base is None:
+            source_title = payload.get("source_title")
+            if isinstance(source_title, str):
+                base = next((chapter for chapter in extracted if chapter.title == source_title), None)
+
+        if base is None:
+            candidate_title = payload.get("title")
+            if isinstance(candidate_title, str):
+                base = next((chapter for chapter in extracted if chapter.title == candidate_title), None)
+
+        text_override = payload.get("text")
+        if text_override is not None:
+            text_value = str(text_override)
+        elif base is not None:
+            text_value = base.text
+        else:
+            diagnostics.append(
+                f"Skipped chapter override at position {position + 1}: no text provided and no matching source chapter found."
+            )
+            continue
+
+        title_override = payload.get("title")
+        if title_override is not None:
+            title_value = str(title_override)
+        elif base is not None:
+            title_value = base.title
+        else:
+            title_value = f"Chapter {position + 1}"
+
+        if base and not payload.get("source_title"):
+            payload["source_title"] = base.title
+
+        payload["title"] = title_value
+        payload["text"] = text_value
+        payload["characters"] = len(text_value)
+        payload.setdefault("order", payload.get("order", position))
+
+        selected.append(ExtractedChapter(title=title_value, text=text_value))
+
+    return selected, metadata_updates, diagnostics

abogen/domain/chapter_titles.py

@@ -0,0 +1,204 @@
+from __future__ import annotations
+
+import re
+from typing import List, Tuple
+
+
+_HEADING_SANITIZE_RE = re.compile(r"[^a-z0-9]+")
+_HEADING_NUMBER_PREFIX_RE = re.compile(
+    r"^\s*(?P<number>(?:\d+|[ivxlcdm]+))(?P<suffix>(?:[\s.:;-].*)?)$",
+    re.IGNORECASE,
+)
+_ACRONYM_ALLOWLIST = {
+    "AI", "API", "CPU", "DIY", "GPU", "HTML", "HTTP", "HTTPS", "ID",
+    "JSON", "MP3", "MP4", "M4B", "NASA", "OCR", "PDF", "SQL", "TV",
+    "TTS", "UK", "UN", "UFO", "OK", "URL", "USA", "US", "VR",
+}
+_ROMAN_NUMERAL_CHARS = frozenset("IVXLCDM")
+_CAPS_WORD_RE = re.compile(r"[A-Z][A-Z0-9'\u2019-]*")
+
+
+def simplify_heading_text(text: str) -> str:
+    raw = str(text or "").strip().lower()
+    if not raw:
+        return ""
+    simplified = _HEADING_SANITIZE_RE.sub("", raw)
+    if simplified.startswith("chapter"):
+        simplified = simplified[7:]
+    return simplified
+
+
+def headings_equivalent(left: str, right: str) -> bool:
+    simple_left = simplify_heading_text(left)
+    simple_right = simplify_heading_text(right)
+    if not simple_left or not simple_right:
+        return False
+    if simple_left == simple_right:
+        return True
+    if simple_right.startswith(simple_left):
+        return True
+    if simple_left.startswith(simple_right):
+        return True
+    if len(simple_left) > 5 and simple_left in simple_right:
+        return True
+    return False
+
+
+def strip_duplicate_heading_line(text: str, heading: str) -> Tuple[str, bool]:
+    source_text = str(text or "")
+    if not source_text:
+        return source_text, False
+    normalized_heading = simplify_heading_text(heading)
+    if not normalized_heading:
+        return source_text, False
+    lines = source_text.splitlines()
+    new_lines: List[str] = []
+    removed = False
+    for line in lines:
+        stripped = line.strip()
+        if not removed and stripped:
+            if headings_equivalent(stripped, heading):
+                removed = True
+                continue
+        new_lines.append(line)
+    if not removed:
+        return source_text, False
+    while new_lines and not new_lines[0].strip():
+        new_lines.pop(0)
+    return "\n".join(new_lines), True
+
+
+def normalize_caps_word(word: str) -> str:
+    upper = word.upper()
+    letters = [char for char in upper if char.isalpha()]
+    if not letters:
+        return word
+    if upper in _ACRONYM_ALLOWLIST:
+        return word
+    if len(letters) <= 1:
+        return word
+    if all(char in _ROMAN_NUMERAL_CHARS for char in letters) and len(letters) <= 7:
+        return word
+
+    parts = re.split(r"(['\-\u2019])", word)
+    normalized_parts: List[str] = []
+    for part in parts:
+        if part in {"'", "-", "\u2019"}:
+            normalized_parts.append(part)
+            continue
+        if not part:
+            continue
+        normalized_parts.append(part[0].upper() + part[1:].lower())
+    return "".join(normalized_parts) or word
+
+
+def normalize_chapter_opening_caps(text: str) -> Tuple[str, bool]:
+    if not text:
+        return text, False
+
+    leading_len = len(text) - len(text.lstrip())
+    leading = text[:leading_len]
+    working = text[leading_len:]
+    if not working:
+        return text, False
+
+    builder: List[str] = []
+    pos = 0
+    changed = False
+
+    while pos < len(working):
+        char = working[pos]
+        if char in "\r\n":
+            builder.append(working[pos:])
+            pos = len(working)
+            break
+        if char.isspace():
+            builder.append(char)
+            pos += 1
+            continue
+        if char.islower():
+            builder.append(working[pos:])
+            pos = len(working)
+            break
+        if not char.isalpha():
+            builder.append(char)
+            pos += 1
+            continue
+
+        match = _CAPS_WORD_RE.match(working, pos)
+        if not match:
+            builder.append(char)
+            pos += 1
+            continue
+
+        word = match.group(0)
+        if any(ch.islower() for ch in word):
+            builder.append(working[pos:])
+            pos = len(working)
+            break
+
+        normalized = normalize_caps_word(word)
+        if normalized != word:
+            changed = True
+        builder.append(normalized)
+        pos = match.end()
+
+    if pos < len(working):
+        builder.append(working[pos:])
+
+    if not changed:
+        return text, False
+
+    return leading + "".join(builder), True
+
+
+def format_spoken_chapter_title(title: str, index: int, apply_prefix: bool) -> str:
+    base = str(title or "").strip()
+    if not base:
+        return f"Chapter {index}" if apply_prefix else ""
+    if not apply_prefix:
+        return base
+    lowered = base.lower()
+    if lowered.startswith("chapter") and (len(lowered) == 7 or not lowered[7].isalpha()):
+        return base
+    match = _HEADING_NUMBER_PREFIX_RE.match(base)
+    if match:
+        number = match.group("number") or ""
+        suffix = match.group("suffix") or ""
+        cleaned_suffix = suffix.lstrip(" .,:;-_ \t\u2013\u2014\u00b7\u2022")
+        if cleaned_suffix:
+            return f"Chapter {number}. {cleaned_suffix}"
+        return f"Chapter {number}"
+    return base
+
+
+def apply_chapter_text_transforms(
+    text: str,
+    *,
+    heading_text: str,
+    raw_title: str,
+    strip_heading: bool,
+    normalize_caps: bool,
+) -> Tuple[str, bool, bool]:
+    """Strip duplicate heading and normalize opening caps.
+
+    Returns ``(text, heading_removed, caps_changed)``.
+    The caller is responsible for state updates (pending flags, logging,
+    dict mutation, ``continue``).
+    """
+    heading_removed = False
+    caps_changed = False
+
+    if strip_heading and heading_text:
+        text, heading_removed = strip_duplicate_heading_line(text, heading_text)
+        if not heading_removed and raw_title:
+            match = _HEADING_NUMBER_PREFIX_RE.match(raw_title)
+            if match:
+                number = match.group("number")
+                if number:
+                    text, heading_removed = strip_duplicate_heading_line(text, number)
+
+    if normalize_caps and text:
+        text, caps_changed = normalize_chapter_opening_caps(text)
+
+    return text, heading_removed, caps_changed

abogen/domain/chunk_utils.py

@@ -0,0 +1,75 @@
+"""Chunk processing utilities.
+
+Functions for grouping chunks, recording override usage, and selecting
+text for TTS synthesis.
+"""
+
+from __future__ import annotations
+
+from collections import defaultdict
+from typing import Any, Dict, Iterable, Mapping, Optional
+
+from abogen.pronunciation_store import increment_usage
+
+
+def safe_int(value: Any, default: int = 0) -> int:
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        return default
+
+
+def group_chunks_by_chapter(chunks: Iterable[Dict[str, Any]]) -> Dict[int, List[Dict[str, Any]]]:
+    grouped: Dict[int, List[Dict[str, Any]]] = defaultdict(list)
+    for entry in chunks or []:
+        if not isinstance(entry, dict):
+            continue
+        try:
+            chapter_index = int(entry.get("chapter_index", 0))
+        except (TypeError, ValueError):
+            chapter_index = 0
+        grouped[chapter_index].append(dict(entry))
+
+    for chapter_index, items in grouped.items():
+        items.sort(key=lambda payload: safe_int(payload.get("chunk_index")))
+
+    return grouped
+
+
+def record_override_usage(
+    job: Any,
+    usage_counter: Mapping[str, int],
+    token_map: Mapping[str, str],
+) -> None:
+    if not usage_counter:
+        return
+
+    language = getattr(job, "language", "") or "a"
+    for normalized, amount in usage_counter.items():
+        if amount <= 0:
+            continue
+        token_value = token_map.get(normalized, normalized)
+        try:
+            increment_usage(language=language, token=token_value, amount=int(amount))
+        except Exception:  # pragma: no cover - defensive logging
+            job.add_log(f"Failed to record usage for override {token_value}", level="warning")
+
+
+def chunk_text_for_tts(entry: Mapping[str, Any]) -> str:
+    """Choose the best source text for synthesis.
+
+    We must prefer the raw chunk text (``text`` / ``original_text``) so
+    manual/pronunciation overrides can match against the original tokens
+    (e.g. censored words like ``Unfu*k``).  ``normalized_text`` may have
+    already been run through ``normalize_for_pipeline``, which can remove
+    punctuation and prevent overrides from triggering.
+    """
+
+    if not isinstance(entry, Mapping):
+        return ""
+    return str(
+        entry.get("text")
+        or entry.get("original_text")
+        or entry.get("normalized_text")
+        or ""
+    ).strip()

abogen/domain/device.py

@@ -0,0 +1,31 @@
+from __future__ import annotations
+
+import platform as _platform
+
+
+def select_device() -> str:
+    """Return the best available compute device (``"mps"``, ``"cuda"``, or ``"cpu"``).
+
+    Checks ``torch`` availability at runtime so this can be called from
+    any context without requiring torch at import time.
+    """
+    try:
+        import torch  # type: ignore[import-not-found]
+    except Exception:
+        return "cpu"
+
+    system = _platform.system()
+    if system == "Darwin" and _platform.processor() == "arm":
+        try:
+            if torch.backends.mps.is_available():  # type: ignore[union-attr]
+                return "mps"
+        except Exception:
+            pass
+        return "cpu"
+
+    try:
+        if torch.cuda.is_available():  # type: ignore[union-attr]
+            return "cuda"
+    except Exception:
+        pass
+    return "cpu"

abogen/domain/file_type.py

@@ -0,0 +1,136 @@
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Dict, List, Tuple
+
+from abogen.text_extractor import ExtractedChapter
+
+
+_SIGNIFICANT_LENGTH_THRESHOLDS: Dict[str, int] = {"epub": 1000, "markdown": 500}
+_MIN_SHORT_CONTENT: Dict[str, int] = {"epub": 240, "markdown": 160}
+_STRUCTURAL_KEYWORDS = (
+    "preface",
+    "prologue",
+    "introduction",
+    "foreword",
+    "epilogue",
+    "afterword",
+    "appendix",
+    "acknowledgment",
+    "acknowledgement",
+)
+_STRUCTURAL_MIN_LENGTH = 120
+_MAX_SHORT_CHAPTERS = 2
+
+
+@dataclass
+class ChapterFilterResult:
+    kept: List[ExtractedChapter]
+    skipped: List[Tuple[str, int]]
+
+
+def infer_file_type(path: Path) -> str:
+    suffix = path.suffix.lower()
+    if suffix == ".epub":
+        return "epub"
+    if suffix in {".md", ".markdown"}:
+        return "markdown"
+    if suffix == ".pdf":
+        return "pdf"
+    if suffix == ".txt":
+        return "text"
+    return suffix.lstrip(".") or "text"
+
+
+def looks_structural(title: str) -> bool:
+    lowered = title.strip().lower()
+    if not lowered:
+        return False
+    return any(keyword in lowered for keyword in _STRUCTURAL_KEYWORDS)
+
+
+def chapter_label(file_type: str) -> str:
+    return "chapters" if file_type.lower() in {"epub", "markdown"} else "pages"
+
+
+def auto_select_relevant_chapters(
+    chapters: List[ExtractedChapter],
+    file_type: str,
+) -> ChapterFilterResult:
+    if not chapters:
+        return ChapterFilterResult(kept=[], skipped=[])
+
+    normalized = file_type.lower()
+    threshold = _SIGNIFICANT_LENGTH_THRESHOLDS.get(normalized, 0)
+    min_short = _MIN_SHORT_CONTENT.get(normalized, 0)
+
+    kept: List[ExtractedChapter] = []
+    skipped: List[Tuple[str, int]] = []
+    short_kept = 0
+
+    for chapter in chapters:
+        stripped = chapter.text.strip()
+        length = len(stripped)
+        if length == 0:
+            skipped.append((chapter.title, length))
+            continue
+
+        keep = False
+        if threshold == 0:
+            keep = True
+        elif length >= threshold:
+            keep = True
+        elif not kept:
+            keep = True
+        elif min_short and length >= min_short and short_kept < _MAX_SHORT_CHAPTERS:
+            keep = True
+            short_kept += 1
+        elif looks_structural(chapter.title) and length >= _STRUCTURAL_MIN_LENGTH:
+            keep = True
+
+        if keep:
+            kept.append(chapter)
+        else:
+            skipped.append((chapter.title, length))
+
+    if kept:
+        return ChapterFilterResult(kept=kept, skipped=skipped)
+
+    longest_idx = None
+    longest_length = 0
+    for idx, chapter in enumerate(chapters):
+        stripped = chapter.text.strip()
+        if stripped and len(stripped) > longest_length:
+            longest_length = len(stripped)
+            longest_idx = idx
+
+    if longest_idx is not None:
+        longest = chapters[longest_idx]
+        fallback_skipped = [
+            (chapter.title, len(chapter.text.strip()))
+            for idx, chapter in enumerate(chapters)
+            if idx != longest_idx and chapter.text.strip()
+        ]
+        return ChapterFilterResult(kept=[longest], skipped=fallback_skipped)
+
+    return ChapterFilterResult(kept=[], skipped=skipped)
+
+
+def update_metadata_for_chapter_count(
+    metadata: Dict[str, Any], count: int, file_type: str
+) -> None:
+    if not metadata or count <= 0:
+        return
+
+    label = "Chapters" if file_type.lower() in {"epub", "markdown"} else "Pages"
+    metadata["chapter_count"] = str(count)
+
+    pattern = re.compile(r"\(\d+\s+(Chapters?|Pages?)\)")
+    replacement = f"({count} {label})"
+    for key in ("album", "ALBUM"):
+        value = metadata.get(key)
+        if not isinstance(value, str):
+            continue
+        metadata[key] = pattern.sub(replacement, value)

abogen/domain/metadata_helpers.py

@@ -0,0 +1,138 @@
+from __future__ import annotations
+
+import re
+from typing import Any, Dict, List, Mapping, Optional, Tuple
+
+
+_SERIES_NAME_KEYS = (
+    "series",
+    "series_name",
+    "series_title",
+)
+_SERIES_NUMBER_KEYS = (
+    "series_index",
+    "series_position",
+    "series_sequence",
+    "book_number",
+    "series_number",
+)
+_SERIES_NUMBER_RE = re.compile(r"\d+(?:\.\d+)?")
+
+
+def normalize_metadata_map(values: Optional[Mapping[str, Any]]) -> Dict[str, str]:
+    normalized: Dict[str, str] = {}
+    if not values:
+        return normalized
+    for key, value in values.items():
+        if value is None:
+            continue
+        text = str(value).strip()
+        if not text:
+            continue
+        normalized[str(key).casefold()] = text
+    return normalized
+
+
+def format_author_sentence(raw: Optional[str]) -> str:
+    if raw is None:
+        return ""
+    normalized = str(raw).strip()
+    if not normalized:
+        return ""
+    lowered = normalized.casefold()
+    if lowered in {"unknown", "various"}:
+        return ""
+
+    working = normalized.replace("&", " and ")
+    segments = [segment.strip() for segment in working.split(",") if segment.strip()]
+    tokens: List[str] = []
+
+    if segments:
+        for segment in segments:
+            parts = [part.strip() for part in re.split(r"\band\b", segment, flags=re.IGNORECASE) if part.strip()]
+            if parts:
+                tokens.extend(parts)
+            else:
+                tokens.append(segment)
+    else:
+        parts = [part.strip() for part in re.split(r"\band\b", working, flags=re.IGNORECASE) if part.strip()]
+        tokens.extend(parts or [normalized])
+
+    cleaned = [token for token in tokens if token and token.casefold() not in {"unknown", "various"}]
+    if not cleaned:
+        return ""
+    if len(cleaned) == 1:
+        return f"By {cleaned[0]}"
+    if len(cleaned) == 2:
+        return f"By {cleaned[0]} and {cleaned[1]}"
+    return f"By {', '.join(cleaned[:-1])}, and {cleaned[-1]}"
+
+
+def ensure_sentence(text: str) -> str:
+    cleaned = text.strip()
+    if not cleaned:
+        return ""
+    if cleaned[-1] in ".!?":
+        return cleaned
+    return f"{cleaned}."
+
+
+def normalize_series_number(value: Any) -> Optional[str]:
+    text = str(value or "").strip()
+    if not text:
+        return None
+    candidate = text.replace(",", ".")
+    if candidate.replace(".", "", 1).isdigit():
+        if "." in candidate:
+            normalized = candidate.rstrip("0").rstrip(".")
+            return normalized or "0"
+        try:
+            return str(int(candidate))
+        except ValueError:
+            pass
+    match = _SERIES_NUMBER_RE.search(candidate)
+    if not match:
+        return None
+    normalized = match.group(0)
+    if "." in normalized:
+        normalized = normalized.rstrip("0").rstrip(".")
+        return normalized or "0"
+    try:
+        return str(int(normalized))
+    except ValueError:
+        return normalized
+
+
+def extract_series_metadata(values: Mapping[str, str]) -> Tuple[Optional[str], Optional[str]]:
+    series_name: Optional[str] = None
+    for key in _SERIES_NAME_KEYS:
+        raw = values.get(key)
+        if raw:
+            cleaned = str(raw).strip()
+            if cleaned:
+                series_name = cleaned
+                break
+
+    series_number: Optional[str] = None
+    for key in _SERIES_NUMBER_KEYS:
+        raw = values.get(key)
+        if raw is None:
+            continue
+        normalized = normalize_series_number(raw)
+        if normalized:
+            series_number = normalized
+            break
+
+    return series_name, series_number
+
+
+def format_series_sentence(series_name: Optional[str], series_number: Optional[str]) -> str:
+    if not series_name or not series_number:
+        return ""
+    name = series_name.strip()
+    number = series_number.strip()
+    if not name or not number:
+        return ""
+    article = "the " if not name.lower().startswith("the ") else ""
+    phrase = f"Book {number} of {article}{name}"
+    return re.sub(r"\s+", " ", phrase).strip()

abogen/domain/metadata_merge.py

@@ -0,0 +1,23 @@
+from __future__ import annotations
+
+from typing import Any, Dict, Optional
+
+
+def merge_metadata(
+    extracted: Optional[Dict[str, Any]],
+    overrides: Optional[Dict[str, Any]],
+) -> Dict[str, str]:
+    merged: Dict[str, str] = {}
+    if extracted:
+        for key, value in extracted.items():
+            if value is None:
+                continue
+            merged[str(key)] = str(value)
+    if overrides:
+        for key, value in overrides.items():
+            key_str = str(key)
+            if value is None:
+                merged.pop(key_str, None)
+            else:
+                merged[key_str] = str(value)
+    return merged

abogen/domain/normalization.py

@@ -0,0 +1,30 @@
+"""Text normalization convenience helpers."""
+
+from __future__ import annotations
+
+from typing import Any, Mapping, Optional
+
+from abogen.kokoro_text_normalization import (
+    ApostropheConfig,
+    normalize_for_pipeline as _normalize_for_pipeline,
+)
+from abogen.normalization_settings import (
+    build_apostrophe_config,
+    get_runtime_settings,
+    apply_overrides as _apply_overrides,
+)
+
+_BASE_APOSTROPHE_CONFIG = ApostropheConfig()
+
+
+def normalize_text_for_pipeline(
+    text: str,
+    *,
+    normalization_overrides: Optional[Mapping[str, Any]] = None,
+) -> str:
+    """Normalize text using runtime settings with optional overrides."""
+    runtime_settings = get_runtime_settings()
+    if normalization_overrides:
+        runtime_settings = _apply_overrides(runtime_settings, normalization_overrides)
+    apostrophe_config = build_apostrophe_config(settings=runtime_settings, base=_BASE_APOSTROPHE_CONFIG)
+    return _normalize_for_pipeline(text, config=apostrophe_config, settings=runtime_settings)

abogen/domain/output_paths.py

@@ -0,0 +1,91 @@
+"""Output path resolution utilities.
+
+Pure functions for resolving output directories, building file paths,
+and computing project folder layouts.
+"""
+
+from __future__ import annotations
+
+import re
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Callable, List, Optional, Tuple
+
+from abogen.text_extractor import ExtractedChapter
+
+
+_OUTPUT_SANITIZE_RE = re.compile(r"[^\w\-_.]+")
+
+
+def slugify(title: str, index: int) -> str:
+    sanitized = re.sub(r"[^\w\-]+", "_", title.lower()).strip("_")
+    if not sanitized:
+        sanitized = f"chapter_{index:02d}"
+    return sanitized[:80]
+
+
+def sanitize_output_stem(name: str) -> str:
+    base = Path(name or "").stem
+    sanitized = _OUTPUT_SANITIZE_RE.sub("_", base).strip("_")
+    return sanitized or "output"
+
+
+def output_timestamp_token() -> str:
+    return datetime.now().strftime("%Y%m%d-%H%M%S")
+
+
+def build_output_path(directory: Path, original_name: str, extension: str) -> Path:
+    sanitized = sanitize_output_stem(original_name)
+    return directory / f"{sanitized}.{extension}"
+
+
+def apply_newline_policy(chapters: List[ExtractedChapter], replace_single_newlines: bool) -> None:
+    if not replace_single_newlines:
+        return
+    newline_regex = re.compile(r"(?<!\n)\n(?!\n)")
+    for chapter in chapters:
+        chapter.text = newline_regex.sub(" ", chapter.text)
+
+
+def resolve_output_directory(
+    *,
+    save_mode: str,
+    stored_path: Path,
+    output_folder: Optional[str],
+    desktop_dir: Optional[Path],
+    user_output_path: Optional[Path],
+    user_cache_outputs: Optional[Path],
+) -> Path:
+    if save_mode == "Save to Desktop" and desktop_dir:
+        return desktop_dir
+    if save_mode == "Save next to input file":
+        return stored_path.parent
+    if save_mode == "Choose output folder" and output_folder:
+        return Path(output_folder)
+    if save_mode == "Use default save location" and user_output_path:
+        return user_output_path
+    return user_cache_outputs or Path(".")
+
+
+def resolve_project_layout(
+    *,
+    original_filename: str,
+    save_as_project: bool,
+    base_dir: Path,
+    timestamp_fn: Callable[[], str] = output_timestamp_token,
+    sanitize_fn: Callable[[str, int], str] = sanitize_output_stem,
+) -> Tuple[Path, Path, Path, Optional[Path]]:
+    sanitized = sanitize_fn(original_filename, 0)
+    folder_name = f"{timestamp_fn()}_{sanitized}"
+    project_root = base_dir / folder_name
+    project_root.mkdir(parents=True, exist_ok=True)
+
+    if save_as_project:
+        audio_dir = project_root / "audio"
+        subtitle_dir = project_root / "subtitles"
+        metadata_dir = project_root / "metadata"
+        for directory in (audio_dir, subtitle_dir, metadata_dir):
+            directory.mkdir(parents=True, exist_ok=True)
+        return project_root, audio_dir, subtitle_dir, metadata_dir
+
+    return project_root, project_root, project_root, None

abogen/domain/pronunciation.py

@@ -0,0 +1,261 @@
+"""Pronunciation rule compilation and application.
+
+Pure functions for compiling token-level and sentence-level pronunciation
+overrides into regex patterns, applying them to text, and merging multiple
+override sources with precedence rules.
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Any, Dict, Iterable, List, Mapping, Optional
+
+from abogen.entity_analysis import normalize_token as normalize_entity_token
+from abogen.entity_analysis import normalize_manual_override_token
+
+
+def compile_pronunciation_rules(
+    overrides: Optional[Iterable[Mapping[str, Any]]],
+) -> List[Dict[str, Any]]:
+    if not overrides:
+        return []
+
+    candidates: List[Dict[str, Any]] = []
+    seen: set[str] = set()
+
+    for entry in overrides:
+        if not isinstance(entry, Mapping):
+            continue
+        pronunciation_value = str(entry.get("pronunciation") or "").strip()
+        if not pronunciation_value:
+            continue
+
+        token_values: List[str] = []
+        token_raw = entry.get("token")
+        if token_raw:
+            token_value = str(token_raw).strip()
+            if token_value:
+                token_values.append(token_value)
+        normalized_raw = entry.get("normalized")
+        if normalized_raw:
+            normalized_value = str(normalized_raw).strip()
+            if normalized_value:
+                token_values.append(normalized_value)
+        if token_raw and not token_values:
+            fallback = normalize_entity_token(str(token_raw))
+            if fallback:
+                token_values.append(fallback)
+
+        if not token_values:
+            continue
+
+        usage_normalized = str(entry.get("normalized") or "").strip()
+        if not usage_normalized and token_values:
+            usage_normalized = normalize_entity_token(token_values[0]) or token_values[0]
+        usage_token = str(entry.get("token") or token_values[0])
+
+        for token_value in token_values:
+            key = token_value.casefold()
+            if key in seen:
+                continue
+            seen.add(key)
+            candidates.append(
+                {
+                    "token": token_value,
+                    "normalized": usage_normalized,
+                    "replacement": pronunciation_value,
+                }
+            )
+
+    if not candidates:
+        return []
+
+    candidates.sort(key=lambda item: len(item["token"]), reverse=True)
+    compiled: List[Dict[str, Any]] = []
+    for candidate in candidates:
+        token_value = candidate["token"]
+        pronunciation_value = candidate["replacement"]
+        escaped = re.escape(token_value)
+        pattern = re.compile(rf"(?i)(?<!\w){escaped}(?P<possessive>'s|\u2019s|\u2019)?(?!\w)")
+        compiled.append(
+            {
+                "pattern": pattern,
+                "replacement": pronunciation_value,
+                "normalized": candidate.get("normalized") or token_value,
+                "token": candidate.get("token") or token_value,
+            }
+        )
+
+    return compiled
+
+
+def compile_heteronym_sentence_rules(
+    overrides: Optional[Iterable[Mapping[str, Any]]],
+) -> List[Dict[str, Any]]:
+    if not overrides:
+        return []
+
+    compiled: List[Dict[str, Any]] = []
+    seen: set[str] = set()
+
+    for entry in overrides:
+        if not isinstance(entry, Mapping):
+            continue
+        sentence = str(entry.get("sentence") or "").strip()
+        if not sentence:
+            continue
+        choice = str(entry.get("choice") or "").strip()
+        if not choice:
+            continue
+
+        replacement_sentence = ""
+        options = entry.get("options")
+        if isinstance(options, list):
+            for opt in options:
+                if not isinstance(opt, Mapping):
+                    continue
+                if str(opt.get("key") or "").strip() == choice:
+                    replacement_sentence = str(opt.get("replacement_sentence") or "").strip()
+                    break
+        if not replacement_sentence:
+            continue
+
+        rule_key = f"{sentence}\n{choice}".casefold()
+        if rule_key in seen:
+            continue
+        seen.add(rule_key)
+
+        parts = [p for p in re.split(r"\s+", sentence) if p]
+        if not parts:
+            continue
+        pattern_text = r"\s+".join(re.escape(p) for p in parts)
+        pattern = re.compile(pattern_text)
+        compiled.append({"pattern": pattern, "replacement": replacement_sentence})
+
+    compiled.sort(key=lambda item: len(item["pattern"].pattern), reverse=True)
+    return compiled
+
+
+def apply_heteronym_sentence_rules(text: str, rules: List[Dict[str, Any]]) -> str:
+    if not text or not rules:
+        return text
+    result = text
+    for rule in rules:
+        pattern = rule["pattern"]
+        replacement = rule["replacement"]
+        result = pattern.sub(replacement, result)
+    return result
+
+
+def apply_pronunciation_rules(
+    text: str,
+    rules: List[Dict[str, Any]],
+    usage_counter: Optional[Dict[str, int]] = None,
+) -> str:
+    if not text or not rules:
+        return text
+
+    result = text
+    for rule in rules:
+        pattern = rule["pattern"]
+        pronunciation_value = rule["replacement"]
+        usage_key = str(rule.get("normalized") or "").strip()
+
+        def _replacement(match: re.Match[str]) -> str:
+            suffix = match.group("possessive") or ""
+            if usage_counter is not None and usage_key:
+                usage_counter[usage_key] = usage_counter.get(usage_key, 0) + 1
+            return pronunciation_value + suffix
+
+        result = pattern.sub(_replacement, result)
+
+    return result
+
+
+def merge_pronunciation_overrides(job: Any) -> List[Dict[str, Any]]:
+    """Return pronunciation override entries, ensuring manual overrides are included.
+
+    Pending jobs keep both ``manual_overrides`` and ``pronunciation_overrides``, but the
+    latter can be stale if the UI didn't resync before enqueue. During conversion,
+    we must merge manual overrides so they always apply (before TTS).
+
+    Precedence: manual overrides win over existing entries for the same normalized key.
+    """
+
+    collected: Dict[str, Dict[str, Any]] = {}
+
+    existing = getattr(job, "pronunciation_overrides", None)
+    if isinstance(existing, list):
+        for entry in existing:
+            if not isinstance(entry, Mapping):
+                continue
+            token_value = str(entry.get("token") or "").strip()
+            pronunciation_value = str(entry.get("pronunciation") or "").strip()
+            if not token_value or not pronunciation_value:
+                continue
+            normalized = str(entry.get("normalized") or "").strip() or normalize_entity_token(token_value)
+            if not normalized:
+                continue
+            collected[normalized] = {
+                "token": token_value,
+                "normalized": normalized,
+                "pronunciation": pronunciation_value,
+                "voice": str(entry.get("voice") or "").strip() or None,
+                "notes": str(entry.get("notes") or "").strip() or None,
+                "context": str(entry.get("context") or "").strip() or None,
+                "source": str(entry.get("source") or "pronunciation"),
+                "language": getattr(job, "language", None),
+            }
+
+    speakers = getattr(job, "speakers", None)
+    if isinstance(speakers, dict):
+        for payload in speakers.values():
+            if not isinstance(payload, Mapping):
+                continue
+            token_value = str(payload.get("token") or "").strip()
+            pronunciation_value = str(payload.get("pronunciation") or "").strip()
+            if not token_value or not pronunciation_value:
+                continue
+            normalized = normalize_entity_token(token_value)
+            if not normalized:
+                continue
+            collected[normalized] = {
+                "token": token_value,
+                "normalized": normalized,
+                "pronunciation": pronunciation_value,
+                "voice": str(
+                    payload.get("resolved_voice")
+                    or payload.get("voice")
+                    or getattr(job, "voice", "")
+                ).strip()
+                or None,
+                "notes": None,
+                "context": None,
+                "source": "speaker",
+                "language": getattr(job, "language", None),
+            }
+
+    manual = getattr(job, "manual_overrides", None)
+    if isinstance(manual, list):
+        for entry in manual:
+            if not isinstance(entry, Mapping):
+                continue
+            token_value = str(entry.get("token") or "").strip()
+            pronunciation_value = str(entry.get("pronunciation") or "").strip()
+            if not token_value or not pronunciation_value:
+                continue
+            normalized = str(entry.get("normalized") or "").strip() or normalize_manual_override_token(token_value)
+            if not normalized:
+                continue
+            collected[normalized] = {
+                "token": token_value,
+                "normalized": normalized,
+                "pronunciation": pronunciation_value,
+                "voice": str(entry.get("voice") or "").strip() or None,
+                "notes": str(entry.get("notes") or "").strip() or None,
+                "context": str(entry.get("context") or "").strip() or None,
+                "source": str(entry.get("source") or "manual"),
+                "language": getattr(job, "language", None),
+            }
+
+    return list(collected.values())

abogen/domain/split_pattern.py

@@ -0,0 +1,40 @@
+from __future__ import annotations
+
+"""Unified split pattern logic extracted from 3 copies."""
+import re
+
+
+PUNCTUATION_SENTENCE = r".!?。！？"
+PUNCTUATION_SENTENCE_COMMA = r".!?,。！？、，"
+
+
+def get_split_pattern(language: str, subtitle_mode: str) -> str:
+    """Get the appropriate split pattern based on language and subtitle mode.
+
+    Args:
+        language: Language code (a, b, e, f, etc.)
+        subtitle_mode: Subtitle mode ("Sentence", "Sentence + Comma", "Line", etc.)
+
+    Returns:
+        Split pattern string
+    """
+    # For English, always use newline splitting only
+    if language in ("a", "b"):
+        return "\n"
+
+    # Determine spacing pattern based on language
+    spacing = r"\s*" if language in ("z", "j") else r"\s+"
+
+    # For CJK languages, when subtitle mode is Disabled or Line, prefer
+    # punctuation-based splitting instead of plain newline splitting.
+    if subtitle_mode in ("Disabled", "Line") and language in ("z", "j"):
+        return rf"(?<=[{PUNCTUATION_SENTENCE}]){spacing}|\n+"
+
+    if subtitle_mode == "Line":
+        return "\n"
+    elif subtitle_mode == "Sentence":
+        return rf"(?<=[{PUNCTUATION_SENTENCE}]){spacing}|\n+"
+    elif subtitle_mode == "Sentence + Comma":
+        return rf"(?<=[{PUNCTUATION_SENTENCE_COMMA}]){spacing}|\n+"
+    else:
+        return r"\n+"

abogen/domain/title_builder.py

@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Any, Dict, List, Mapping, Optional
+
+from .metadata_helpers import (
+    ensure_sentence,
+    extract_series_metadata,
+    format_author_sentence,
+    format_series_sentence,
+    normalize_metadata_map,
+)
+
+
+def build_title_intro_text(
+    metadata: Optional[Mapping[str, Any]],
+    fallback_basename: str,
+) -> str:
+    """Build the title introduction text from metadata."""
+    normalized = normalize_metadata_map(metadata)
+    fallback_title = Path(fallback_basename).stem if fallback_basename else ""
+    title = (
+        normalized.get("title")
+        or normalized.get("book_title")
+        or normalized.get("album")
+        or fallback_title
+    )
+    if not title:
+        title = fallback_title
+    subtitle = normalized.get("subtitle") or normalized.get("sub_title")
+    if subtitle and title and subtitle.casefold() == title.casefold():
+        subtitle = ""
+
+    author_value = ""
+    for candidate in ("artist", "album_artist", "author", "authors", "writer", "composer"):
+        value = normalized.get(candidate)
+        if value:
+            author_value = value
+            break
+
+    series_name, series_number = extract_series_metadata(normalized)
+    series_sentence = format_series_sentence(series_name, series_number)
+
+    sentences: List[str] = []
+    if series_sentence:
+        sentences.append(ensure_sentence(series_sentence))
+    if title:
+        sentences.append(ensure_sentence(title))
+    if subtitle:
+        sentences.append(ensure_sentence(subtitle))
+    author_sentence = format_author_sentence(author_value)
+    if author_sentence:
+        sentences.append(ensure_sentence(author_sentence))
+    return " ".join(sentences).strip()
+
+
+def build_outro_text(
+    metadata: Optional[Mapping[str, Any]],
+    fallback_basename: str,
+) -> str:
+    """Build the outro/closing text from metadata."""
+    normalized = normalize_metadata_map(metadata)
+    fallback_title = Path(fallback_basename).stem if fallback_basename else ""
+    title = (
+        normalized.get("title")
+        or normalized.get("book_title")
+        or normalized.get("album")
+        or fallback_title
+    )
+    author_value = ""
+    for candidate in ("authors", "author", "album_artist", "artist", "writer", "composer"):
+        value = normalized.get(candidate)
+        if value:
+            author_value = value
+            break
+    author_sentence = format_author_sentence(author_value)
+    authors_fragment = (
+        author_sentence[3:].strip() if author_sentence.lower().startswith("by ") else author_sentence.strip()
+    )
+
+    if title and authors_fragment:
+        closing_line = f"The end of {title} from {authors_fragment}"
+    elif title:
+        closing_line = f"The end of {title}"
+    elif authors_fragment:
+        closing_line = f"The end from {authors_fragment}"
+    else:
+        closing_line = "The end"
+
+    series_name, series_number = extract_series_metadata(normalized)
+    series_sentence = format_series_sentence(series_name, series_number)
+
+    sentences: List[str] = [ensure_sentence(closing_line)]
+    if series_sentence:
+        sentences.append(ensure_sentence(series_sentence))
+
+    return " ".join(sentence for sentence in sentences if sentence).strip()

abogen/domain/voice_resolution.py

@@ -0,0 +1,190 @@
+"""Voice resolution helpers.
+
+Functions for resolving voice specifications, collecting required voice IDs,
+and determining the voice to use for chapters and chunks.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Dict, Optional, Set
+
+from abogen.tts_plugin.utils import get_voices, get_default_voice
+from abogen.voice_formulas import extract_voice_ids
+from abogen.voice_cache import ensure_voice_assets
+
+
+def spec_to_voice_ids(spec: Any) -> Set[str]:
+    text = str(spec or "").strip()
+    if not text:
+        return set()
+    if text == "__custom_mix":
+        return set()
+    if "*" in text:
+        try:
+            return set(extract_voice_ids(text))
+        except ValueError:
+            return set()
+    if text in get_voices("kokoro"):
+        return {text}
+    return set()
+
+
+def job_voice_fallback(job: Any) -> str:
+    base = str(getattr(job, "voice", "") or "").strip()
+    if base and base != "__custom_mix":
+        return base
+
+    speakers = getattr(job, "speakers", None)
+    if isinstance(speakers, dict):
+        narrator = speakers.get("narrator")
+        if isinstance(narrator, dict):
+            for key in ("resolved_voice", "voice_formula", "voice"):
+                value = narrator.get(key)
+                candidate = str(value or "").strip()
+                if candidate and candidate != "__custom_mix":
+                    return candidate
+        for payload in speakers.values() or []:
+            if not isinstance(payload, dict):
+                continue
+            for key in ("resolved_voice", "voice_formula", "voice"):
+                value = payload.get(key)
+                candidate = str(value or "").strip()
+                if candidate and candidate != "__custom_mix":
+                    return candidate
+
+    for chapter in getattr(job, "chapters", []) or []:
+        if not isinstance(chapter, dict):
+            continue
+        for key in ("resolved_voice", "voice_formula", "voice"):
+            candidate = str(chapter.get(key) or "").strip()
+            if candidate and candidate != "__custom_mix":
+                return candidate
+
+    return ""
+
+
+def collect_required_voice_ids(job: Any) -> Set[str]:
+    voices: Set[str] = set()
+    voices.update(spec_to_voice_ids(job.voice))
+    voices.update(spec_to_voice_ids(job_voice_fallback(job)))
+
+    for chapter in getattr(job, "chapters", []) or []:
+        if not isinstance(chapter, dict):
+            continue
+        for key in ("resolved_voice", "voice_formula", "voice"):
+            voices.update(spec_to_voice_ids(chapter.get(key)))
+
+    for chunk in getattr(job, "chunks", []) or []:
+        if not isinstance(chunk, dict):
+            continue
+        for key in ("resolved_voice", "voice_formula", "voice"):
+            voices.update(spec_to_voice_ids(chunk.get(key)))
+
+    speakers = getattr(job, "speakers", {})
+    if isinstance(speakers, dict):
+        for payload in speakers.values() or []:
+            if not isinstance(payload, dict):
+                continue
+            for key in ("resolved_voice", "voice_formula", "voice"):
+                voices.update(spec_to_voice_ids(payload.get(key)))
+
+    voices.update(get_voices("kokoro"))
+    return voices
+
+
+def initialize_voice_cache(job: Any) -> None:
+    try:
+        targets = collect_required_voice_ids(job)
+        downloaded, errors = ensure_voice_assets(
+            targets,
+            on_progress=lambda message: job.add_log(message, level="debug"),
+        )
+    except RuntimeError as exc:
+        job.add_log(f"Voice cache unavailable: {exc}", level="warning")
+        return
+
+    if downloaded:
+        job.add_log(
+            f"Cached {len(downloaded)} voice asset{'s' if len(downloaded) != 1 else ''} locally.",
+            level="info",
+        )
+
+    for voice_id, error in errors.items():
+        job.add_log(f"Failed to cache voice '{voice_id}': {error}", level="warning")
+
+
+def chapter_voice_spec(job: Any, override: Optional[Dict[str, Any]]) -> str:
+    if not override:
+        return job_voice_fallback(job)
+
+    resolved = str(override.get("resolved_voice", "")).strip()
+    if resolved:
+        return resolved
+
+    formula = str(override.get("voice_formula", "")).strip()
+    if formula:
+        return formula
+
+    voice = str(override.get("voice", "")).strip()
+    if voice:
+        return voice
+
+    return job_voice_fallback(job)
+
+
+def chunk_voice_spec(job: Any, chunk: Dict[str, Any], fallback: str) -> str:
+    for key in ("resolved_voice", "voice_formula", "voice"):
+        value = chunk.get(key)
+        if value:
+            return str(value)
+
+    speaker_id = chunk.get("speaker_id")
+    speakers = getattr(job, "speakers", None)
+    if isinstance(speakers, dict) and speaker_id in speakers:
+        speaker_entry = speakers.get(speaker_id) or {}
+        if isinstance(speaker_entry, dict):
+            for key in ("resolved_voice", "voice_formula", "voice"):
+                value = speaker_entry.get(key)
+                if value:
+                    return str(value)
+            profile_formula = speaker_entry.get("voice_formula")
+            if profile_formula:
+                return str(profile_formula)
+
+    profile_name = chunk.get("voice_profile")
+    if profile_name:
+        if isinstance(speakers, dict):
+            speaker_entry = speakers.get(profile_name)
+            if isinstance(speaker_entry, dict):
+                for key in ("resolved_voice", "voice_formula", "voice"):
+                    value = speaker_entry.get(key)
+                    if value:
+                        return str(value)
+
+    if fallback:
+        return fallback
+    return job_voice_fallback(job)
+
+
+def resolve_fallback_voice_spec(
+    base_spec: str,
+    job_voice: str,
+    voice_cache_keys: list[str],
+    provider: str = "kokoro",
+) -> str:
+    """Resolve the voice spec for intro/outro with a priority fallback chain.
+
+    Priority: base_spec → job_voice → first voice_cache key → default voice.
+    ``"__custom_mix"`` is treated as empty (it is not a usable voice spec).
+    """
+    spec = base_spec or job_voice
+    if spec == "__custom_mix":
+        spec = job_voice or ""
+    if not spec:
+        for key in voice_cache_keys:
+            if key and key != "__custom_mix":
+                spec = key.split(":", 1)[-1]
+                break
+    if not spec:
+        spec = get_default_voice(provider)
+    return spec

abogen/domain/voice_utils.py

@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+from typing import Any, Mapping, Optional, Tuple, Set
+
+from abogen.voice_formulas import extract_voice_ids, get_new_voice
+from abogen.tts_plugin.utils import get_voices
+
+
+def infer_provider_from_spec(value: Any, fallback: str = "kokoro") -> str:
+    """Infer TTS provider from voice specification."""
+    raw = str(value or "").strip()
+    if not raw:
+        return fallback
+    if raw.upper() == raw and raw.replace("_", "").isalnum():
+        return "supertonic"
+    if raw == "__custom_mix" or "*" in raw or "+" in raw:
+        return "kokoro"
+    if raw in get_voices("kokoro"):
+        return "kokoro"
+    return fallback
+
+
+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"
+
+    return upper
+
+
+def split_speaker_reference(value: Any) -> Tuple[Optional[str], str]:
+    """Parse speaker/profile reference from string.
+
+    Expected format: "speaker:name" or "profile:name"
+    Returns (name, original) or (None, original) if not a valid reference.
+    """
+    raw = str(value or "").strip()
+    if not raw or ":" not in raw:
+        return None, raw
+    prefix, remainder = raw.split(":", 1)
+    prefix = prefix.strip().lower()
+    if prefix not in {"speaker", "profile"}:
+        return None, raw
+    name = remainder.strip()
+    return (name or None), raw
+
+
+def formula_from_kokoro_entry(entry: Mapping[str, Any]) -> str:
+    """Build voice formula string from kokoro entry."""
+    voices = entry.get("voices") or []
+    if not voices:
+        return ""
+    total = 0.0
+    parts: list[tuple[str, float]] = []
+    for item in voices:
+        if not isinstance(item, (list, tuple)) or len(item) < 2:
+            continue
+        name = str(item[0] or "").strip()
+        try:
+            weight = float(item[1])
+        except (TypeError, ValueError):
+            continue
+        if name and weight > 0:
+            parts.append((name, weight))
+            total += weight
+
+    if not parts:
+        return ""
+
+    normalized = [(name, weight / total) for name, weight in parts]
+    return " + ".join(f"{name}*{weight:.6f}" for name, weight in normalized)
+
+
+def coerce_truthy(value: Any, default: bool = True) -> bool:
+    """Coerce a value to boolean with default."""
+    if isinstance(value, bool):
+        return value
+    if isinstance(value, str):
+        return value.lower() not in {"false", "0", "no", "off", ""}
+    if value is None:
+        return default
+    return bool(value)

abogen/infrastructure/exporters.py

@@ -0,0 +1,673 @@
+from __future__ import annotations
+
+import json
+import logging
+import tempfile
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Mapping, Sequence
+
+import static_ffmpeg
+
+from abogen.epub3.exporter import build_epub3_package
+from abogen.integrations.audiobookshelf import (
+    AudiobookshelfClient,
+    AudiobookshelfConfig,
+    AudiobookshelfUploadError,
+)
+from abogen.utils import create_process
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class ExportConfig:
+    """Configuration for export operations."""
+    ffmpeg_path: str = "ffmpeg"
+    verify_ssl: bool = True
+
+
+class ExportService:
+    """Unified service for audiobook exports (M4B, FFMETADATA, EPUB3, Audiobookshelf)."""
+    
+    def __init__(self, config: Optional[ExportConfig] = None):
+        self.config = config or ExportConfig()
+        static_ffmpeg.add_paths()
+    
+    # ----------------------------------------------------------------------
+    # FFMETADATA
+    # ----------------------------------------------------------------------
+    
+    def render_ffmetadata(
+        self,
+        metadata: Dict[str, Any],
+        chapters: List[Dict[str, Any]],
+    ) -> str:
+        """Render FFMETADATA content."""
+        lines = [";FFMETADATA1"]
+        
+        for key, value in (metadata or {}).items():
+            if value is None:
+                continue
+            key_str = str(key).strip()
+            if not key_str:
+                continue
+            lines.append(f"{key_str}={self._escape_ffmetadata_value(value)}")
+        
+        for chapter in chapters or []:
+            start = chapter.get("start")
+            end = chapter.get("end")
+            if start is None or end is None:
+                continue
+            try:
+                start_ms = max(0, int(round(float(start) * 1000)))
+                end_ms = int(round(float(end) * 1000))
+            except (TypeError, ValueError):
+                continue
+            if end_ms <= start_ms:
+                end_ms = start_ms + 1
+            lines.append("[CHAPTER]")
+            lines.append("TIMEBASE=1/1000")
+            lines.append(f"START={start_ms}")
+            lines.append(f"END={end_ms}")
+            title = chapter.get("title")
+            if title:
+                lines.append(f"title={self._escape_ffmetadata_value(title)}")
+            voice = chapter.get("voice")
+            if voice:
+                lines.append(f"voice={self._escape_ffmetadata_value(voice)}")
+        
+        return "\n".join(lines) + "\n"
+    
+    @staticmethod
+    def _escape_ffmetadata_value(value: Any) -> str:
+        escaped = str(value).replace("\\", "\\\\").replace("\n", "\\n")
+        escaped = escaped.replace("=", "\\=").replace(";", "\\;").replace("#", "\\#")
+        return escaped
+    
+    def write_ffmetadata_file(
+        self,
+        audio_path: Path,
+        metadata: Dict[str, Any],
+        chapters: List[Dict[str, Any]],
+    ) -> Optional[Path]:
+        """Write FFMETADATA file to temp location."""
+        content = self.render_ffmetadata(metadata, chapters)
+        if content.strip() == ";FFMETADATA1":
+            return None
+        
+        directory = audio_path.parent if audio_path.parent.exists() else Path(tempfile.gettempdir())
+        with tempfile.NamedTemporaryFile(
+            mode="w",
+            encoding="utf-8",
+            suffix=".ffmeta",
+            delete=False,
+            dir=str(directory),
+        ) as handle:
+            handle.write(content)
+            return Path(handle.name)
+    
+    # ----------------------------------------------------------------------
+    # M4B Export
+    # ----------------------------------------------------------------------
+    
+    def embed_m4b_metadata(
+        self,
+        audio_path: Path,
+        metadata: Dict[str, Any],
+        chapters: List[Dict[str, Any]],
+        cover_path: Optional[Path] = None,
+        cover_mime: Optional[str] = None,
+        log_callback: Optional[callable] = None,
+    ) -> None:
+        """Embed metadata and chapters into M4B file using FFmpeg + Mutagen."""
+        ffmetadata_path = self.write_ffmetadata_file(audio_path, metadata, chapters)
+        
+        metadata_args = self._metadata_to_ffmpeg_args(metadata)
+        
+        cmd = ["ffmpeg", "-y", "-i", str(audio_path)]
+        
+        if ffmetadata_path:
+            cmd.extend(["-f", "ffmetadata", "-i", str(ffmetadata_path)])
+        
+        if cover_path and cover_path.exists():
+            cmd.extend(["-i", str(cover_path)])
+            cmd.extend(["-map", "0:a"])
+            cmd.extend(["-map", "1:v:0", "-c:v:0", "mjpeg", "-disposition:v:0", "attached_pic"])
+            if cover_mime:
+                cmd.extend(["-metadata:s:v:0", f"mimetype={cover_mime}"])
+            cmd.extend(["-metadata:s:v:0", "title=Cover Art"])
+        else:
+            cmd.extend(["-map", "0:a"])
+        
+        cmd.extend(["-c:a", "copy"])
+        
+        if ffmetadata_path:
+            cmd.extend(["-map_metadata", "1", "-map_chapters", "1"])
+        else:
+            cmd.extend(["-map_metadata", "0"])
+        
+        if metadata_args:
+            cmd.extend(metadata_args)
+        
+        cmd.extend(["-movflags", "+faststart+use_metadata_tags"])
+        
+        temp_output = audio_path.with_suffix(audio_path.suffix + ".tmp")
+        if audio_path.suffix.lower() in {".m4b", ".mp4", ".m4a"}:
+            cmd.extend(["-f", "mp4"])
+        cmd.append(str(temp_output))
+        
+        if log_callback:
+            log_callback("Embedding metadata into M4B output")
+        
+        process = create_process(cmd, text=True)
+        return_code = process.wait()
+        
+        if ffmetadata_path and ffmetadata_path.exists():
+            try:
+                ffmetadata_path.unlink()
+            except OSError:
+                pass
+        
+        if return_code != 0:
+            if temp_output.exists():
+                temp_output.unlink(missing_ok=True)
+            raise RuntimeError(f"ffmpeg failed to embed metadata (exit code {return_code})")
+        
+        temp_output.replace(audio_path)
+        
+        if log_callback:
+            log_callback("Embedded metadata and chapters into M4B output", "info")
+        
+        # Apply chapters via Mutagen for better compatibility
+        self._apply_m4b_chapters_mutagen(audio_path, chapters, log_callback)
+    
+    @staticmethod
+    def _metadata_to_ffmpeg_args(metadata: Dict[str, Any]) -> List[str]:
+        args = []
+        for key, value in (metadata or {}).items():
+            if value in (None, ""):
+                continue
+            key_str = str(key).strip()
+            if not key_str:
+                continue
+            normalized_key = key_str.lower()
+            if normalized_key == "year":
+                ffmpeg_key = "date"
+            else:
+                ffmpeg_key = key_str
+            args.extend(["-metadata", f"{ffmpeg_key}={value}"])
+        return args
+    
+    def _apply_m4b_chapters_mutagen(
+        self,
+        audio_path: Path,
+        chapters: List[Dict[str, Any]],
+        log_callback: Optional[callable] = None,
+    ) -> bool:
+        """Apply chapter atoms using Mutagen."""
+        if not chapters:
+            return False
+        
+        try:
+            from fractions import Fraction
+            from mutagen.mp4 import MP4, MP4Chapter
+        except ImportError:
+            if log_callback:
+                log_callback("Unable to write MP4 chapter atoms because mutagen is not installed.", "warning")
+            return False
+        
+        try:
+            mp4 = MP4(str(audio_path))
+        except Exception as exc:
+            if log_callback:
+                log_callback(f"Failed to open m4b for chapter embedding: {exc}", "warning")
+            return False
+        
+        chapter_objects = []
+        for index, entry in enumerate(sorted(chapters, key=lambda item: float(item.get("start") or 0.0))):
+            start_raw = entry.get("start")
+            if start_raw is None:
+                continue
+            try:
+                start_seconds = max(0.0, float(start_raw))
+            except (TypeError, ValueError):
+                continue
+            
+            title_value = entry.get("title")
+            title_text = str(title_value) if title_value else f"Chapter {index + 1}"
+            
+            start_fraction = Fraction(int(round(start_seconds * 1000)), 1000)
+            chapter_atom = MP4Chapter(start_fraction, title_text)
+            
+            end_raw = entry.get("end")
+            if end_raw is not None:
+                try:
+                    end_seconds = float(end_raw)
+                except (TypeError, ValueError):
+                    end_seconds = None
+                if end_seconds is not None and end_seconds > start_seconds:
+                    chapter_atom.end = Fraction(int(round(end_seconds * 1000)), 1000)
+            
+            chapter_objects.append(chapter_atom)
+        
+        if not chapter_objects:
+            return False
+        
+        try:
+            mp4.chapters = chapter_objects
+            mp4.save()
+        except Exception as exc:
+            if log_callback:
+                log_callback(f"Failed to persist MP4 chapter atoms: {exc}", "warning")
+            return False
+        
+        if log_callback:
+            log_callback(f"Applied {len(chapter_objects)} chapter markers via mutagen", "info")
+        return True
+    
+    # ----------------------------------------------------------------------
+    # EPUB3 Export
+    # ----------------------------------------------------------------------
+    
+    def export_epub3(
+        self,
+        output_path: Path,
+        book_id: str,
+        extraction: Any,  # ExtractionResult
+        metadata_tags: Dict[str, Any],
+        chapter_markers: Sequence[Dict[str, Any]],
+        chunk_markers: Sequence[Dict[str, Any]],
+        chunks: Iterable[Dict[str, Any]],
+        audio_path: Path,
+        speaker_mode: str = "single",
+        cover_path: Optional[Path] = None,
+        cover_mime: Optional[str] = None,
+    ) -> Path:
+        """Export EPUB3 with media overlays."""
+        return build_epub3_package(
+            output_path=output_path,
+            book_id=book_id,
+            extraction=extraction,
+            metadata_tags=metadata_tags,
+            chapter_markers=chapter_markers,
+            chunk_markers=chunk_markers,
+            chunks=chunks,
+            audio_path=audio_path,
+            speaker_mode=speaker_mode,
+            cover_image_path=cover_path,
+            cover_image_mime=cover_mime,
+        )
+    
+    # ----------------------------------------------------------------------
+    # Audiobookshelf Integration
+    # ----------------------------------------------------------------------
+    
+    def build_audiobookshelf_metadata(self, job: Any) -> Dict[str, Any]:
+        """Build Audiobookshelf metadata from job."""
+        tags = self._normalize_metadata_casefold(getattr(job, "metadata_tags", {}))
+        filename = Path(getattr(job, "original_filename", "") or "").stem or "Audiobook"
+        
+        title = self._first_nonempty(
+            tags.get("title"),
+            tags.get("book_title"),
+            tags.get("name"),
+            tags.get("album"),
+            filename,
+        )
+        authors = self._split_people_field(
+            tags.get("authors")
+            or tags.get("author")
+            or tags.get("album_artist")
+            or tags.get("artist")
+        )
+        narrators = self._split_people_field(tags.get("narrators") or tags.get("narrator"))
+        description = self._first_nonempty(
+            tags.get("description"), tags.get("summary"), tags.get("comment")
+        )
+        genres = self._split_simple_list(tags.get("genre"))
+        keywords = self._split_simple_list(tags.get("tags") or tags.get("keywords"))
+        language = self._first_nonempty(tags.get("language"), tags.get("lang")) or getattr(job, "language", "") or ""
+        series_name = self._first_nonempty(
+            tags.get("series"),
+            tags.get("series_name"),
+            tags.get("seriesname"),
+            tags.get("series_title"),
+            tags.get("seriestitle"),
+        )
+        
+        series_sequence = None
+        for key in ("series_index", "series_position", "series_sequence", "series_number", "seriesnumber", "book_number", "booknumber"):
+            raw = tags.get(key)
+            normalized = self._normalize_series_sequence(raw)
+            if normalized:
+                series_sequence = normalized
+                break
+        if not series_name:
+            series_sequence = None
+        
+        data = {
+            "title": title,
+            "subtitle": tags.get("subtitle"),
+            "authors": authors,
+            "narrators": narrators,
+            "description": description,
+            "publisher": tags.get("publisher"),
+            "genres": genres,
+            "tags": keywords,
+            "language": language,
+            "publishedYear": self._extract_year(
+                tags.get("published") or tags.get("publication_year") or tags.get("date") or tags.get("year")
+            ),
+            "seriesName": series_name,
+            "seriesSequence": series_sequence,
+            "isbn": self._first_nonempty(tags.get("isbn"), tags.get("asin")),
+        }
+        
+        published_date = self._first_nonempty(
+            tags.get("published"), tags.get("publication_date"), tags.get("date")
+        )
+        if published_date:
+            data["publishedDate"] = published_date
+        
+        rating_text = self._first_nonempty(tags.get("rating"), tags.get("my_rating"))
+        if rating_text:
+            try:
+                data["rating"] = float(str(rating_text).strip())
+            except ValueError:
+                pass
+            rating_max_text = self._first_nonempty(tags.get("rating_max"), tags.get("rating_scale"))
+            if rating_max_text:
+                try:
+                    data["ratingMax"] = float(str(rating_max_text).strip())
+                except ValueError:
+                    pass
+        
+        # Remove empty values
+        cleaned = {}
+        for key, value in data.items():
+            if value is None:
+                continue
+            if isinstance(value, str) and not value.strip():
+                continue
+            if isinstance(value, (list, tuple)) and not value:
+                continue
+            cleaned[key] = value
+        return cleaned
+    
+    def load_audiobookshelf_chapters(self, job: Any) -> Optional[List[Dict[str, Any]]]:
+        """Load chapters from job artifacts for Audiobookshelf."""
+        metadata_ref = job.result.artifacts.get("metadata") if getattr(job, "result", None) else None
+        if not metadata_ref:
+            return None
+        metadata_path = metadata_ref if isinstance(metadata_ref, Path) else Path(str(metadata_ref))
+        if not metadata_path.exists():
+            return None
+        try:
+            payload = json.loads(metadata_path.read_text(encoding="utf-8"))
+        except (OSError, json.JSONDecodeError):
+            return None
+        chapters = payload.get("chapters")
+        if not isinstance(chapters, list):
+            return None
+        cleaned = []
+        for entry in chapters:
+            if not isinstance(entry, Mapping):
+                continue
+            title = self._first_nonempty(entry.get("title"), entry.get("original_title"))
+            start = entry.get("start")
+            end = entry.get("end")
+            if title is None or not isinstance(start, (int, float)):
+                continue
+            chapter_payload = {"title": title, "start": float(start)}
+            if isinstance(end, (int, float)):
+                chapter_payload["end"] = float(end)
+            cleaned.append(chapter_payload)
+        return cleaned or None
+    
+    def upload_audiobookshelf(
+        self,
+        job: Any,
+        audio_path: Path,
+        subtitle_paths: List[Path],
+        chapters: List[Dict[str, Any]],
+        metadata: Dict[str, Any],
+        cover_path: Optional[Path] = None,
+        config: Optional[AudiobookshelfConfig] = None,
+        log_callback: Optional[callable] = None,
+    ) -> None:
+        """Upload to Audiobookshelf."""
+        if config is None:
+            # Load from job or global config
+            cfg = getattr(job, "_abs_config", None)
+            if cfg is None:
+                from abogen.utils import load_config
+                global_cfg = load_config() or {}
+                abs_cfg = global_cfg.get("audiobookshelf")
+                if isinstance(abs_cfg, Mapping):
+                    config = AudiobookshelfConfig(
+                        base_url=str(abs_cfg.get("base_url") or "").strip(),
+                        api_token=str(abs_cfg.get("api_token") or "").strip(),
+                        library_id=str(abs_cfg.get("library_id") or "").strip(),
+                        collection_id=(str(abs_cfg.get("collection_id") or "").strip() or None),
+                        folder_id=str(abs_cfg.get("folder_id") or "").strip(),
+                        verify_ssl=self._coerce_bool(abs_cfg.get("verify_ssl"), True),
+                        send_cover=self._coerce_bool(abs_cfg.get("send_cover"), True),
+                        send_chapters=self._coerce_bool(abs_cfg.get("send_chapters"), True),
+                        send_subtitles=self._coerce_bool(abs_cfg.get("send_subtitles"), False),
+                        timeout=float(abs_cfg.get("timeout", 3600.0)),
+                    )
+                else:
+                    if log_callback:
+                        log_callback("Audiobookshelf upload skipped: not configured", "warning")
+                    return
+        
+        if not config.base_url or not config.api_token or not config.library_id:
+            if log_callback:
+                log_callback("Audiobookshelf upload skipped: configure base URL, API token, and library ID first", "warning")
+            return
+        if not config.folder_id:
+            if log_callback:
+                log_callback("Audiobookshelf upload skipped: enter folder name or ID in settings", "warning")
+            return
+        
+        if not audio_path.exists():
+            if log_callback:
+                log_callback("Audiobookshelf upload skipped: audio output not found", "warning")
+            return
+        
+        existing_subtitles = [p for p in subtitle_paths if p.exists()] if config.send_subtitles else None
+        chapters_to_send = chapters if config.send_chapters else None
+        
+        client = AudiobookshelfClient(config)
+        
+        display_title = metadata.get("title") or audio_path.stem
+        try:
+            existing_items = client.find_existing_items(display_title, folder_id=config.folder_id)
+        except AudiobookshelfUploadError as exc:
+            if log_callback:
+                log_callback(f"Audiobookshelf lookup failed: {exc}", "error")
+            return
+        
+        if existing_items:
+            if log_callback:
+                log_callback(f"Removing existing Audiobookshelf item(s) for '{display_title}' before upload.", "info")
+            try:
+                client.delete_items(existing_items)
+            except Exception as exc:
+                if log_callback:
+                    log_callback(f"Failed to remove existing item(s): {exc}", "warning")
+        
+        cover_to_send = cover_path
+        if config.send_cover and cover_to_send:
+            if isinstance(cover_to_send, str):
+                cover_to_send = Path(cover_to_send)
+            if not cover_to_send.exists():
+                cover_to_send = None
+        
+        client.upload_audiobook(
+            audio_path,
+            metadata=metadata,
+            cover_path=cover_to_send,
+            chapters=chapters_to_send,
+            subtitles=existing_subtitles,
+        )
+        
+        if log_callback:
+            log_callback("Audiobookshelf upload queued.", "info")
+    
+    # ----------------------------------------------------------------------
+    # Helpers
+    # ----------------------------------------------------------------------
+    
+    @staticmethod
+    def _normalize_metadata_casefold(values: Optional[Mapping[str, Any]]) -> Dict[str, Any]:
+        normalized = {}
+        if not values:
+            return normalized
+        for key, value in values.items():
+            if value is None:
+                continue
+            key_text = str(key).strip().lower()
+            if not key_text:
+                continue
+            if isinstance(value, (list, tuple, set)):
+                normalized[key_text] = value
+            else:
+                text = str(value).strip()
+                if text:
+                    normalized[key_text] = text
+        return normalized
+    
+    @staticmethod
+    def _split_people_field(raw: Any) -> List[str]:
+        if raw is None:
+            return []
+        if isinstance(raw, (list, tuple, set)):
+            results = []
+            for item in raw:
+                results.extend(ExportService._split_people_field(item))
+            return results
+        text = str(raw or "").strip()
+        if not text:
+            return []
+        import re
+        tokens = [token.strip() for token in re.split(r"[;,/&]|\band\b", text, flags=re.IGNORECASE) if token.strip()]
+        seen = set()
+        ordered = []
+        for token in tokens:
+            key = token.casefold()
+            if key in seen:
+                continue
+            seen.add(key)
+            ordered.append(token)
+        return ordered
+    
+    @staticmethod
+    def _split_simple_list(raw: Any) -> List[str]:
+        if raw is None:
+            return []
+        if isinstance(raw, (list, tuple, set)):
+            results = []
+            for item in raw:
+                results.extend(ExportService._split_simple_list(item))
+            return results
+        text = str(raw or "").strip()
+        if not text:
+            return []
+        import re
+        tokens = [token.strip() for token in re.split(r"[;,\n]", text) if token.strip()]
+        seen = set()
+        ordered = []
+        for token in tokens:
+            key = token.casefold()
+            if key in seen:
+                continue
+            seen.add(key)
+            ordered.append(token)
+        return ordered
+    
+    @staticmethod
+    def _first_nonempty(*values: Any) -> Optional[str]:
+        for value in values:
+            if value is None:
+                continue
+            if isinstance(value, (list, tuple, set)):
+                items = list(value)
+                if not items:
+                    continue
+                value = items[0]
+            text = str(value).strip()
+            if text:
+                return text
+        return None
+    
+    @staticmethod
+    def _extract_year(raw: Optional[str]) -> Optional[int]:
+        if not raw:
+            return None
+        text = str(raw).strip()
+        if not text:
+            return None
+        import re
+        match = re.search(r"(19|20)\d{2}", text)
+        if match:
+            try:
+                return int(match.group(0))
+            except ValueError:
+                return None
+        try:
+            parsed = int(text)
+        except ValueError:
+            return None
+        if 0 < parsed < 3000:
+            return parsed
+        return None
+    
+    @staticmethod
+    def _normalize_series_sequence(raw: Any) -> Optional[str]:
+        if raw is None:
+            return None
+        if isinstance(raw, (int, float)):
+            if isinstance(raw, float) and (raw != raw or raw == float("inf") or raw == float("-inf")):
+                return None
+            text = str(raw)
+        else:
+            text = str(raw).strip()
+        if not text:
+            return None
+        candidate = text.replace(",", ".")
+        import re
+        match = re.search(r"\d+(?:\.\d+)?", candidate)
+        if not match:
+            return None
+        normalized = match.group(0)
+        if "." in normalized:
+            normalized = normalized.rstrip("0").rstrip(".")
+            return normalized or "0"
+        try:
+            return str(int(normalized))
+        except ValueError:
+            cleaned = normalized.lstrip("0")
+            return cleaned or "0"
+    
+    @staticmethod
+    def _coerce_bool(value: Any, default: bool = True) -> bool:
+        if isinstance(value, bool):
+            return value
+        if isinstance(value, str):
+            lowered = value.strip().lower()
+            if lowered in {"true", "1", "yes", "on"}:
+                return True
+            if lowered in {"false", "0", "no", "off"}:
+                return False
+            return default
+        if value is None:
+            return default
+        return bool(value)
+
+
+__all__ = [
+    "ExportConfig",
+    "ExportService",
+]

abogen/infrastructure/subtitle_writer.py

@@ -0,0 +1,303 @@
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from enum import Enum
+from pathlib import Path
+from typing import List, Optional, TextIO
+
+from abogen.subtitle_utils import clean_subtitle_text
+
+
+class SubtitleFormat(Enum):
+    SRT = "srt"
+    ASS = "ass"
+    VTT = "vtt"
+
+
+class SubtitleMode(Enum):
+    DISABLED = "Disabled"
+    LINE = "Line"
+    SENTENCE = "Sentence"
+    SENTENCE_COMMA = "Sentence + Comma"
+    SENTENCE_HIGHLIGHT = "Sentence + Highlighting"
+
+
+class SubtitleAlignment(Enum):
+    LEFT = "left"
+    CENTER = "center"
+    NARROW = "narrow"
+    CENTER_NARROW = "center_narrow"
+
+
+@dataclass
+class SubtitleConfig:
+    """Configuration for subtitle writer."""
+    format: SubtitleFormat
+    mode: SubtitleMode
+    alignment: SubtitleAlignment = SubtitleAlignment.LEFT
+    max_words: int = 50
+    highlight_color: str = "&H00FFFF00"  # ASS highlight color
+
+
+class SubtitleWriter(ABC):
+    """Abstract base class for subtitle writers."""
+    
+    def __init__(self, path: Path, config: SubtitleConfig):
+        self.path = path
+        self.config = config
+        self._file: Optional[TextIO] = None
+        self._index = 0
+        self._opened = False
+    
+    def open(self) -> None:
+        """Open the subtitle file and write header."""
+        if self._opened:
+            return
+        self._file = open(self.path, "w", encoding="utf-8", errors="replace")
+        self._write_header()
+        self._opened = True
+    
+    @abstractmethod
+    def _write_header(self) -> None:
+        pass
+    
+    def write_entry(
+        self,
+        start: float,
+        end: float,
+        text: str,
+        voice: Optional[str] = None,
+    ) -> None:
+        """Write a subtitle entry."""
+        if not self._opened:
+            self.open()
+        
+        text = clean_subtitle_text(text)
+        if not text:
+            return
+        
+        self._index += 1
+        self._write_entry(self._index, start, end, text, voice)
+    
+    @abstractmethod
+    def _write_entry(
+        self,
+        index: int,
+        start: float,
+        end: float,
+        text: str,
+        voice: Optional[str],
+    ) -> None:
+        pass
+    
+    def close(self) -> None:
+        """Close the subtitle file."""
+        if self._file:
+            self._file.close()
+            self._file = None
+            self._opened = False
+    
+    def __enter__(self) -> "SubtitleWriter":
+        self.open()
+        return self
+    
+    def __exit__(self, exc_type, exc_val, exc_tb) -> None:
+        self.close()
+
+
+class SrtWriter(SubtitleWriter):
+    """SRT subtitle writer."""
+    
+    def _write_header(self) -> None:
+        pass  # SRT has no header
+    
+    def _write_entry(
+        self,
+        index: int,
+        start: float,
+        end: float,
+        text: str,
+        voice: Optional[str],
+    ) -> None:
+        start_str = self._format_time(start)
+        end_str = self._format_time(end)
+        
+        if voice:
+            text = f"[{voice}] {text}"
+        
+        self._file.write(f"{index}\n")
+        self._file.write(f"{start_str} --> {end_str}\n")
+        self._file.write(f"{text}\n\n")
+    
+    @staticmethod
+    def _format_time(seconds: float) -> str:
+        hours = int(seconds // 3600)
+        minutes = int((seconds % 3600) // 60)
+        secs = int(seconds % 60)
+        millis = int((seconds - int(seconds)) * 1000)
+        return f"{hours:02d}:{minutes:02d}:{secs:02d},{millis:03d}"
+
+
+class VttWriter(SubtitleWriter):
+    """WebVTT subtitle writer."""
+    
+    def _write_header(self) -> None:
+        self._file.write("WEBVTT\n\n")
+    
+    def _write_entry(
+        self,
+        index: int,
+        start: float,
+        end: float,
+        text: str,
+        voice: Optional[str],
+    ) -> None:
+        start_str = self._format_time(start)
+        end_str = self._format_time(end)
+        
+        if voice:
+            text = f"[{voice}] {text}"
+        
+        self._file.write(f"{index}\n")
+        self._file.write(f"{start_str} --> {end_str}\n")
+        self._file.write(f"{text}\n\n")
+    
+    @staticmethod
+    def _format_time(seconds: float) -> str:
+        hours = int(seconds // 3600)
+        minutes = int((seconds % 3600) // 60)
+        secs = seconds % 60
+        return f"{hours:02d}:{minutes:02d}:{secs:06.3f}".replace(".", ".")
+
+
+class AssWriter(SubtitleWriter):
+    """ASS subtitle writer with karaoke highlighting support."""
+    
+    def __init__(self, path: Path, config: SubtitleConfig):
+        super().__init__(path, config)
+        self._is_centered = config.alignment in (SubtitleAlignment.CENTER, SubtitleAlignment.CENTER_NARROW)
+        self._is_narrow = config.alignment in (SubtitleAlignment.NARROW, SubtitleAlignment.CENTER_NARROW)
+    
+    def _write_header(self) -> None:
+        margin = "90" if self._is_narrow else "10"
+        alignment = "5" if self._is_centered else "2"
+        
+        self._file.write("[Script Info]\n")
+        self._file.write("Title: Generated by Abogen\n")
+        self._file.write("ScriptType: v4.00+\n\n")
+        
+        # Styles
+        self._file.write("[V4+ Styles]\n")
+        self._file.write(
+            "Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, "
+            "OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, "
+            "ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, "
+            "Alignment, MarginL, MarginR, MarginV, Encoding\n"
+        )
+        
+        if self.config.mode == SubtitleMode.SENTENCE_HIGHLIGHT:
+            # Karaoke style with highlighting
+            self._file.write(
+                f"Style: Default,Arial,24,&H00FFFFFF,&H00808080,&H00000000,&H00404040,"
+                f"0,0,0,0,100,100,0,0,3,2,0,{alignment},{margin},{margin},10,1\n"
+            )
+            self._file.write(
+                f"Style: Highlight,Arial,24,&H0000FFFF,&H00808080,&H00000000,&H00404040,"
+                f"0,0,0,0,100,100,0,0,3,2,0,{alignment},{margin},{margin},10,1\n\n"
+            )
+        else:
+            self._file.write(
+                f"Style: Default,Arial,24,&H00FFFFFF,&H00808080,&H00000000,&H00404040,"
+                f"0,0,0,0,100,100,0,0,3,2,0,{alignment},{margin},{margin},10,1\n\n"
+            )
+        
+        self._file.write("[Events]\n")
+        self._file.write(
+            "Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\n"
+        )
+    
+    def _write_entry(
+        self,
+        index: int,
+        start: float,
+        end: float,
+        text: str,
+        voice: Optional[str],
+    ) -> None:
+        start_str = self._format_time(start)
+        end_str = self._format_time(end)
+        
+        if voice:
+            text = f"[{voice}] {text}"
+        
+        style = "Default"
+        if self.config.mode == SubtitleMode.SENTENCE_HIGHLIGHT:
+            # Add karaoke tags for highlighting
+            text = self._add_karaoke_tags(text)
+            style = "Highlight"
+        
+        alignment_tag = r"{\an5}" if self._is_centered else ""
+        self._file.write(
+            f"Dialogue: 0,{start_str},{end_str},{style},,0,0,0,,{alignment_tag}{text}\n"
+        )
+    
+    def _add_karaoke_tags(self, text: str) -> str:
+        """Add karaoke highlighting tags to text."""
+        # Simple word-level karaoke timing
+        words = text.split()
+        if not words:
+            return text
+        
+        # This is a simplified version - real karaoke needs per-word timing
+        # For now, just return the text with the highlight color
+        return r"{\k100}" + r"{\k100}".join(words) + r"{\k0}"
+    
+    @staticmethod
+    def _format_time(seconds: float) -> str:
+        hours = int(seconds // 3600)
+        minutes = int((seconds % 3600) // 60)
+        secs = seconds % 60
+        return f"{hours}:{minutes:02d}:{secs:05.2f}"
+
+
+def create_subtitle_writer(
+    path: Path,
+    format: str,
+    mode: str,
+    alignment: str = "left",
+    max_words: int = 50,
+) -> SubtitleWriter:
+    """Factory function to create subtitle writer."""
+    fmt = SubtitleFormat(format.lower())
+    mode = SubtitleMode(mode)
+    align = SubtitleAlignment(alignment.lower())
+    
+    config = SubtitleConfig(
+        format=fmt,
+        mode=mode,
+        alignment=align,
+        max_words=max_words,
+    )
+    
+    if fmt == SubtitleFormat.SRT:
+        return SrtWriter(path, config)
+    elif fmt == SubtitleFormat.VTT:
+        return VttWriter(path, config)
+    elif fmt == SubtitleFormat.ASS:
+        return AssWriter(path, config)
+    else:
+        raise ValueError(f"Unsupported subtitle format: {format}")
+
+
+__all__ = [
+    "SubtitleFormat",
+    "SubtitleMode",
+    "SubtitleAlignment",
+    "SubtitleConfig",
+    "SubtitleWriter",
+    "SrtWriter",
+    "VttWriter",
+    "AssWriter",
+    "create_subtitle_writer",
+]

abogen/main.py

@@ -2,13 +2,14 @@
 
 from __future__ import annotations
 
-import atexit
 import os
 import platform
-import signal
-import sys
 
-from abogen.utils import load_config, prevent_sleep_end
+# Initialise global shutdown handling (atexit, signals, Qt) as early as possible.
+from abogen import shutdown  # noqa: F401
+shutdown.register_shutdown()
+
+from abogen.utils import load_config
 from abogen.webui.app import main as _run_web_ui
 
 # Configure Hugging Face Hub behaviour (mirrors legacy GUI defaults).
@@ -27,17 +28,6 @@ os.environ.setdefault("MIOPEN_CONV_PRECISE_ROCM_TUNING", "0")
 if platform.system() == "Darwin" and platform.processor() == "arm":
     os.environ.setdefault("PYTORCH_ENABLE_MPS_FALLBACK", "1")
 
-atexit.register(prevent_sleep_end)
-
-
-def _cleanup_sleep(signum, _frame):
-    prevent_sleep_end()
-    sys.exit(0)
-
-
-signal.signal(signal.SIGINT, _cleanup_sleep)
-signal.signal(signal.SIGTERM, _cleanup_sleep)
-
 
 def main() -> None:
     """Launch the Flask-based web UI."""

abogen/predownload_gui.py

@@ -21,7 +21,8 @@ from PyQt6.QtWidgets import (
 )
 from PyQt6.QtCore import QThread, pyqtSignal
 
-from abogen.constants import COLORS, VOICES_INTERNAL
+from abogen.constants import COLORS
+from abogen.tts_plugin.utils import get_voices
 from abogen.spacy_utils import SPACY_MODELS
 import abogen.hf_tracker
 
@@ -114,7 +115,7 @@ class PreDownloadWorker(QThread):
             self._voices_success = False
             return
 
-        voice_list = VOICES_INTERNAL
+        voice_list = get_voices("kokoro")
         for idx, voice in enumerate(voice_list, start=1):
             if self._cancelled:
                 self._voices_success = False
@@ -462,14 +463,14 @@ class PreDownloadDialog(QDialog):
         try:
             from huggingface_hub import try_to_load_from_cache
 
-            for voice in VOICES_INTERNAL:
+            for voice in get_voices("kokoro"):
                 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(VOICES_INTERNAL)
+            return False, list(get_voices("kokoro"))
         return (len(missing) == 0), missing
 
     def _check_kokoro_model(self) -> bool:

abogen/pyqt/conversion.py

@@ -5,6 +5,7 @@ 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,
@@ -13,20 +14,25 @@ from abogen.utils import (
 )
 from abogen.constants import (
     LANGUAGE_DESCRIPTIONS,
-    SAMPLE_VOICE_TEXTS,
     COLORS,
     CHAPTER_OPTIONS_COUNTDOWN,
     SUBTITLE_FORMATS,
     SUPPORTED_SOUND_FORMATS,
     SUPPORTED_SUBTITLE_FORMATS,
 )
-from abogen.tts_supertonic import SupertonicPipeline, SUPERTONIC_AVAILABLE_LANGS, DEFAULT_SUPERTONIC_VOICES
 from abogen.voice_formulas import get_new_voice
+from abogen.infrastructure.subtitle_writer import _format_timestamp
+from abogen.domain.split_pattern import get_split_pattern
+from abogen.domain.output_paths import (
+    resolve_output_directory,
+    build_output_path,
+    sanitize_output_stem,
+)
+from abogen.domain.audio_helpers import build_ffmpeg_command, to_float32
 import abogen.hf_tracker as hf_tracker
 import static_ffmpeg
 import threading  # for efficient waiting
 import subprocess
-import platform
 
 # Configuration constants
 _USER_RESPONSE_TIMEOUT = (
@@ -43,10 +49,7 @@ from abogen.subtitle_utils import (
     get_sample_voice_text,
     sanitize_name_for_os,
     _CHAPTER_MARKER_SEARCH_PATTERN,
-    _VOICE_MARKER_PATTERN,
-    _VOICE_MARKER_SEARCH_PATTERN,
-    split_text_by_voice_markers,
-    validate_voice_name,
+    split_text_by_voice_markers
 )
 
 class CountdownDialog(QDialog):
@@ -216,40 +219,6 @@ class ConversionThread(QThread):
     PUNCTUATION_SENTENCE_COMMA = ".!?,।。！？、，"
     PUNCTUATION_COMMAS = ",，、"
 
-    def _get_split_pattern(self, lang_code, subtitle_mode):
-        """
-        Get the appropriate split pattern based on language and subtitle mode.
-
-        Args:
-            lang_code: Language code (a, b, e, f, etc.)
-            subtitle_mode: Subtitle mode ("Sentence", "Sentence + Comma", "Line", etc.)
-
-        Returns:
-            Split pattern string
-        """
-        # For English, always use newline splitting only
-        if lang_code in ["a", "b"]:
-            return "\n"
-
-        # Determine spacing pattern based on language
-        spacing_pattern = r"\s*" if lang_code in ["z", "j"] else r"\s+"
-
-        # For Chinese/Japanese, when subtitle mode is Disabled or Line, prefer
-        # punctuation-based splitting instead of plain newline splitting.
-        if subtitle_mode in ("Disabled", "Line") and lang_code in ["z", "j"]:
-            return r"(?<=[{}]){}|\n+".format(self.PUNCTUATION_SENTENCE, spacing_pattern)
-
-        if subtitle_mode == "Line":
-            return "\n"
-        elif subtitle_mode == "Sentence":
-            return r"(?<=[{}]){}|\n+".format(self.PUNCTUATION_SENTENCE, spacing_pattern)
-        elif subtitle_mode == "Sentence + Comma":
-            return r"(?<=[{}]){}|\n+".format(
-                self.PUNCTUATION_SENTENCE_COMMA, spacing_pattern
-            )
-        else:
-            return r"\n+"  # Default to line breaks
-
     def __init__(
         self,
         file_name,
@@ -260,30 +229,21 @@ class ConversionThread(QThread):
         output_folder,
         subtitle_mode,
         output_format,
-        np_module,
-        kpipeline_class,
+        backend,
         start_time,
         total_char_count,
         use_gpu=True,
         from_queue=False,
         save_base_path=None,
-        tts_provider="kokoro",
-        supertonic_language="en",
-        supertonic_total_steps=8,
-    ):
+    ):  # Add use_gpu parameter
         super().__init__()
         self._chapter_options_event = threading.Event()
         self._timestamp_response_event = threading.Event()
-        self.np = np_module
-        self.KPipeline = kpipeline_class
+        self.backend = backend
         self.file_name = file_name
         self.lang_code = lang_code
         self.speed = speed
         self.voice = voice
-        self.tts_provider = tts_provider
-        self.supertonic_language = supertonic_language
-        self.supertonic_total_steps = supertonic_total_steps
-        self.sample_rate = 44100 if tts_provider == "supertonic" else 24000
         self.save_option = save_option
         self.output_folder = output_folder
         self.subtitle_mode = subtitle_mode
@@ -307,7 +267,7 @@ class ConversionThread(QThread):
         self.silence_duration = 2.0  # Default value, will be overridden from GUI
         self.use_spacy_segmentation = True  # Default, will be overridden from GUI
         # Set split pattern based on language and subtitle mode
-        self.split_pattern = self._get_split_pattern(lang_code, subtitle_mode)
+        self.split_pattern = get_split_pattern(lang_code, subtitle_mode)
         self.voice_cache = {}  # Cache for loaded voices
 
     def load_voice_cached(self, voice_name, tts):
@@ -435,10 +395,6 @@ class ConversionThread(QThread):
             )
             self.log_updated.emit(f"- Voice: {self.voice}")
             self.log_updated.emit(f"- Speed: {self.speed}")
-            tts_provider_label = self.tts_provider.capitalize()
-            if self.tts_provider == "supertonic":
-                tts_provider_label += f" (lang={self.supertonic_language}, steps={self.supertonic_total_steps})"
-            self.log_updated.emit(f"- TTS Engine: {tts_provider_label}")
             self.log_updated.emit(f"- Subtitle mode: {self.subtitle_mode}")
             self.log_updated.emit(f"- Output format: {self.output_format}")
             self.log_updated.emit(
@@ -502,26 +458,6 @@ 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"
-
-            if self.tts_provider == "supertonic":
-                tts = SupertonicPipeline(
-                    sample_rate=self.sample_rate,
-                    lang=self.supertonic_language,
-                    total_steps=self.supertonic_total_steps,
-                )
-            else:
-                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
@@ -557,7 +493,7 @@ class ConversionThread(QThread):
 
             # Process subtitle files separately
             if is_subtitle_file or is_timestamp_text:
-                self._process_subtitle_file(tts, base_path, is_timestamp_text)
+                self._process_subtitle_file(self.backend, base_path, is_timestamp_text)
                 return
 
             if self.is_direct_text:
@@ -709,15 +645,17 @@ class ConversionThread(QThread):
                 base_path = self.display_path if self.display_path else self.file_name
 
             base_name = os.path.splitext(os.path.basename(base_path))[0]
-            # Sanitize base_name for folder/file creation based on OS
             sanitized_base_name = sanitize_name_for_os(base_name, is_folder=True)
 
-            if self.save_option == "Save to Desktop":
-                parent_dir = user_desktop_dir()
-            elif self.save_option == "Save next to input file":
-                parent_dir = os.path.dirname(base_path)
-            else:
-                parent_dir = self.output_folder or os.getcwd()
+            parent_dir = resolve_output_directory(
+                save_mode=self.save_option,
+                stored_path=Path(base_path),
+                output_folder=getattr(self, "output_folder", None),
+                desktop_dir=Path(user_desktop_dir()),
+                user_output_path=None,
+                user_cache_outputs=Path(os.getcwd()),
+            )
+            parent_dir = str(parent_dir)
             # Ensure the output folder exists, error if it doesn't
             if not os.path.exists(parent_dir):
                 self.log_updated.emit(
@@ -766,7 +704,7 @@ class ConversionThread(QThread):
                 merged_out_path = f"{base_filepath_no_ext}.{self.output_format}"
                 subtitle_entries = []
                 current_time = 0.0
-                rate = self.sample_rate
+                rate = 24000
                 subtitle_mode = self.subtitle_mode
                 self.etr_start_time = time.time()
                 self.processed_char_count = 0
@@ -782,38 +720,45 @@ class ConversionThread(QThread):
                     merged_out_file = sf.SoundFile(
                         merged_out_path,
                         "w",
-                        samplerate=self.sample_rate,
+                        samplerate=24000,
                         channels=1,
                         format=self.output_format,
                     )
                     ffmpeg_proc = None
-                elif self.output_format == "m4b":
-                    # Real-time M4B generation using FFmpeg pipe
+                elif self.output_format in ("m4b", "opus"):
+                    # Real-time generation using FFmpeg pipe
                     static_ffmpeg.add_paths()
                     merged_out_file = None
                     ffmpeg_proc = None
                     metadata_options, cover_path = (
                         self._extract_and_add_metadata_tags_to_ffmpeg_cmd()
+                        if self.output_format == "m4b"
+                        else ([], None)
                     )
-                    # Prepare ffmpeg command for m4b output
-                    cmd = [
-                        "ffmpeg",
-                            "-y",
-                            "-thread_queue_size",
-                            "32768",
-                            "-f",
-                            "f32le",
-                            "-ar",
-                            str(self.sample_rate),
-                            "-ac",
-                            "1",
-                            "-i",
-                            "pipe:0",
-                        ]
-                    cmd.extend(["-c:a", "libopus", "-b:a", "24000"])
-                    cmd.append(merged_out_path)
+                    cmd = build_ffmpeg_command(
+                        Path(merged_out_path),
+                        self.output_format,
+                    )
+                    # Insert thread queue size after ffmpeg header
+                    cmd.insert(2, "-thread_queue_size")
+                    cmd.insert(3, "32768")
+                    if self.output_format == "m4b" and cover_path and os.path.exists(cover_path):
+                        # Insert cover image input before the output path
+                        output_path = cmd.pop()
+                        cmd.extend([
+                            "-i", cover_path,
+                            "-map", "0:a",
+                            "-map", "1",
+                            "-c:v", "copy",
+                            "-disposition:v", "attached_pic",
+                        ])
+                        cmd.extend(metadata_options)
+                        cmd.append(output_path)
+                    elif self.output_format == "m4b":
+                        output_path = cmd.pop()
+                        cmd.extend(metadata_options)
+                        cmd.append(output_path)
                     ffmpeg_proc = create_process(cmd, stdin=subprocess.PIPE, text=False)
-                    merged_out_file = None
                 else:
                     self.log_updated.emit(
                         (f"Unsupported output format: {self.output_format}", "red")
@@ -891,7 +836,7 @@ class ConversionThread(QThread):
                 merged_out_path = None
                 subtitle_entries = []
                 current_time = 0.0
-                rate = self.sample_rate
+                rate = 24000
                 subtitle_mode = self.subtitle_mode
                 self.etr_start_time = time.time()
                 self.processed_char_count = 0
@@ -944,7 +889,7 @@ class ConversionThread(QThread):
                         chapter_out_file = sf.SoundFile(
                             chapter_out_path,
                             "w",
-                            samplerate=self.sample_rate,
+                            samplerate=24000,
                             channels=1,
                             format=separate_chapters_format,
                         )
@@ -959,7 +904,7 @@ class ConversionThread(QThread):
                             "-f",
                             "f32le",
                             "-ar",
-                            str(self.sample_rate),
+                            "24000",
                             "-ac",
                             "1",
                             "-i",
@@ -1046,7 +991,7 @@ 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, tts)
+                        loaded_voice = self.load_voice_cached(voice_name, self.backend)
                         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"))
@@ -1055,7 +1000,7 @@ 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, tts)
+                            loaded_voice = self.load_voice_cached(self.voice, self.backend)
 
                     # Determine if spaCy segmentation should be used for PRE-TTS segmentation
                     # Only non-English languages use spaCy for pre-segmentation
@@ -1141,7 +1086,7 @@ class ConversionThread(QThread):
                         print("Using split pattern: (unprintable)")
 
                     for text_segment in text_segments:
-                        for result in tts(
+                        for result in self.backend(
                             text_segment,
                             voice=loaded_voice,
                             speed=self.speed,
@@ -1247,8 +1192,8 @@ class ConversionThread(QThread):
                                         )
                                         if "ass" in subtitle_format:
                                             for start, end, text in new_entries:
-                                                start_time = self._ass_time(start)
-                                                end_time = self._ass_time(end)
+                                                start_time = _format_timestamp(start, ass=True)
+                                                end_time = _format_timestamp(end, ass=True)
                                                 # Use karaoke effect for highlighting mode
                                                 effect = (
                                                     "karaoke"
@@ -1263,7 +1208,7 @@ class ConversionThread(QThread):
                                             for entry in new_entries:
                                                 start, end, text = entry
                                                 merged_subtitle_file.write(
-                                                    f"{merged_srt_index}\n{self._srt_time(start)} --> {self._srt_time(end)}\n{text}\n\n"
+                                                    f"{merged_srt_index}\n{_format_timestamp(start)} --> {_format_timestamp(end)}\n{text}\n\n"
                                                 )
                                                 merged_srt_index += 1
                                 # Per-chapter subtitle processing for both file and ffmpeg_proc
@@ -1281,8 +1226,8 @@ class ConversionThread(QThread):
                                         )
                                         if "ass" in subtitle_format:
                                             for start, end, text in new_chapter_entries:
-                                                start_time = self._ass_time(start)
-                                                end_time = self._ass_time(end)
+                                                start_time = _format_timestamp(start, ass=True)
+                                                end_time = _format_timestamp(end, ass=True)
                                                 # Use karaoke effect for highlighting mode
                                                 effect = (
                                                     "karaoke"
@@ -1297,7 +1242,7 @@ class ConversionThread(QThread):
                                             for entry in new_chapter_entries:
                                                 start, end, text = entry
                                                 chapter_subtitle_file.write(
-                                                    f"{chapter_srt_index}\n{self._srt_time(start)} --> {self._srt_time(end)}\n{text}\n\n"
+                                                    f"{chapter_srt_index}\n{_format_timestamp(start)} --> {_format_timestamp(end)}\n{text}\n\n"
                                                 )
                                                 chapter_srt_index += 1
                             if merge_chapters_at_end:
@@ -1341,9 +1286,9 @@ class ConversionThread(QThread):
                 # Add silence between chapters for merged output (except after the last chapter)
                 if merge_chapters_at_end and chapter_idx < total_chapters:
                     silence_samples = int(
-                        self.silence_duration * self.sample_rate
+                        self.silence_duration * 24000
                     )  # Silence duration at 24,000 Hz
-                    silence_audio = self.np.zeros(silence_samples, dtype="float32")
+                    silence_audio = np.zeros(silence_samples, dtype="float32")
                     silence_bytes = silence_audio.tobytes()
 
                     if merged_out_file:
@@ -1572,7 +1517,7 @@ class ConversionThread(QThread):
                 parent_dir, f"{sanitized_base_name}{suffix}"
             )
             merged_out_path = f"{base_filepath_no_ext}.{self.output_format}"
-            rate = self.sample_rate
+            rate = 24000
 
             # Setup audio output
             merged_out_file, ffmpeg_proc = None, None
@@ -1586,58 +1531,27 @@ class ConversionThread(QThread):
                 )
             else:
                 static_ffmpeg.add_paths()
-                cmd = [
-                    "ffmpeg",
-                    "-y",
-                    "-thread_queue_size",
-                    "32768",
-                    "-f",
-                    "f32le",
-                    "-ar",
-                    str(rate),
-                    "-ac",
-                    "1",
-                    "-i",
-                    "pipe:0",
-                ]
+                cmd = build_ffmpeg_command(
+                    Path(merged_out_path),
+                    self.output_format,
+                )
+                cmd.insert(2, "-thread_queue_size")
+                cmd.insert(3, "32768")
                 if self.output_format == "m4b":
                     metadata_options, cover_path = (
                         self._extract_and_add_metadata_tags_to_ffmpeg_cmd()
                     )
                     if cover_path and os.path.exists(cover_path):
-                        cmd.extend(
-                            [
-                                "-i",
-                                cover_path,
-                                "-map",
-                                "0:a",
-                                "-map",
-                                "1",
-                                "-c:v",
-                                "copy",
-                                "-disposition:v",
-                                "attached_pic",
-                            ]
-                        )
-                    cmd.extend(
-                        [
-                            "-c:a",
-                            "aac",
-                            "-q:a",
-                            "2",
-                            "-movflags",
-                            "+faststart+use_metadata_tags",
-                        ]
-                    )
+                        output_path = cmd.pop()
+                        cmd.extend([
+                            "-i", cover_path,
+                            "-map", "0:a",
+                            "-map", "1",
+                            "-c:v", "copy",
+                            "-disposition:v", "attached_pic",
+                        ])
+                        cmd.append(output_path)
                     cmd.extend(metadata_options)
-                elif self.output_format == "opus":
-                    cmd.extend(["-c:a", "libopus", "-b:a", "24000"])
-                else:
-                    self.log_updated.emit(
-                        (f"Unsupported output format: {self.output_format}", "red")
-                    )
-                    return
-                cmd.append(merged_out_path)
                 ffmpeg_proc = create_process(cmd, stdin=subprocess.PIPE, text=False)
 
             # Always generate subtitles for subtitle input files
@@ -1682,7 +1596,7 @@ class ConversionThread(QThread):
             max_end_time = max(
                 (end for _, end, _ in subtitles if end is not None), default=0
             )
-            audio_buffer = self.np.zeros(
+            audio_buffer = np.zeros(
                 int(max_end_time * rate) + rate, dtype="float32"
             )
 
@@ -1746,7 +1660,7 @@ class ConversionThread(QThread):
                 # Generate TTS audio
                 tts_results = [
                     r
-                    for r in tts(
+                    for r in self.backend(
                         processed_text,
                         voice=loaded_voice,
                         speed=self.speed,
@@ -1764,11 +1678,11 @@ class ConversionThread(QThread):
 
                 # Concatenate audio and determine duration
                 full_audio = (
-                    self.np.concatenate(
+                    np.concatenate(
                         [a.numpy() if hasattr(a, "numpy") else a for a in audio_chunks]
                     )
                     if audio_chunks
-                    else self.np.zeros(
+                    else np.zeros(
                         int((subtitle_duration or 0) * rate), dtype="float32"
                     )
                 )
@@ -1802,8 +1716,8 @@ class ConversionThread(QThread):
                         num_stages = max(
                             1,
                             int(
-                                self.np.ceil(
-                                    self.np.log(speed_factor) / self.np.log(2.0)
+                                np.ceil(
+                                    np.log(speed_factor) / np.log(2.0)
                                 )
                             ),
                         )
@@ -1836,7 +1750,7 @@ class ConversionThread(QThread):
                             stdout=subprocess.PIPE,
                             stderr=subprocess.PIPE,
                         )
-                        full_audio = self.np.frombuffer(
+                        full_audio = np.frombuffer(
                             speed_proc.communicate(input=full_audio.tobytes())[0],
                             dtype="float32",
                         )
@@ -1850,7 +1764,7 @@ class ConversionThread(QThread):
 
                         tts_results = [
                             r
-                            for r in tts(
+                            for r in self.backend(
                                 processed_text,
                                 voice=loaded_voice,
                                 speed=new_speed,
@@ -1861,14 +1775,14 @@ class ConversionThread(QThread):
                         audio_chunks = [r.audio for r in tts_results]
 
                         full_audio = (
-                            self.np.concatenate(
+                            np.concatenate(
                                 [
                                     a.numpy() if hasattr(a, "numpy") else a
                                     for a in audio_chunks
                                 ]
                             )
                             if audio_chunks
-                            else self.np.zeros(
+                            else np.zeros(
                                 int(subtitle_duration * rate), dtype="float32"
                             )
                         )
@@ -1885,10 +1799,10 @@ class ConversionThread(QThread):
                 # Pad or trim to subtitle duration
                 target_samples = int(subtitle_duration * rate)
                 if len(full_audio) < target_samples:
-                    full_audio = self.np.concatenate(
+                    full_audio = np.concatenate(
                         [
                             full_audio,
-                            self.np.zeros(
+                            np.zeros(
                                 target_samples - len(full_audio), dtype="float32"
                             ),
                         ]
@@ -1901,10 +1815,10 @@ class ConversionThread(QThread):
                 end_sample = start_sample + len(full_audio)
                 if end_sample > len(audio_buffer):
                     # Extend buffer if needed
-                    audio_buffer = self.np.concatenate(
+                    audio_buffer = np.concatenate(
                         [
                             audio_buffer,
-                            self.np.zeros(
+                            np.zeros(
                                 end_sample - len(audio_buffer), dtype="float32"
                             ),
                         ]
@@ -1927,11 +1841,11 @@ class ConversionThread(QThread):
                             else processed_text.replace("\n", "\\N")
                         )
                         subtitle_file.write(
-                            f"Dialogue: 0,{self._ass_time(start_time)},{self._ass_time(end_time)},Default,,{margin},{margin},0,{effect},{alignment}{ass_text}\n"
+                            f"Dialogue: 0,{_format_timestamp(start_time, ass=True)},{_format_timestamp(end_time, ass=True)},Default,,{margin},{margin},0,{effect},{alignment}{ass_text}\n"
                         )
                     else:
                         subtitle_file.write(
-                            f"{srt_index}\n{self._srt_time(start_time)} --> {self._srt_time(end_time)}\n{processed_text}\n\n"
+                            f"{srt_index}\n{_format_timestamp(start_time)} --> {_format_timestamp(end_time)}\n{processed_text}\n\n"
                         )
                         srt_index += 1
 
@@ -1946,7 +1860,7 @@ class ConversionThread(QThread):
                 self.progress_updated.emit(percent, etr_str)
 
             # Normalize audio buffer to prevent clipping from mixed overlaps
-            max_amplitude = self.np.abs(audio_buffer).max()
+            max_amplitude = np.abs(audio_buffer).max()
             if max_amplitude > 1.0:
                 self.log_updated.emit(
                     f"\n  -> Normalizing audio (peak: {max_amplitude:.2f})"
@@ -2093,22 +2007,6 @@ class ConversionThread(QThread):
         # Add these to ffmpeg command
         return metadata_options, cover_path
 
-    def _srt_time(self, t):
-        """Helper function to format time for SRT files"""
-        h = int(t // 3600)
-        m = int((t % 3600) // 60)
-        s = int(t % 60)
-        ms = int((t - int(t)) * 1000)
-        return f"{h:02d}:{m:02d}:{s:02d},{ms:03d}"
-
-    def _ass_time(self, t):
-        """Helper function to format time for ASS files"""
-        h = int(t // 3600)
-        m = int((t % 3600) // 60)
-        s = int(t % 60)
-        cs = int((t - int(t)) * 100)  # Centiseconds for ASS format
-        return f"{h:01d}:{m:02d}:{s:02d}.{cs:02d}"
-
     def _process_subtitle_tokens(
         self,
         tokens_with_timestamps,
@@ -2386,6 +2284,10 @@ class ConversionThread(QThread):
         self.cancel_requested = True
         self.should_cancel = True
         self.waiting_for_user_input = False
+        # Clear voice cache (instance and module-level)
+        self.voice_cache.clear()
+        from abogen.voice_cache import clear_voice_cache
+        clear_voice_cache()
         # Terminate subprocess if running
         if self.process:
             try:
@@ -2415,28 +2317,19 @@ class VoicePreviewThread(QThread):
 
     def __init__(
         self,
-        np_module,
-        kpipeline_class,
+        backend,
         lang_code,
         voice,
         speed,
         use_gpu=False,
         parent=None,
-        tts_provider="kokoro",
-        supertonic_language="en",
-        supertonic_total_steps=8,
     ):
         super().__init__(parent)
-        self.np_module = np_module
-        self.kpipeline_class = kpipeline_class
+        self.backend = backend
         self.lang_code = lang_code
         self.voice = voice
         self.speed = speed
         self.use_gpu = use_gpu
-        self.tts_provider = tts_provider
-        self.supertonic_language = supertonic_language
-        self.supertonic_total_steps = supertonic_total_steps
-        self.sample_rate = 44100 if tts_provider == "supertonic" else 24000
 
         # Cache location for preview audio
         self.cache_dir = get_user_cache_path("preview_cache")
@@ -2446,11 +2339,6 @@ class VoicePreviewThread(QThread):
 
     def _get_cache_path(self):
         """Generate a unique filename for the voice with its parameters"""
-        if self.tts_provider == "supertonic":
-            voice_id = self.voice or "M1"
-            filename = f"st_{voice_id}_{self.supertonic_language}_steps{self.supertonic_total_steps}_{self.speed:.2f}.wav"
-            return os.path.join(self.cache_dir, filename)
-
         # For a voice formula, use a hash of the formula
         if "*" in self.voice:
             voice_id = (
@@ -2465,56 +2353,27 @@ class VoicePreviewThread(QThread):
 
     def run(self):
         print(
-            f"\nVoice: {self.voice}\nLanguage: {self.lang_code}\nSpeed: {self.speed}\nGPU: {self.use_gpu}\nTTS Provider: {self.tts_provider}\n"
+            f"\nVoice: {self.voice}\nLanguage: {self.lang_code}\nSpeed: {self.speed}\nGPU: {self.use_gpu}\n"
         )
 
         # Generate the preview and save to cache
         try:
-            if self.tts_provider == "supertonic":
-                from abogen.tts_supertonic import SupertonicPipeline
 
-                tts = SupertonicPipeline(
-                    sample_rate=self.sample_rate,
-                    lang=self.supertonic_language,
-                    total_steps=self.supertonic_total_steps,
-                )
-                loaded_voice = self.voice or "M1"
-                sample_text = "Hello, this is a sample of the selected voice."
-                audio_segments = []
-                for result in tts(
-                    sample_text,
-                    voice=loaded_voice,
-                    speed=self.speed,
-                    split_pattern=None,
-                ):
-                    audio_segments.append(result.audio)
-            else:
-                # Set device based on use_gpu setting and platform
-                if self.use_gpu:
-                    if platform.system() == "Darwin" and platform.processor() == "arm":
-                        device = "mps"
-                    else:
-                        device = "cuda"
-                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(tts, self.voice, self.use_gpu)
+                loaded_voice = get_new_voice(self.backend, self.voice, self.use_gpu)
             else:
                 loaded_voice = self.voice
             sample_text = get_sample_voice_text(self.lang_code)
             audio_segments = []
-                for result in tts(
+            for result in self.backend(
                 sample_text, voice=loaded_voice, speed=self.speed, split_pattern=None
             ):
                 audio_segments.append(result.audio)
-
             if audio_segments:
-                audio = self.np_module.concatenate(audio_segments)
-                sf.write(self.cache_path, audio, self.sample_rate)
+                audio = np.concatenate(audio_segments)
+                # Save directly to the cache path
+                sf.write(self.cache_path, audio, 24000)
                 self.temp_wav = self.cache_path
             self.finished.emit()
         except Exception as e:

abogen/pyqt/gui.py

@@ -7,9 +7,9 @@ import base64
 import re
 from abogen.pyqt.queue_manager_gui import QueueManager
 from abogen.pyqt.queued_item import QueuedItem
+from abogen.domain.device import select_device as _select_device
 import abogen.hf_tracker as hf_tracker
 import hashlib  # Added for cache path generation
-from abogen.tts_supertonic import SUPERTONIC_AVAILABLE_LANGS, DEFAULT_SUPERTONIC_VOICES
 from PyQt6.QtWidgets import (
     QApplication,
     QWidget,
@@ -83,13 +83,11 @@ from abogen.constants import (
     GITHUB_URL,
     PROGRAM_DESCRIPTION,
     LANGUAGE_DESCRIPTIONS,
-    VOICES_INTERNAL,
-    KOKORO_LANG_TO_COUNTRY,
-    SUPERTONIC_LANG_TO_COUNTRY,
     SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
     COLORS,
     SUBTITLE_FORMATS,
 )
+from abogen.tts_plugin.utils import get_voices
 import threading
 from abogen.pyqt.voice_formula_gui import VoiceFormulaDialog
 from abogen.voice_profiles import load_profiles
@@ -973,9 +971,6 @@ class abogen(QWidget):
         self.fix_nonstandard_punctuation = self.config.get(
             "fix_nonstandard_punctuation", False
         )
-        self.tts_provider_config = self.config.get("tts_provider", "kokoro")
-        self.supertonic_language_config = self.config.get("supertonic_language", "en")
-        self.supertonic_total_steps_config = self.config.get("supertonic_total_steps", 8)
         self._pending_close_event = None
         self.gpu_ok = False  # Initialize GPU availability status
 
@@ -1024,16 +1019,6 @@ class abogen(QWidget):
                 else:
                     self.mixed_voice_state = entry
                     self.selected_lang = entry[0][0] if entry and entry[0] else None
-        # Restore TTS provider and supertonic settings from config
-        provider_text = "Supertonic" if self.tts_provider_config == "supertonic" else "Kokoro"
-        idx_st = self.provider_combo.findText(provider_text)
-        if idx_st >= 0:
-            self.provider_combo.setCurrentIndex(idx_st)
-        self.st_lang_combo.setCurrentText(self.supertonic_language_config)
-        idx_steps = self.st_steps_combo.findData(self.supertonic_total_steps_config)
-        if idx_steps >= 0:
-            self.st_steps_combo.setCurrentIndex(idx_steps)
-
         if self.save_option == "Choose output folder" and self.selected_output_folder:
             self.save_path_label.setText(self.selected_output_folder)
             self.save_path_row_widget.show()
@@ -1123,53 +1108,6 @@ class abogen(QWidget):
         speed_layout.addWidget(self.speed_label)
         controls_layout.addLayout(speed_layout)
         self.speed_slider.valueChanged.connect(self.update_speed_label)
-
-        # TTS Provider selection
-        provider_layout = QHBoxLayout()
-        provider_layout.setSpacing(7)
-        provider_label = QLabel("TTS Engine:", self)
-        provider_layout.addWidget(provider_label)
-        self.provider_combo = QComboBox(self)
-        self.provider_combo.addItem("Kokoro", "kokoro")
-        self.provider_combo.addItem("Supertonic", "supertonic")
-        self.provider_combo.setStyleSheet("QComboBox { min-height: 20px; padding: 6px 12px; }")
-        self.provider_combo.currentIndexChanged.connect(self.on_provider_changed)
-        provider_layout.addWidget(self.provider_combo)
-        controls_layout.addLayout(provider_layout)
-
-        # Supertonic-specific controls (language + steps), hidden by default
-        self.supertonic_row = QWidget()
-        supertonic_row_layout = QHBoxLayout(self.supertonic_row)
-        supertonic_row_layout.setContentsMargins(0, 0, 0, 0)
-        supertonic_row_layout.setSpacing(7)
-
-        st_lang_label = QLabel("Language:", self)
-        supertonic_row_layout.addWidget(st_lang_label)
-        self.st_lang_combo = QComboBox(self)
-        for code in SUPERTONIC_AVAILABLE_LANGS:
-            country_code = SUPERTONIC_LANG_TO_COUNTRY.get(code, code)
-            flag = get_resource_path("abogen.assets.flags", f"{country_code}.png")
-            icon_st = QIcon(flag) if flag and os.path.exists(flag) else QIcon()
-            self.st_lang_combo.addItem(icon_st, code, code)
-        self.st_lang_combo.setCurrentText("en")
-        self.st_lang_combo.setStyleSheet("QComboBox { min-height: 20px; padding: 6px 12px; }")
-        self.st_lang_combo.currentTextChanged.connect(self._on_st_lang_changed)
-        supertonic_row_layout.addWidget(self.st_lang_combo)
-
-        st_steps_label = QLabel("Steps:", self)
-        supertonic_row_layout.addWidget(st_steps_label)
-        self.st_steps_combo = QComboBox(self)
-        for val in range(2, 16):
-            self.st_steps_combo.addItem(str(val), val)
-        self.st_steps_combo.setCurrentIndex(self.st_steps_combo.findData(8))
-        self.st_steps_combo.setStyleSheet("QComboBox { min-height: 20px; padding: 6px 12px; }")
-        self.st_steps_combo.currentIndexChanged.connect(self._on_st_steps_changed)
-        supertonic_row_layout.addWidget(self.st_steps_combo)
-
-        supertonic_row_layout.addStretch()
-        self.supertonic_row.hide()
-        controls_layout.addWidget(self.supertonic_row)
-
         # Voice selection
         voice_layout = QHBoxLayout()
         voice_layout.setSpacing(7)
@@ -1827,12 +1765,6 @@ class abogen(QWidget):
         Update the enabled state of subtitle options based on the selected language.
         For non-English languages, only sentence-based and line-based modes are supported.
         """
-        provider = self.provider_combo.currentData()
-        if provider == "supertonic":
-            self.subtitle_combo.setEnabled(False)
-            self.subtitle_format_combo.setEnabled(False)
-            return
-
         # Check if current file is a subtitle file
         is_subtitle_input = False
         if self.selected_file and self.selected_file.lower().endswith(
@@ -1892,48 +1824,6 @@ class abogen(QWidget):
         # Enable/disable subtitle options based on language
         self.update_subtitle_options_availability()
 
-    def on_provider_changed(self, index):
-        provider = self.provider_combo.itemData(index)
-        self.config["tts_provider"] = provider
-        save_config(self.config)
-        is_supertonic = provider == "supertonic"
-
-        # Show/hide Supertonic controls
-        self.supertonic_row.setVisible(is_supertonic)
-
-        # Update subtitles availability
-        self.update_subtitle_options_availability()
-
-        # Repopulate voice list
-        self.populate_profiles_in_voice_combo()
-
-        # Clear/reset mixed voice state when switching provider
-        if is_supertonic:
-            self.mixed_voice_state = None
-            self.btn_voice_formula_mixer.setEnabled(False)
-            self.voice_combo.setToolTip(
-                "Supertonic voices:\n"
-                "M1-M5 = Male voices\n"
-                "F1-F5 = Female voices"
-            )
-        else:
-            self.btn_voice_formula_mixer.setEnabled(True)
-            self.voice_combo.setToolTip(
-                "The first character represents the language:\n"
-                '"a" => American English\n"b" => British English\n"e" => Spanish\n"f" => French\n"h" => Hindi\n"i" => Italian\n"j" => Japanese\n"p" => Brazilian Portuguese\n"z" => Mandarin Chinese\nThe second character represents the gender:\n"m" => Male\n"f" => Female'
-            )
-
-    def _on_st_lang_changed(self, lang):
-        self.config["supertonic_language"] = lang
-        save_config(self.config)
-        if self.provider_combo.currentData() == "supertonic":
-            self.selected_lang = lang
-            self.update_subtitle_options_availability()
-
-    def _on_st_steps_changed(self):
-        self.config["supertonic_total_steps"] = self.st_steps_combo.currentData()
-        save_config(self.config)
-
     def on_voice_combo_changed(self, index):
         data = self.voice_combo.itemData(index)
         if isinstance(data, str) and data.startswith("profile:"):
@@ -1942,24 +1832,8 @@ class abogen(QWidget):
             from abogen.voice_profiles import load_profiles
 
             entry = load_profiles().get(pname, {})
+            # set mixed voices and language
             if isinstance(entry, dict):
-                entry_provider = str(entry.get("provider", "")).strip().lower()
-                if entry_provider == "supertonic":
-                    # Switch provider to Supertonic if not already
-                    if self.provider_combo.currentData() != "supertonic":
-                        self.provider_combo.setCurrentIndex(1)
-                    self.mixed_voice_state = None
-                    self.selected_lang = entry.get("language", self.st_lang_combo.currentText())
-                    # Sync supertonic controls from profile
-                    profile_steps = entry.get("total_steps")
-                    if profile_steps is not None:
-                        idx_steps = self.st_steps_combo.findData(int(profile_steps))
-                        if idx_steps >= 0:
-                            self.st_steps_combo.setCurrentIndex(idx_steps)
-                    profile_lang = entry.get("language")
-                    if profile_lang and profile_lang in SUPERTONIC_AVAILABLE_LANGS:
-                        self.st_lang_combo.setCurrentText(profile_lang)
-                else:
                 self.mixed_voice_state = entry.get("voices", [])
                 self.selected_lang = entry.get("language")
             else:
@@ -1974,12 +1848,7 @@ class abogen(QWidget):
         else:
             self.mixed_voice_state = None
             self.selected_profile_name = None
-            self.selected_voice = data
-            provider = self.provider_combo.currentData()
-            if provider == "supertonic":
-                self.selected_lang = self.st_lang_combo.currentText()
-            else:
-                self.selected_lang = data[0] if data else ""
+            self.selected_voice, self.selected_lang = data, data[0]
             self.config["selected_voice"] = data
             if "selected_profile_name" in self.config:
                 del self.config["selected_profile_name"]
@@ -1998,37 +1867,16 @@ class abogen(QWidget):
     def populate_profiles_in_voice_combo(self):
         # preserve current voice or profile
         current = self.voice_combo.currentData()
-        provider = self.provider_combo.currentData()
         self.voice_combo.blockSignals(True)
         self.voice_combo.clear()
-        # re-add profiles matching current provider
+        # re-add profiles
         profile_icon = QIcon(get_resource_path("abogen.assets", "profile.png"))
-        for pname, entry in load_profiles().items():
-            entry_provider = ""
-            if isinstance(entry, dict):
-                entry_provider = str(entry.get("provider", "")).strip().lower()
-            if provider == "supertonic":
-                if entry_provider == "supertonic":
-                    self.voice_combo.addItem(profile_icon, pname, f"profile:{pname}")
-            else:
-                if entry_provider != "supertonic":
+        for pname in load_profiles().keys():
             self.voice_combo.addItem(profile_icon, pname, f"profile:{pname}")
         # re-add voices
-        if provider == "supertonic":
-            for v in DEFAULT_SUPERTONIC_VOICES:
+        for v in get_voices("kokoro"):
             icon = QIcon()
-                if v.startswith("F"):
-                    icon_path = get_resource_path("abogen.assets", "female.png")
-                else:
-                    icon_path = get_resource_path("abogen.assets", "male.png")
-                if icon_path and os.path.exists(icon_path):
-                    icon = QIcon(icon_path)
-                self.voice_combo.addItem(icon, f"{v}", v)
-        else:
-            for v in VOICES_INTERNAL:
-                icon = QIcon()
-                country_code = KOKORO_LANG_TO_COUNTRY.get(v[0], v[0])
-                flag_path = get_resource_path("abogen.assets.flags", f"{country_code}.png")
+            flag_path = get_resource_path("abogen.assets.flags", f"{v[0]}.png")
             if flag_path and os.path.exists(flag_path):
                 icon = QIcon(flag_path)
             self.voice_combo.addItem(icon, f"{v}", v)
@@ -2222,9 +2070,6 @@ class abogen(QWidget):
             save_base_path=save_base_path,
             save_chapters_separately=getattr(self, "save_chapters_separately", None),
             merge_chapters_at_end=getattr(self, "merge_chapters_at_end", None),
-            tts_provider=self.provider_combo.currentData(),
-            supertonic_language=self.st_lang_combo.currentText(),
-            supertonic_total_steps=self.st_steps_combo.currentData(),
         )
 
         # Prevent adding duplicate items to the queue
@@ -2368,15 +2213,6 @@ class abogen(QWidget):
                 self.config["replace_numerals"] = self.replace_numerals
                 self.config["fix_nonstandard_punctuation"] = self.fix_nonstandard_punctuation
 
-                # TTS provider settings
-                tts_provider = getattr(queued_item, "tts_provider", "kokoro")
-                self.provider_combo.setCurrentText("Supertonic" if tts_provider == "supertonic" else "Kokoro")
-                self.st_lang_combo.setCurrentText(getattr(queued_item, "supertonic_language", "en"))
-                steps_val = getattr(queued_item, "supertonic_total_steps", 8)
-                idx_steps = self.st_steps_combo.findData(steps_val)
-                if idx_steps >= 0:
-                    self.st_steps_combo.setCurrentIndex(idx_steps)
-
                 # Sync Voice/Profile in config
                 self.config["selected_voice"] = self.selected_voice
                 if "selected_profile_name" in self.config:
@@ -2399,8 +2235,6 @@ class abogen(QWidget):
             self.current_queue_index = 0  # Reset for next time
 
     def get_voice_formula(self) -> str:
-        if self.provider_combo.currentData() == "supertonic":
-            return self._get_supertonic_voice()
         if self.mixed_voice_state:
             formula_components = [
                 f"{name}*{weight}" for name, weight in self.mixed_voice_state
@@ -2410,8 +2244,6 @@ class abogen(QWidget):
             return self.selected_voice
 
     def get_selected_lang(self, voice_formula) -> str:
-        if self.provider_combo.currentData() == "supertonic":
-            return self.st_lang_combo.currentText()
         if self.selected_profile_name:
             from abogen.voice_profiles import load_profiles
 
@@ -2485,9 +2317,9 @@ class abogen(QWidget):
             file_size_str = "Unknown"
 
         # pipeline_loaded_callback remains unchanged
-        def pipeline_loaded_callback(np_module, kpipeline_class, error):
+        def pipeline_loaded_callback(backend, error):
             if error:
-                self.update_log((f"Error loading numpy or KPipeline: {error}", "red"))
+                self.update_log((f"Error loading TTS backend: {error}", "red"))
                 prevent_sleep_end()
                 return
 
@@ -2501,10 +2333,6 @@ class abogen(QWidget):
             # determine selected language: use profile setting if profile selected, else voice code
             selected_lang = self.get_selected_lang(voice_formula)
 
-            tts_provider = self.provider_combo.currentData()
-            supertonic_language = self.st_lang_combo.currentText()
-            supertonic_total_steps = self.st_steps_combo.currentData()
-
             self.conversion_thread = ConversionThread(
                 self.selected_file,
                 selected_lang,
@@ -2514,17 +2342,13 @@ class abogen(QWidget):
                 self.selected_output_folder,
                 subtitle_mode=actual_subtitle_mode,
                 output_format=self.selected_format,
-                np_module=np_module,
-                kpipeline_class=kpipeline_class,
+                backend=backend,
                 start_time=self.start_time,
                 total_char_count=self.char_count,
                 use_gpu=self.gpu_ok,
                 from_queue=from_queue,
-                save_base_path=self.displayed_file_path,
-                tts_provider=tts_provider,
-                supertonic_language=supertonic_language,
-                supertonic_total_steps=supertonic_total_steps,
-            )
+                save_base_path=self.displayed_file_path,  # Pass the save base path (original file for EPUB)
+            )  # Use gpu_ok status
             # Pass the displayed file path to the log_updated signal handler in ConversionThread
             self.conversion_thread.display_path = display_path
             # Pass the file size string
@@ -2601,14 +2425,18 @@ class abogen(QWidget):
             # Store gpu_ok status to use when creating the conversion thread
             self.gpu_ok = gpu_ok
             self.update_log((gpu_msg, gpu_ok))
-            tts_provider = self.provider_combo.currentData()
-            if tts_provider == "supertonic":
-                # Supertonic doesn't need KPipeline, call callback directly
-                import numpy as np
-                pipeline_loaded_callback(np, None, None)
-            else:
             self.update_log("Loading modules...")
-                load_thread = LoadPipelineThread(pipeline_loaded_callback)
+
+            # Determine device based on GPU availability
+            if gpu_ok:
+                device = _select_device()
+            else:
+                device = "cpu"
+
+            lang_code = self.selected_lang or "a"
+            load_thread = LoadPipelineThread(
+                pipeline_loaded_callback, lang_code=lang_code, device=device
+            )
             load_thread.start()
 
         threading.Thread(target=gpu_and_load, daemon=True).start()
@@ -2922,32 +2750,9 @@ class abogen(QWidget):
                 "Open File Error", f"Could not open file:\n{e}"
             )
 
-    def _get_supertonic_voice(self) -> str:
-        """Resolve the effective Supertonic voice from the current combo selection."""
-        if self.selected_profile_name:
-            from abogen.voice_profiles import load_profiles
-            entry = load_profiles().get(self.selected_profile_name, {})
-            if isinstance(entry, dict):
-                return str(entry.get("voice", "M1"))
-            return "M1"
-        current_data = self.voice_combo.currentData()
-        if current_data and isinstance(current_data, str) and not current_data.startswith("profile:"):
-            return current_data
-        return "M1"
-
     def _get_preview_cache_path(self):
         """Generate the expected cache path for the current voice settings."""
         speed = self.speed_slider.value() / 100.0
-        provider = self.provider_combo.currentData()
-
-        if provider == "supertonic":
-            voice_to_cache = self._get_supertonic_voice()
-            lang_to_cache = self.st_lang_combo.currentText()
-            steps = self.st_steps_combo.currentData()
-            cache_dir = get_user_cache_path("preview_cache")
-            filename = f"st_{voice_to_cache}_{lang_to_cache}_steps{steps}_{speed:.2f}.wav"
-            return os.path.join(cache_dir, filename)
-
         voice_to_cache = ""
         lang_to_cache = ""
 
@@ -3052,13 +2857,6 @@ class abogen(QWidget):
         self.btn_voice_formula_mixer.setEnabled(False)  # Disable mixer button
         self.btn_start.setEnabled(False)  # Disable start button during preview
 
-        # For Supertonic, skip KPipeline loading and use SupertonicPipeline directly
-        if self.provider_combo.currentData() == "supertonic":
-            import numpy as np
-            self.loading_movie.start()
-            self._on_pipeline_loaded_for_preview(np, None, None)
-            return
-
         # Start loading animation - ensure signal connection is always active
         if hasattr(self, "loading_movie"):
             # Disconnect previous connections to avoid multiple connections
@@ -3075,18 +2873,24 @@ class abogen(QWidget):
             )
             self.loading_movie.start()
 
-        def pipeline_loaded_callback(np_module, kpipeline_class, error):
-            self._on_pipeline_loaded_for_preview(np_module, kpipeline_class, error)
+        # Determine device based on GPU availability
+        if self.gpu_ok:
+            device = _select_device()
+        else:
+            device = "cpu"
 
-        load_thread = LoadPipelineThread(pipeline_loaded_callback)
+        lang = self.selected_lang or "a"
+        load_thread = LoadPipelineThread(
+            self._on_pipeline_loaded_for_preview, lang_code=lang, device=device
+        )
         load_thread.start()
 
-    def _on_pipeline_loaded_for_preview(self, np_module, kpipeline_class, error):
+    def _on_pipeline_loaded_for_preview(self, backend, 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 numpy or KPipeline: {error}"
+                "Loading Error", f"Error loading TTS backend: {error}"
             )
             self.btn_preview.setIcon(self.play_icon)
             self.btn_preview.setEnabled(True)
@@ -3117,28 +2921,14 @@ class abogen(QWidget):
                     else None
                 )
         else:
-            if self.provider_combo.currentData() == "supertonic":
-                voice = self._get_supertonic_voice()
-            else:
-                voice = self.selected_voice or ""
-
-        tts_provider = self.provider_combo.currentData()
-        supertonic_language = self.st_lang_combo.currentText()
-        supertonic_total_steps = self.st_steps_combo.currentData()
-
-        if tts_provider == "supertonic":
-            lang = supertonic_language
-        else:
-            lang = self.selected_voice[0] if self.selected_voice else ""
+            lang = self.selected_voice[0]
+            voice = self.selected_voice
 
         # use same gpu/cpu logic as in conversion
         gpu_msg, gpu_ok = get_gpu_acceleration(self.use_gpu)
 
         self.preview_thread = VoicePreviewThread(
-            np_module, kpipeline_class, lang, voice, speed, gpu_ok,
-            tts_provider=tts_provider,
-            supertonic_language=supertonic_language,
-            supertonic_total_steps=supertonic_total_steps,
+            backend, lang, voice, speed, gpu_ok
         )
         self.preview_thread.finished.connect(self._play_preview_audio)
         self.preview_thread.error.connect(self._preview_error)
@@ -3441,12 +3231,16 @@ class abogen(QWidget):
             )
             box.setDefaultButton(QMessageBox.StandardButton.No)
             if box.exec() == QMessageBox.StandardButton.Yes:
+                from abogen import shutdown
+                shutdown.request_shutdown()
                 self.cleanup_conversion_thread()
                 self.cleanup_preview_threads()
                 event.accept()
             else:
                 event.ignore()
         else:
+            from abogen import shutdown
+            shutdown.request_shutdown()
             self.cleanup_conversion_thread()
             self.cleanup_preview_threads()
             event.accept()

abogen/pyqt/main.py

@@ -1,10 +1,10 @@
 import os
 import sys
 import platform
-import atexit
-import signal
-from abogen.utils import get_resource_path, load_config, prevent_sleep_end
 
+# Initialise global shutdown handling (atexit, signals, Qt) as early as possible.
+from abogen import shutdown  # noqa: F401
+shutdown.register_shutdown()
 
 # Fix PyTorch DLL loading issue ([WinError 1114]) on Windows before importing PyQt6
 if platform.system() == "Windows":
@@ -94,6 +94,7 @@ os.environ["HF_HUB_DISABLE_TELEMETRY"] = "1"  # Disable Hugging Face telemetry
 os.environ["HF_HUB_ETAG_TIMEOUT"] = "10"  # Metadata request timeout (seconds)
 os.environ["HF_HUB_DOWNLOAD_TIMEOUT"] = "10"  # File download timeout (seconds)
 os.environ["HF_HUB_DISABLE_SYMLINKS_WARNING"] = "1"  # Disable symlinks warning
+from abogen.utils import load_config
 if load_config().get("disable_kokoro_internet", False):
     print("INFO: Kokoro's internet access is disabled.")
     os.environ["HF_HUB_OFFLINE"] = "1"  # Disable Hugging Face Hub internet access
@@ -105,25 +106,6 @@ from abogen.constants import PROGRAM_NAME, VERSION
 os.environ["MIOPEN_FIND_MODE"] = "FAST"
 os.environ["MIOPEN_CONV_PRECISE_ROCM_TUNING"] = "0"
 
-# Reset sleep states
-atexit.register(prevent_sleep_end)
-
-
-# Also handle signals (Ctrl+C, kill, etc.)
-def _cleanup_sleep(signum, frame):
-    prevent_sleep_end()
-    sys.exit(0)
-
-
-signal.signal(signal.SIGINT, _cleanup_sleep)
-signal.signal(signal.SIGTERM, _cleanup_sleep)
-
-# Ensure sys.stdout and sys.stderr are valid in GUI mode
-if sys.stdout is None:
-    sys.stdout = open(os.devnull, "w")
-if sys.stderr is None:
-    sys.stderr = open(os.devnull, "w")
-
 # Enable MPS GPU acceleration on Mac Apple Silicon
 if platform.system() == "Darwin" and platform.processor() == "arm":
     os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"

abogen/pyqt/predownload_gui.py

@@ -21,7 +21,8 @@ from PyQt6.QtWidgets import (
 )
 from PyQt6.QtCore import QThread, pyqtSignal
 
-from abogen.constants import COLORS, VOICES_INTERNAL
+from abogen.constants import COLORS
+from abogen.tts_plugin.utils import get_voices
 from abogen.spacy_utils import SPACY_MODELS
 import abogen.hf_tracker
 
@@ -114,7 +115,7 @@ class PreDownloadWorker(QThread):
             self._voices_success = False
             return
 
-        voice_list = VOICES_INTERNAL
+        voice_list = get_voices("kokoro")
         for idx, voice in enumerate(voice_list, start=1):
             if self._cancelled:
                 self._voices_success = False
@@ -462,14 +463,14 @@ class PreDownloadDialog(QDialog):
         try:
             from huggingface_hub import try_to_load_from_cache
 
-            for voice in VOICES_INTERNAL:
+            for voice in get_voices("kokoro"):
                 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(VOICES_INTERNAL)
+            return False, list(get_voices("kokoro"))
         return (len(missing) == 0), missing
 
     def _check_kokoro_model(self) -> bool:

abogen/pyqt/queued_item.py

@@ -26,7 +26,3 @@ class QueuedItem:
     replace_all_caps: bool = False
     replace_numerals: bool = False
     fix_nonstandard_punctuation: bool = False
-    # TTS Provider fields
-    tts_provider: str = "kokoro"
-    supertonic_language: str = "en"
-    supertonic_total_steps: int = 8

abogen/pyqt/voice_formula_gui.py

@@ -28,12 +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,
-    KOKORO_LANG_TO_COUNTRY,
     COLORS,
 )
+from abogen.tts_plugin.utils import get_voices
 import re
 import platform
 from abogen.utils import get_resource_path
@@ -180,7 +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 VOICES_INTERNAL and self.voice_name[1] == "f"
+        is_female = self.voice_name in get_voices("kokoro") and self.voice_name[1] == "f"
 
         # Icons layout (flag and gender)
         icons_layout = QHBoxLayout()
@@ -190,9 +189,8 @@ class VoiceMixer(QWidget):
         )  # Center the icons horizontally
 
         # Flag icon
-        country_code = KOKORO_LANG_TO_COUNTRY.get(language_code, language_code)
         flag_icon_path = get_resource_path(
-            "abogen.assets.flags", f"{country_code}.png"
+            "abogen.assets.flags", f"{language_code}.png"
         )
         gender_icon_path = get_resource_path(
             "abogen.assets", "female.png" if is_female else "male.png"
@@ -514,8 +512,7 @@ class VoiceFormulaDialog(QDialog):
         header_row.addWidget(QLabel("Language:"))
         self.language_combo = QComboBox()
         for code, desc in LANGUAGE_OPTIONS:
-            country_code = KOKORO_LANG_TO_COUNTRY.get(code, code)
-            flag = get_resource_path("abogen.assets.flags", f"{country_code}.png")
+            flag = get_resource_path("abogen.assets.flags", f"{code}.png")
             if flag and os.path.exists(flag):
                 self.language_combo.addItem(QIcon(flag), desc, code)
             else:
@@ -775,7 +772,7 @@ class VoiceFormulaDialog(QDialog):
 
     def add_voices(self, initial_state):
         first_enabled_voice = None
-        for voice in VOICES_INTERNAL:
+        for voice in get_voices("kokoro"):
             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/shutdown.py

@@ -0,0 +1,160 @@
+"""Graceful shutdown - single module, no over-engineering."""
+from __future__ import annotations
+
+import atexit
+import gc
+import signal
+import sys
+from typing import Callable
+
+_CLEANUP_FUNCS: list[Callable[[], None]] = []
+_EXECUTED = False
+
+
+def register_cleanup(fn: Callable[[], None]) -> None:
+    """Register a cleanup function to run on shutdown."""
+    _CLEANUP_FUNCS.append(fn)
+
+
+def _run_cleanups() -> None:
+    global _EXECUTED
+    if _EXECUTED:
+        return
+    _EXECUTED = True
+    for fn in _CLEANUP_FUNCS:
+        try:
+            fn()
+        except Exception:
+            pass
+
+
+# ---- Register built-in cleanup functions ----
+
+# 1. Restore sleep prevention
+def _restore_sleep() -> None:
+    try:
+        from abogen.utils import prevent_sleep_end
+        prevent_sleep_end()
+    except Exception:
+        pass
+
+register_cleanup(_restore_sleep)
+
+# 2. Shutdown web UI ConversionService
+def _shutdown_conversion_service() -> None:
+    try:
+        from abogen.webui.service import get_service
+        svc = get_service()
+        if svc is not None:
+            svc.shutdown()
+    except Exception:
+        pass
+
+register_cleanup(_shutdown_conversion_service)
+
+# 3. Clear TTS pipelines and GPU memory
+def _cleanup_tts_pipelines() -> None:
+    # Clear web UI pipeline cache
+    try:
+        from abogen.webui.conversion_runner import _PIPELINES
+        _PIPELINES.clear()
+    except Exception:
+        pass
+
+    # Clear PyQt conversion thread voice cache
+    try:
+        from abogen.pyqt.conversion import ConversionThread
+        if hasattr(ConversionThread, "voice_cache"):
+            ConversionThread.voice_cache.clear()
+    except Exception:
+        pass
+
+    gc.collect()
+
+    # Release CUDA cache
+    try:
+        import torch
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+            torch.cuda.ipc_collect()
+    except Exception:
+        pass
+
+register_cleanup(_cleanup_tts_pipelines)
+
+# 4. Clear global voice cache
+def _clear_voice_cache() -> None:
+    try:
+        from abogen.voice_cache import clear_voice_cache
+        clear_voice_cache()
+    except Exception:
+        pass
+
+register_cleanup(_clear_voice_cache)
+
+# 5. Terminate child processes (ffmpeg, etc.)
+def _terminate_subprocesses() -> None:
+    try:
+        import psutil
+    except Exception:
+        return
+
+    try:
+        current = psutil.Process()
+        for child in current.children(recursive=True):
+            try:
+                child.terminate()
+            except Exception:
+                pass
+        gone, alive = psutil.wait_procs(current.children(recursive=True), timeout=3)
+        for proc in alive:
+            try:
+                proc.kill()
+            except Exception:
+                pass
+    except Exception:
+        pass
+
+register_cleanup(_terminate_subprocesses)
+
+
+def register_shutdown() -> None:
+    """Install process-wide shutdown hooks (atexit, signals, Qt)."""
+    if register_shutdown._registered:
+        return
+    register_shutdown._registered = True
+
+    atexit.register(_run_cleanups)
+
+    # POSIX signals
+    for sig in (signal.SIGINT, signal.SIGTERM):
+        try:
+            signal.signal(sig, _on_signal)
+        except Exception:
+            pass
+
+    # Qt hook
+    try:
+        from PyQt6.QtWidgets import QApplication
+
+        app = QApplication.instance()
+        if app is not None:
+            app.aboutToQuit.connect(_run_cleanups)
+    except Exception:
+        pass
+
+
+register_shutdown._registered = False
+
+
+def _on_signal(signum: int, _frame) -> None:
+    _run_cleanups()
+    sys.exit(0)
+
+
+def request_shutdown() -> None:
+    """Programmatically trigger cleanup (e.g., from GUI closeEvent)."""
+    _run_cleanups()
+
+
+__all__ = ["register_shutdown", "request_shutdown", "register_cleanup"]

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 VOICES_INTERNAL list (case-insensitive).
+    """Validate voice name against available voices (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.constants import VOICES_INTERNAL
+    from abogen.tts_plugin.utils import get_voices
 
     # Create case-insensitive lookup set (done once per call)
-    voice_lookup_lower = {v.lower() for v in VOICES_INTERNAL}
+    voice_lookup_lower = {v.lower() for v in get_voices("kokoro")}
     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 VOICES_INTERNAL.
+    Voice names are normalized to lowercase to match canonical voice names.
 
     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.constants import VOICES_INTERNAL
+    from abogen.tts_plugin.utils import get_voices
 
     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 VOICES_INTERNAL if v.lower() == voice_part_lower),
+                            (v for v in get_voices("kokoro") 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 VOICES_INTERNAL if v.lower() == voice_name_lower),
+                    (v for v in get_voices("kokoro") if v.lower() == voice_name_lower),
                     voice_name
                 )
             valid_markers += 1

abogen/tts_plugin/__init__.py

@@ -0,0 +1,170 @@
+"""TTS Plugin Architecture - Public API.
+
+This package defines the frozen Plugin API for the TTS Plugin Architecture.
+All public interfaces are fully defined but contain no business logic.
+
+Public modules:
+- types: Core domain value objects (AudioFormat, Duration, VoiceSelection, etc.)
+- errors: Error hierarchy (EngineError and subtypes)
+- manifest: Plugin manifest types (PluginManifest, EngineManifest, etc.)
+- engine: Engine and EngineSession protocols
+- capabilities: Optional capability interfaces (VoiceLister, PreviewGenerator, etc.)
+- host_context: HostContext dataclass
+- plugin: Plugin contract (create_engine function signature)
+- loader: Plugin discovery and loading
+- plugin_manager: Plugin management and engine creation
+- utils: Direct utility functions (get_voices, create_pipeline, etc.)
+
+Usage:
+    from abogen.tts_plugin import (
+        # Types
+        AudioFormat,
+        Duration,
+        VoiceSelection,
+        ParameterValues,
+        SynthesisRequest,
+        SynthesizedAudio,
+        EngineConfig,
+        # Errors
+        EngineError,
+        ModelNotFoundError,
+        ModelLoadError,
+        NetworkError,
+        InvalidInputError,
+        ConfigurationError,
+        CancelledError,
+        InternalError,
+        # Manifest
+        PluginManifest,
+        EngineManifest,
+        VoiceSourceManifest,
+        VoiceManifest,
+        ParameterManifest,
+        AudioFormatManifest,
+        EnumOption,
+        RequirementManifest,
+        GpuRequirement,
+        ModelManifest,
+        # Engine
+        Engine,
+        EngineSession,
+        # Capabilities
+        VoiceLister,
+        PreviewGenerator,
+        StreamingSynthesizer,
+        CancelableSession,
+        # Host Context
+        HostContext,
+        HttpClient,
+        # Plugin Manager
+        get_plugin_manager,
+        reset_plugin_manager,
+        # Utils
+        get_voices,
+        get_default_voice,
+        is_plugin_registered,
+        resolve_voice_to_plugin,
+        create_pipeline,
+    )
+"""
+
+from abogen.tts_plugin.capabilities import (
+    CancelableSession,
+    PreviewGenerator,
+    StreamingSynthesizer,
+    VoiceLister,
+)
+from abogen.tts_plugin.engine import Engine, EngineSession
+from abogen.tts_plugin.errors import (
+    CancelledError,
+    ConfigurationError,
+    EngineError,
+    InternalError,
+    InvalidInputError,
+    ModelLoadError,
+    ModelNotFoundError,
+    NetworkError,
+)
+from abogen.tts_plugin.host_context import HttpClient, HostContext
+from abogen.tts_plugin.manifest import (
+    AudioFormatManifest,
+    EngineManifest,
+    EnumOption,
+    GpuRequirement,
+    ModelManifest,
+    ParameterManifest,
+    PluginManifest,
+    RequirementManifest,
+    VoiceManifest,
+    VoiceSourceManifest,
+)
+from abogen.tts_plugin.types import (
+    AudioFormat,
+    Duration,
+    EngineConfig,
+    ParameterValues,
+    SynthesisRequest,
+    SynthesizedAudio,
+    VoiceSelection,
+)
+
+# Plugin Manager and Utils
+from abogen.tts_plugin.plugin_manager import get_plugin_manager, reset_plugin_manager
+from abogen.tts_plugin.utils import (
+    create_pipeline,
+    get_default_voice,
+    get_voices,
+    is_plugin_registered,
+    resolve_voice_to_plugin,
+)
+
+__all__ = [
+    # Types
+    "AudioFormat",
+    "Duration",
+    "VoiceSelection",
+    "ParameterValues",
+    "SynthesisRequest",
+    "SynthesizedAudio",
+    "EngineConfig",
+    # Errors
+    "EngineError",
+    "ModelNotFoundError",
+    "ModelLoadError",
+    "NetworkError",
+    "InvalidInputError",
+    "ConfigurationError",
+    "CancelledError",
+    "InternalError",
+    # Manifest
+    "PluginManifest",
+    "EngineManifest",
+    "VoiceSourceManifest",
+    "VoiceManifest",
+    "ParameterManifest",
+    "AudioFormatManifest",
+    "EnumOption",
+    "RequirementManifest",
+    "GpuRequirement",
+    "ModelManifest",
+    # Engine
+    "Engine",
+    "EngineSession",
+    # Capabilities
+    "VoiceLister",
+    "PreviewGenerator",
+    "StreamingSynthesizer",
+    "CancelableSession",
+    # Host Context
+    "HostContext",
+    "HttpClient",
+    # Plugin Manager
+    "get_plugin_manager",
+    "reset_plugin_manager",
+    # Utils
+    "get_voices",
+    "get_default_voice",
+    "is_plugin_registered",
+    "resolve_voice_to_plugin",
+    "create_pipeline",
+]

abogen/tts_plugin/capabilities.py

@@ -0,0 +1,103 @@
+"""Capability interfaces for the TTS Plugin Architecture.
+
+This module defines optional capability interfaces that engines can implement.
+Capabilities are additive; implementing new capabilities doesn't break old plugins.
+"""
+
+from __future__ import annotations
+
+from typing import Iterator, Protocol, runtime_checkable
+
+from abogen.tts_plugin.manifest import VoiceManifest
+from abogen.tts_plugin.types import SynthesisRequest, SynthesizedAudio, VoiceSelection
+
+
+@runtime_checkable
+class VoiceLister(Protocol):
+    """Protocol for listing available voices.
+
+    Engines that support voice listing should implement this interface.
+    """
+
+    def listVoices(self, sourceId: str) -> list[VoiceManifest]:
+        """List available voices for a given source.
+
+        Args:
+            sourceId: The voice source identifier.
+
+        Returns:
+            List of VoiceManifest describing available voices.
+
+        Raises:
+            EngineError: On failure.
+        """
+        ...
+
+
+@runtime_checkable
+class PreviewGenerator(Protocol):
+    """Protocol for generating voice previews.
+
+    Engines that support voice preview should implement this interface.
+    """
+
+    def generatePreview(self, voice: VoiceSelection, text: str) -> SynthesizedAudio:
+        """Generate a preview audio for a voice.
+
+        Args:
+            voice: Voice selection for the preview.
+            text: Text to use for the preview.
+
+        Returns:
+            SynthesizedAudio with the preview audio data.
+
+        Raises:
+            EngineError: On failure.
+        """
+        ...
+
+
+@runtime_checkable
+class StreamingSynthesizer(Protocol):
+    """Protocol for streaming synthesis.
+
+    Optional capability of EngineSession, not Engine.
+    Engines that support streaming synthesis should implement this interface.
+    """
+
+    def synthesizeStream(self, request: SynthesisRequest) -> Iterator[bytes]:
+        """Synthesize audio in streaming mode.
+
+        Args:
+            request: The synthesis request.
+
+        Yields:
+            Audio chunks as they become available.
+
+        Raises:
+            CancelledError: If cancel() is called during iteration.
+            EngineError: On synthesis failure.
+        """
+        ...
+        # This is a generator function; implementation will use yield
+        yield b""  # pragma: no cover
+
+
+@runtime_checkable
+class CancelableSession(Protocol):
+    """Protocol for cancellation support.
+
+    Optional capability for engines that support cancellation.
+    cancel() causes synthesize() to raise CancelledError.
+    """
+
+    def cancel(self) -> None:
+        """Cancel in-progress synthesis.
+
+        After cancellation, synthesize() raises CancelledError.
+        The session remains usable after cancellation.
+
+        Raises:
+            EngineError: If called after dispose().
+        """
+        ...

abogen/tts_plugin/engine.py

@@ -0,0 +1,95 @@
+"""Engine interfaces for the TTS Plugin Architecture.
+
+This module defines the core Engine and EngineSession protocols.
+These are the primary interfaces that plugin implementations must satisfy.
+"""
+
+from __future__ import annotations
+
+from typing import Protocol, runtime_checkable
+
+from abogen.tts_plugin.types import SynthesisRequest, SynthesizedAudio
+
+
+@runtime_checkable
+class EngineSession(Protocol):
+    """Protocol for a session that owns mutable execution state.
+
+    An EngineSession is created by Engine.createSession() and owns
+    mutable execution state isolated from other concurrent work.
+    It is NOT thread-safe.
+
+    Lifecycle:
+        1. Created by Engine.createSession()
+        2. Used for synthesis via synthesize()
+        3. Disposed via dispose()
+
+    After dispose(), all methods except dispose() raise EngineError.
+    """
+
+    def synthesize(self, request: SynthesisRequest) -> SynthesizedAudio:
+        """Synthesize audio from text.
+
+        Args:
+            request: The synthesis request containing text, voice, parameters, and format.
+
+        Returns:
+            SynthesizedAudio with the synthesized audio data.
+
+        Raises:
+            EngineError: On synthesis failure. Session remains usable after error.
+            EngineError: If called after dispose().
+        """
+        ...
+
+    def dispose(self) -> None:
+        """Release session resources.
+
+        This method is idempotent and safe to call multiple times.
+        It never raises exceptions (catches and logs internally).
+        After dispose(), all methods except dispose() raise EngineError.
+        """
+        ...
+
+
+@runtime_checkable
+class Engine(Protocol):
+    """Protocol for a TTS engine that creates sessions.
+
+    An Engine is a factory for EngineSession instances. It is stateless
+    and thread-safe for createSession().
+
+    Lifecycle:
+        1. Created via create_engine() (plugin contract)
+        2. Sessions created via createSession()
+        3. Disposed via dispose()
+
+    Thread Safety:
+        - createSession() is thread-safe and can be called from any thread.
+        - dispose() must be called after all sessions are disposed.
+        - Disposing engine while sessions are alive violates API contract.
+    """
+
+    def createSession(self) -> EngineSession:
+        """Create a new session for synthesis.
+
+        Returns:
+            A new EngineSession instance. Ownership transfers to caller.
+
+        Raises:
+            EngineError: On failure. No partially initialized session is returned.
+        """
+        ...
+
+    def dispose(self) -> None:
+        """Release engine resources.
+
+        Caller must ensure all sessions created by this engine are disposed
+        before calling dispose(). Disposing an engine while any session is
+        still alive violates the API contract; behavior is undefined.
+
+        This method is idempotent and safe to call multiple times.
+        It never raises exceptions (catches and logs internally).
+        After dispose(), all methods except dispose() raise EngineError.
+        """
+        ...

abogen/tts_plugin/errors.py

@@ -0,0 +1,62 @@
+"""Error hierarchy for the TTS Plugin Architecture.
+
+This module defines typed exceptions that engines raise.
+Engines should never raise raw exceptions; they must use EngineError or its subtypes.
+"""
+
+from __future__ import annotations
+
+
+class EngineError(Exception):
+    """Base exception for all engine errors.
+
+    All engine operations that can fail should raise EngineError or one of its subtypes.
+    After dispose(), all methods except dispose() raise EngineError.
+    """
+
+    pass
+
+
+class ModelNotFoundError(EngineError):
+    """Raised when a required model is not found."""
+
+    pass
+
+
+class ModelLoadError(EngineError):
+    """Raised when a model fails to load."""
+
+    pass
+
+
+class NetworkError(EngineError):
+    """Raised when a network operation fails."""
+
+    pass
+
+
+class InvalidInputError(EngineError):
+    """Raised when invalid input is provided to the engine."""
+
+    pass
+
+
+class ConfigurationError(EngineError):
+    """Raised when there is a configuration error."""
+
+    pass
+
+
+class CancelledError(EngineError):
+    """Raised when an operation is cancelled.
+
+    This is raised by synthesize() when cancel() is called during synthesis.
+    """
+
+    pass
+
+
+class InternalError(EngineError):
+    """Raised when an internal engine error occurs."""
+
+    pass

abogen/tts_plugin/host_context.py

@@ -0,0 +1,46 @@
+"""Host context for the TTS Plugin Architecture.
+
+This module defines the HostContext dataclass that provides minimal
+host services to plugins. It is the only interface through which
+plugins can access host functionality.
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Protocol, runtime_checkable
+
+
+@runtime_checkable
+class HttpClient(Protocol):
+    """Protocol for HTTP client provided by host.
+
+    Plugins can use this for network requests (e.g., API-based engines).
+    """
+
+    def get(self, url: str, **kwargs: object) -> object:
+        """Perform an HTTP GET request."""
+        ...
+
+    def post(self, url: str, **kwargs: object) -> object:
+        """Perform an HTTP POST request."""
+        ...
+
+
+@dataclass(frozen=True)
+class HostContext:
+    """Minimal host context provided to plugins.
+
+    Contains only essential host services. No business logic.
+
+    Attributes:
+        config_dir: Directory for API keys, preferences, and configuration.
+        logger: Logger for plugin logging.
+        http_client: HTTP client for network requests.
+    """
+
+    config_dir: Path
+    logger: logging.Logger
+    http_client: HttpClient

abogen/tts_plugin/loader.py

@@ -0,0 +1,365 @@
+"""Plugin loader infrastructure for the TTS Plugin Architecture.
+
+This module provides functionality to discover, import, validate, and load
+TTS plugins. It handles both valid and invalid plugins, providing diagnostic
+messages for errors.
+
+The loader does NOT:
+- Create Engine instances (that's the plugin's create_engine() responsibility)
+- Manage plugin lifecycle (that's the Plugin Manager's responsibility)
+- Implement any TTS engine functionality
+"""
+
+from __future__ import annotations
+
+import importlib
+import re
+import sys
+import types
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Callable
+
+from abogen.tts_plugin.manifest import ModelManifest, PluginManifest
+
+
+# Host API version for compatibility checking
+HOST_API_VERSION = "1.0"
+
+
+@dataclass(frozen=True)
+class PluginLoadError:
+    """Diagnostic information for a failed plugin load.
+
+    Attributes:
+        plugin_id: Plugin identifier if available, otherwise directory name.
+        path: Path to the plugin directory.
+        errors: List of error messages describing what went wrong.
+    """
+
+    plugin_id: str
+    path: Path
+    errors: tuple[str, ...] = field(default_factory=tuple)
+
+
+@dataclass(frozen=True)
+class PluginLoadResult:
+    """Result of loading a plugin.
+
+    Attributes:
+        success: Whether the plugin loaded successfully.
+        manifest: The plugin manifest if successful.
+        model_requirements: Model requirements if successful.
+        create_engine: The create_engine function if successful.
+        module: The plugin module if successful.
+        error: Error information if failed.
+    """
+
+    success: bool
+    manifest: PluginManifest | None = None
+    model_requirements: tuple[ModelManifest, ...] | None = None
+    create_engine: Callable[..., Any] | None = None
+    module: types.ModuleType | None = None
+    error: PluginLoadError | None = None
+
+
+def _parse_api_version(version: str) -> tuple[int, int] | None:
+    """Parse an api_version string into (major, minor) tuple.
+
+    Args:
+        version: Version string in format "MAJOR.MINOR".
+
+    Returns:
+        Tuple of (major, minor) or None if invalid format.
+    """
+    match = re.match(r"^(\d+)\.(\d+)$", version)
+    if match:
+        return int(match.group(1)), int(match.group(2))
+    return None
+
+
+def _check_api_version_compatibility(plugin_version: str) -> str | None:
+    """Check if plugin api_version is compatible with host.
+
+    Architecture spec:
+    - Format: semver (MAJOR.MINOR)
+    - Compatibility: Host rejects plugin if major version differs
+    - Minor version: backward compatible, Host accepts higher minor
+
+    Args:
+        plugin_version: Plugin's api_version string.
+
+    Returns:
+        Error message if incompatible, None if compatible.
+    """
+    plugin_ver = _parse_api_version(plugin_version)
+    if plugin_ver is None:
+        return f"Invalid api_version format: '{plugin_version}'. Expected format: MAJOR.MINOR"
+
+    host_ver = _parse_api_version(HOST_API_VERSION)
+    if host_ver is None:
+        return f"Invalid host api_version format: '{HOST_API_VERSION}'"
+
+    if plugin_ver[0] != host_ver[0]:
+        return (
+            f"api_version major mismatch: plugin={plugin_ver[0]}, host={host_ver[0]}. "
+            f"Major version must match for compatibility."
+        )
+
+    return None
+
+
+def _validate_manifest(module: types.ModuleType, plugin_dir: Path) -> list[str]:
+    """Validate that a plugin module has required exports.
+
+    Args:
+        module: The imported plugin module.
+        plugin_dir: Path to the plugin directory.
+
+    Returns:
+        List of error messages (empty if valid).
+    """
+    errors: list[str] = []
+
+    # Check PLUGIN_MANIFEST
+    manifest = getattr(module, "PLUGIN_MANIFEST", None)
+    if manifest is None:
+        errors.append("Missing PLUGIN_MANIFEST export")
+    elif not isinstance(manifest, PluginManifest):
+        errors.append(
+            f"PLUGIN_MANIFEST must be a PluginManifest instance, "
+            f"got {type(manifest).__name__}"
+        )
+
+    # Check MODEL_REQUIREMENTS
+    model_reqs = getattr(module, "MODEL_REQUIREMENTS", None)
+    if model_reqs is None:
+        errors.append("Missing MODEL_REQUIREMENTS export")
+    elif not isinstance(model_reqs, list):
+        errors.append(
+            f"MODEL_REQUIREMENTS must be a list, got {type(model_reqs).__name__}"
+        )
+    else:
+        for i, req in enumerate(model_reqs):
+            if not isinstance(req, ModelManifest):
+                errors.append(
+                    f"MODEL_REQUIREMENTS[{i}] must be a ModelManifest instance, "
+                    f"got {type(req).__name__}"
+                )
+
+    # Check create_engine
+    create_engine = getattr(module, "create_engine", None)
+    if create_engine is None:
+        errors.append("Missing create_engine export")
+    elif not callable(create_engine):
+        errors.append(
+            f"create_engine must be callable, got {type(create_engine).__name__}"
+        )
+
+    return errors
+
+
+def _validate_capabilities(manifest: PluginManifest) -> list[str]:
+    """Validate plugin capabilities.
+
+    Args:
+        manifest: The plugin manifest to validate.
+
+    Returns:
+        List of error messages (empty if valid).
+    """
+    errors: list[str] = []
+
+    # Known capabilities (can be extended)
+    known_capabilities = frozenset({
+        "voice_list",
+        "preview",
+        "voice_clone",
+        "voice_blend",
+        "streaming",
+        "cancel",
+    })
+
+    for cap in manifest.capabilities:
+        if cap not in known_capabilities:
+            errors.append(f"Unknown capability: '{cap}'")
+
+    return errors
+
+
+def _validate_api_version(manifest: PluginManifest) -> list[str]:
+    """Validate api_version compatibility.
+
+    Args:
+        manifest: The plugin manifest to validate.
+
+    Returns:
+        List of error messages (empty if valid).
+    """
+    errors: list[str] = []
+    error = _check_api_version_compatibility(manifest.api_version)
+    if error:
+        errors.append(error)
+    return errors
+
+
+def load_plugin_from_dir(plugin_dir: Path) -> PluginLoadResult:
+    """Load and validate a plugin from a directory.
+
+    The plugin directory must contain an __init__.py that exports:
+    - PLUGIN_MANIFEST: PluginManifest
+    - MODEL_REQUIREMENTS: list[ModelManifest]
+    - create_engine: Callable
+
+    Args:
+        plugin_dir: Path to the plugin directory.
+
+    Returns:
+        PluginLoadResult with success status and either plugin data or error info.
+    """
+    plugin_id = plugin_dir.name
+    errors: list[str] = []
+
+    # Check if directory exists
+    if not plugin_dir.exists():
+        return PluginLoadResult(
+            success=False,
+            error=PluginLoadError(
+                plugin_id=plugin_id,
+                path=plugin_dir,
+                errors=(f"Plugin directory does not exist: {plugin_dir}",),
+            ),
+        )
+
+    # Check for __init__.py
+    init_file = plugin_dir / "__init__.py"
+    if not init_file.exists():
+        return PluginLoadResult(
+            success=False,
+            error=PluginLoadError(
+                plugin_id=plugin_id,
+                path=plugin_dir,
+                errors=("Missing __init__.py in plugin directory",),
+            ),
+        )
+
+    # Import the module
+    module_name = f"abogen.tts_plugin._loaded.{plugin_id}"
+    try:
+        # Remove from cache if already imported (for testing)
+        if module_name in sys.modules:
+            del sys.modules[module_name]
+
+        spec = importlib.util.spec_from_file_location(
+            module_name, init_file, submodule_search_locations=[]
+        )
+        if spec is None or spec.loader is None:
+            return PluginLoadResult(
+                success=False,
+                error=PluginLoadError(
+                    plugin_id=plugin_id,
+                    path=plugin_dir,
+                    errors=(f"Failed to create module spec for {init_file}",),
+                ),
+            )
+
+        module = importlib.util.module_from_spec(spec)
+        sys.modules[module_name] = module
+        spec.loader.exec_module(module)
+    except Exception as e:
+        # Clean up module from sys.modules on import failure
+        if module_name in sys.modules:
+            del sys.modules[module_name]
+        return PluginLoadResult(
+            success=False,
+            error=PluginLoadError(
+                plugin_id=plugin_id,
+                path=plugin_dir,
+                errors=(f"Failed to import plugin module: {e}",),
+            ),
+        )
+
+    # Validate manifest
+    manifest_errors = _validate_manifest(module, plugin_dir)
+    errors.extend(manifest_errors)
+
+    # If manifest is valid, perform additional validation
+    manifest = getattr(module, "PLUGIN_MANIFEST", None)
+    if isinstance(manifest, PluginManifest):
+        # Validate api_version
+        api_errors = _validate_api_version(manifest)
+        errors.extend(api_errors)
+
+        # Validate capabilities
+        cap_errors = _validate_capabilities(manifest)
+        errors.extend(cap_errors)
+
+        # Use manifest id if available
+        plugin_id = manifest.id
+
+    # Check if any errors occurred
+    if errors:
+        # Clean up module from sys.modules
+        if module_name in sys.modules:
+            del sys.modules[module_name]
+
+        return PluginLoadResult(
+            success=False,
+            error=PluginLoadError(
+                plugin_id=plugin_id,
+                path=plugin_dir,
+                errors=tuple(errors),
+            ),
+        )
+
+    # Get MODEL_REQUIREMENTS
+    model_requirements = tuple(getattr(module, "MODEL_REQUIREMENTS", []))
+    create_engine = getattr(module, "create_engine", None)
+
+    return PluginLoadResult(
+        success=True,
+        manifest=manifest,
+        model_requirements=model_requirements,
+        create_engine=create_engine,
+        module=module,
+    )
+
+
+def discover_plugins(plugin_dirs: list[Path]) -> list[PluginLoadResult]:
+    """Discover and load plugins from multiple directories.
+
+    Args:
+        plugin_dirs: List of directories to scan for plugins.
+
+    Returns:
+        List of PluginLoadResult, one per plugin directory found.
+    """
+    results: list[PluginLoadResult] = []
+
+    for plugin_dir in plugin_dirs:
+        if not plugin_dir.exists():
+            continue
+
+        # Scan for subdirectories (each is a potential plugin)
+        for item in sorted(plugin_dir.iterdir()):
+            if item.is_dir() and not item.name.startswith("."):
+                result = load_plugin_from_dir(item)
+                results.append(result)
+
+    return results
+
+
+def load_plugin(
+    plugin_dir: Path,
+) -> PluginLoadResult:
+    """Load a single plugin from a directory.
+
+    This is the main entry point for loading a plugin.
+
+    Args:
+        plugin_dir: Path to the plugin directory.
+
+    Returns:
+        PluginLoadResult with success status and either plugin data or error info.
+    """
+    return load_plugin_from_dir(plugin_dir)

abogen/tts_plugin/manifest.py

@@ -0,0 +1,189 @@
+"""Plugin manifest types for the TTS Plugin Architecture.
+
+This module contains static metadata types that describe plugins.
+These types have no dependencies and are immutable.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+
+@dataclass(frozen=True)
+class AudioFormatManifest:
+    """Manifest describing an audio format.
+
+    Attributes:
+        mime: MIME type of the audio.
+        extension: File extension.
+    """
+
+    mime: str
+    extension: str
+
+
+@dataclass(frozen=True)
+class EnumOption:
+    """Manifest describing an enum option for a parameter.
+
+    Attributes:
+        value: The enum value.
+        label: Human-readable label.
+    """
+
+    value: str
+    label: str
+
+
+@dataclass(frozen=True)
+class ParameterManifest:
+    """Manifest describing a synthesis parameter.
+
+    Attributes:
+        id: Parameter identifier.
+        name: Human-readable name.
+        description: Parameter description.
+        type: Parameter type ("float", "int", "string", "boolean", "enum").
+        default: Default value.
+        min: Minimum value (optional, for numeric types).
+        max: Maximum value (optional, for numeric types).
+        step: Step size (optional, for numeric types).
+        options: Available options (optional, for enum type).
+        unit: Unit of measurement (optional).
+        group: Parameter group (optional).
+    """
+
+    id: str
+    name: str
+    description: str
+    type: str
+    default: Any
+    min: float | None = None
+    max: float | None = None
+    step: float | None = None
+    options: tuple[EnumOption, ...] = field(default_factory=tuple)
+    unit: str | None = None
+    group: str | None = None
+
+
+@dataclass(frozen=True)
+class VoiceManifest:
+    """Manifest describing a voice.
+
+    Attributes:
+        id: Voice identifier.
+        name: Human-readable name.
+        tags: Voice tags (e.g., language, style).
+    """
+
+    id: str
+    name: str
+    tags: tuple[str, ...] = field(default_factory=tuple)
+
+
+@dataclass(frozen=True)
+class VoiceSourceManifest:
+    """Manifest describing a voice source.
+
+    Attributes:
+        id: Voice source identifier.
+        name: Human-readable name.
+        type: Source type ("list", "speaker_id", "clone", "blend", "generate", "none").
+        config: Source-specific configuration.
+    """
+
+    id: str
+    name: str
+    type: str
+    config: Any = None
+
+
+@dataclass(frozen=True)
+class EngineManifest:
+    """Manifest describing engine capabilities.
+
+    Attributes:
+        voiceSources: Available voice sources.
+        parameters: Available synthesis parameters.
+        audioFormats: Supported audio formats.
+    """
+
+    voiceSources: tuple[VoiceSourceManifest, ...] = field(default_factory=tuple)
+    parameters: tuple[ParameterManifest, ...] = field(default_factory=tuple)
+    audioFormats: tuple[AudioFormatManifest, ...] = field(default_factory=tuple)
+
+
+@dataclass(frozen=True)
+class GpuRequirement:
+    """Manifest describing GPU requirements.
+
+    Attributes:
+        required: Whether GPU is required.
+        type: GPU type (e.g., "cuda", "rocm").
+        memory: Required GPU memory in GB.
+    """
+
+    required: bool = False
+    type: str | None = None
+    memory: float | None = None
+
+
+@dataclass(frozen=True)
+class RequirementManifest:
+    """Manifest describing plugin requirements.
+
+    Attributes:
+        gpu: GPU requirements (optional).
+        memory: Required RAM in GB (optional).
+        internet: Whether internet is required (optional).
+    """
+
+    gpu: GpuRequirement | None = None
+    memory: float | None = None
+    internet: bool | None = None
+
+
+@dataclass(frozen=True)
+class ModelManifest:
+    """Manifest describing a model requirement.
+
+    Attributes:
+        id: Model identifier.
+        name: Human-readable name.
+        size: Model size as string (e.g., "100MB", "2GB").
+    """
+
+    id: str
+    name: str
+    size: str
+
+
+@dataclass(frozen=True)
+class PluginManifest:
+    """Main manifest for a TTS plugin.
+
+    Attributes:
+        id: Plugin identifier (unique).
+        name: Human-readable name.
+        version: Plugin version.
+        api_version: API version (semver format: MAJOR.MINOR).
+        description: Plugin description.
+        author: Plugin author.
+        capabilities: List of capability identifiers.
+        requires: Plugin requirements.
+        engine: Engine manifest.
+        voices: Optional static voice catalog. None = not declared (use VoiceLister),
+            empty tuple = explicitly no static voices, non-empty = static catalog.
+    """
+
+    id: str
+    name: str
+    version: str
+    api_version: str
+    description: str
+    author: str
+    capabilities: tuple[str, ...] = field(default_factory=tuple)
+    requires: RequirementManifest = field(default_factory=RequirementManifest)
+    engine: EngineManifest = field(default_factory=EngineManifest)
+    voices: tuple[VoiceManifest, ...] | None = None

abogen/tts_plugin/plugin.py

@@ -0,0 +1,55 @@
+"""Plugin contract for the TTS Plugin Architecture.
+
+This module defines the plugin contract that all TTS plugins must implement.
+Each plugin must export:
+- PLUGIN_MANIFEST: PluginManifest instance
+- MODEL_REQUIREMENTS: list of ModelManifest instances
+- create_engine(): Factory function that creates an Engine
+
+The create_engine() function is the entry point for plugin activation.
+It must be atomic: succeed fully or raise and clean up.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Protocol, runtime_checkable
+
+from abogen.tts_plugin.engine import Engine
+from abogen.tts_plugin.host_context import HostContext
+from abogen.tts_plugin.types import EngineConfig
+
+
+@runtime_checkable
+class Plugin(Protocol):
+    """Protocol defining the plugin contract.
+
+    Every TTS plugin must implement this protocol by exporting:
+    - PLUGIN_MANIFEST: PluginManifest
+    - MODEL_REQUIREMENTS: list[ModelManifest]
+    - create_engine: Callable[[HostContext, Path | None, EngineConfig], Engine]
+    """
+
+    def create_engine(
+        self,
+        context: HostContext,
+        model_path: Path | None,
+        config: EngineConfig,
+    ) -> Engine:
+        """Create an engine instance.
+
+        This is the factory function that creates an Engine from a plugin.
+        It must be atomic: succeed fully or raise EngineError and clean up.
+
+        Args:
+            context: Host services (config dir, logger, http client).
+            model_path: Resolved model path, or None for cloud/no-model engines.
+            config: Engine initialization settings.
+
+        Returns:
+            A fully initialized Engine instance.
+
+        Raises:
+            EngineError: On failure. Cleans up partially created resources.
+        """
+        ...

abogen/tts_plugin/plugin_manager.py

@@ -0,0 +1,153 @@
+"""Plugin Manager
+
+Provides a simple interface for consumers to access TTS engines via the
+new Plugin Architecture. Discovers, loads, and manages plugins from the
+plugins directory.
+
+Usage:
+    from abogen.tts_plugin.plugin_manager import get_plugin_manager
+
+    manager = get_plugin_manager()
+    engine = manager.create_engine("kokoro", lang_code="a", device="cpu")
+    session = engine.create_session()
+    try:
+        result = session.synthesize("Hello world")
+    finally:
+        session.dispose()
+"""
+
+from typing import Any, Dict, List, Optional, Type
+
+from abogen.tts_plugin.engine import Engine, EngineSession
+from abogen.tts_plugin.manifest import PluginManifest
+from abogen.tts_plugin.types import AudioFormat
+
+
+class PluginManager:
+    """Manages TTS plugins and provides a simple interface for consumers."""
+
+    def __init__(self) -> None:
+        self._plugins: Dict[str, dict] = {}
+        self._engines: Dict[str, Engine] = {}
+        self._loaded = False
+
+    def discover(self, plugins_dir: str = "plugins") -> None:
+        """Discover and load all plugins from the given directory."""
+        import os
+        from pathlib import Path
+        from abogen.tts_plugin.loader import load_plugin_from_dir
+
+        self._plugins.clear()
+        self._engines.clear()
+
+        plugins_path = Path(plugins_dir)
+        if not plugins_path.exists():
+            self._loaded = True
+            return
+
+        for entry in plugins_path.iterdir():
+            if entry.is_dir() and (entry / "__init__.py").exists():
+                try:
+                    result = load_plugin_from_dir(entry)
+                    if result.success and result.manifest is not None:
+                        self._plugins[result.manifest.id] = {
+                            "manifest": result.manifest,
+                            "create_engine": result.create_engine,
+                            "module": result.module,
+                        }
+                except Exception as e:
+                    # Log error but continue with other plugins
+                    print(f"Warning: Failed to load plugin from {entry}: {e}")
+
+        self._loaded = True
+
+    def _ensure_loaded(self) -> None:
+        """Ensure plugins have been discovered."""
+        if not self._loaded:
+            self.discover()
+
+    def list_plugins(self) -> List[PluginManifest]:
+        """Return manifests for all loaded plugins."""
+        self._ensure_loaded()
+        return [info["manifest"] for info in self._plugins.values()]
+
+    def get_plugin(self, plugin_id: str) -> Optional[dict]:
+        """Get plugin info by ID."""
+        self._ensure_loaded()
+        return self._plugins.get(plugin_id)
+
+    def has_plugin(self, plugin_id: str) -> bool:
+        """Check if a plugin is loaded."""
+        self._ensure_loaded()
+        return plugin_id in self._plugins
+
+    def create_engine(self, plugin_id: str, **kwargs: Any) -> Engine:
+        """Create an engine instance for the given plugin.
+
+        Args:
+            plugin_id: The plugin identifier (e.g., "kokoro")
+            **kwargs: Arguments passed to the engine constructor
+
+        Returns:
+            An Engine instance
+
+        Raises:
+            KeyError: If plugin_id is not found
+            Exception: If engine creation fails
+        """
+        self._ensure_loaded()
+
+        if plugin_id not in self._plugins:
+            raise KeyError(f"Plugin not found: {plugin_id}")
+
+        plugin_info = self._plugins[plugin_id]
+        create_engine_func = plugin_info["create_engine"]
+
+        # Create engine using the plugin's factory
+        engine = create_engine_func(**kwargs)
+        return engine
+
+    def get_or_create_engine(self, plugin_id: str, **kwargs: Any) -> Engine:
+        """Get an existing engine or create a new one.
+
+        Engines are cached by plugin_id. If you need multiple instances
+        with different parameters, use create_engine() directly.
+        """
+        self._ensure_loaded()
+
+        cache_key = plugin_id
+        if cache_key in self._engines:
+            return self._engines[cache_key]
+
+        engine = self.create_engine(plugin_id, **kwargs)
+        self._engines[cache_key] = engine
+        return engine
+
+    def dispose_all(self) -> None:
+        """Dispose all cached engines."""
+        for engine in self._engines.values():
+            try:
+                engine.dispose()
+            except Exception:
+                pass  # dispose() should never raise
+        self._engines.clear()
+
+
+# Global singleton
+_manager: Optional[PluginManager] = None
+
+
+def get_plugin_manager() -> PluginManager:
+    """Get the global PluginManager instance."""
+    global _manager
+    if _manager is None:
+        _manager = PluginManager()
+    return _manager
+
+
+def reset_plugin_manager() -> None:
+    """Reset the global PluginManager (for testing)."""
+    global _manager
+    if _manager is not None:
+        _manager.dispose_all()
+    _manager = None

abogen/tts_plugin/types.py

@@ -0,0 +1,111 @@
+"""Core domain types for the TTS Plugin Architecture.
+
+This module contains immutable value objects that form the core domain.
+These types have zero dependencies and are used across the plugin system.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any, Mapping
+
+
+@dataclass(frozen=True)
+class AudioFormat:
+    """Immutable value object representing an audio format.
+
+    Attributes:
+        mime: MIME type of the audio (e.g., "audio/wav", "audio/mpeg").
+        extension: File extension (e.g., "wav", "mp3").
+    """
+
+    mime: str
+    extension: str
+
+
+@dataclass(frozen=True)
+class Duration:
+    """Immutable value object representing a time duration.
+
+    Attributes:
+        seconds: Duration in seconds.
+    """
+
+    seconds: float
+
+
+@dataclass(frozen=True)
+class VoiceSelection:
+    """Immutable value object for voice selection. Opaque to engine.
+
+    Attributes:
+        source: Voice source identifier (e.g., "builtin", "clone").
+        key: Voice key within the source.
+        payload: Optional payload for clone/blend sources.
+    """
+
+    source: str
+    key: str
+    payload: Any = None
+
+
+@dataclass(frozen=True)
+class ParameterValues:
+    """Immutable value object for synthesis parameters. Behaves like Mapping[str, Any].
+
+    Attributes:
+        values: Mapping of parameter names to their values.
+    """
+
+    values: Mapping[str, Any] = field(default_factory=dict)
+
+
+@dataclass(frozen=True)
+class SynthesisRequest:
+    """Immutable value object for a synthesis request.
+
+    Attributes:
+        text: Text to synthesize.
+        voice: Voice selection.
+        parameters: Synthesis parameters.
+        format: Desired audio output format.
+    """
+
+    text: str
+    voice: VoiceSelection
+    parameters: ParameterValues
+    format: AudioFormat
+
+
+@dataclass(frozen=True)
+class SynthesizedAudio:
+    """Immutable value object for synthesized audio result.
+
+    Attributes:
+        data: Raw audio bytes.
+        format: Audio format of the result.
+        duration: Duration of the audio.
+    """
+
+    data: bytes
+    format: AudioFormat
+    duration: Duration
+
+
+@dataclass(frozen=True)
+class EngineConfig:
+    """Immutable configuration of an Engine instance.
+
+    Contains parameters that define how a particular Engine instance is
+    created and that remain constant throughout the lifetime of that Engine.
+
+    Plugin implementations may ignore fields that are not applicable to them.
+
+    Attributes:
+        device: Device to use (e.g., "cpu", "cuda:0").
+        lang_code: Language code for the engine (e.g., "a" for Kokoro English).
+            Plugins that do not require a language code ignore this field.
+    """
+
+    device: str = "cpu"
+    lang_code: str = "a"

abogen/tts_plugin/utils.py

@@ -0,0 +1,235 @@
+"""TTS Plugin Architecture — direct utility functions.
+
+Provides helpers that replace the former compatibility adapter by
+calling the Plugin Manager directly.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Iterator
+
+import numpy as np
+
+from abogen.tts_plugin.plugin_manager import get_plugin_manager
+
+
+def get_voices(plugin_id: str) -> tuple[str, ...]:
+    """Return the voice-id tuple for *plugin_id*.
+
+    Uses the official Plugin Architecture: PluginManager → Engine → VoiceLister.
+    First checks plugin manifest for static voice catalog.
+    """
+    import logging
+    import tempfile
+    from pathlib import Path
+
+    from abogen.tts_plugin.host_context import HostContext
+    from abogen.tts_plugin.types import EngineConfig
+
+    manager = get_plugin_manager()
+    if not manager.has_plugin(plugin_id):
+        return ()
+
+    # Check manifest for static voice catalog
+    plugin_info = manager.get_plugin(plugin_id)
+    if plugin_info is not None:
+        manifest = plugin_info.get("manifest")
+        if manifest is not None and manifest.voices is not None:
+            return tuple(v.id for v in manifest.voices)
+
+    ctx = HostContext(
+        config_dir=Path(tempfile.gettempdir()),
+        logger=logging.getLogger(f"abogen.utils.{plugin_id}"),
+        http_client=type("_StubHttpClient", (), {
+            "get": staticmethod(lambda url, **kw: None),
+            "post": staticmethod(lambda url, **kw: None),
+        })(),
+    )
+
+    try:
+        engine = manager.create_engine(
+            plugin_id,
+            context=ctx,
+            model_path=None,
+            config=EngineConfig(device="cpu"),
+        )
+    except Exception:
+        return ()
+
+    try:
+        from abogen.tts_plugin.capabilities import VoiceLister
+
+        if isinstance(engine, VoiceLister):
+            manifests = engine.listVoices("builtin")
+            return tuple(v.id for v in manifests)
+        return ()
+    except Exception:
+        return ()
+    finally:
+        engine.dispose()
+
+
+def get_default_voice(plugin_id: str, fallback: str = "") -> str:
+    """Return the first voice of *plugin_id*, or *fallback*."""
+    voices = get_voices(plugin_id)
+    return voices[0] if voices else fallback
+
+
+def is_plugin_registered(plugin_id: str) -> bool:
+    """Check whether *plugin_id* is loaded by the Plugin Manager."""
+    return get_plugin_manager().has_plugin(plugin_id)
+
+
+def resolve_voice_to_plugin(spec: str, fallback: str = "kokoro") -> str:
+    """Determine which plugin owns the given voice specification.
+
+    Resolution rules:
+    1. Empty spec -> fallback
+    2. Kokoro formula (contains '*' or '+') -> "kokoro"
+    3. Exact voice-id match against loaded plugins -> plugin 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()
+    manager = get_plugin_manager()
+
+    for manifest in manager.list_plugins():
+        for voice_source in manifest.engine.voiceSources:
+            if voice_source.type == "list" and isinstance(voice_source.config, dict):
+                try:
+                    engine = manager.create_engine(manifest.id)
+                    try:
+                        if hasattr(engine, "listVoices"):
+                            voice_manifests = engine.listVoices(voice_source.id)
+                            voice_ids = [v.id.upper() for v in voice_manifests]
+                            if upper in voice_ids:
+                                return manifest.id
+                    finally:
+                        engine.dispose()
+                except Exception:
+                    continue
+
+    return fallback
+
+
+class Pipeline:
+    """Callable wrapper around Engine / EngineSession.
+
+    Presents the same interface that old callers expect::
+
+        pipeline = create_pipeline("kokoro", lang_code="a", device="cpu")
+        for segment in pipeline(text, voice="af_nova", speed=1.0):
+            audio = segment.audio
+    """
+
+    def __init__(self, engine: Any, **engine_kwargs: Any) -> None:
+        self._engine = engine
+        self._engine_kwargs = engine_kwargs
+        self._session: Any = None
+
+    def _ensure_session(self) -> Any:
+        if self._session is None:
+            self._session = self._engine.createSession()
+        return self._session
+
+    def __call__(
+        self,
+        text: str,
+        voice: str = "default",
+        speed: float = 1.0,
+        split_pattern: str | None = None,
+        **kwargs: Any,
+    ) -> Iterator[Any]:
+        from abogen.tts_plugin.types import (
+            AudioFormat,
+            ParameterValues,
+            SynthesisRequest,
+            VoiceSelection,
+        )
+
+        session = self._ensure_session()
+
+        params: dict[str, Any] = {"speed": speed}
+        if split_pattern is not None:
+            params["split_pattern"] = split_pattern
+        params.update(kwargs)
+
+        request = SynthesisRequest(
+            text=text,
+            voice=VoiceSelection(source="builtin", key=voice),
+            parameters=ParameterValues(values=params),
+            format=AudioFormat(mime="audio/wav", extension="wav"),
+        )
+
+        result = session.synthesize(request)
+        audio_array = np.frombuffer(result.data, dtype=np.float32)
+
+        from dataclasses import dataclass
+
+        @dataclass
+        class Segment:
+            graphemes: str
+            audio: np.ndarray
+
+        yield Segment(graphemes=text, audio=audio_array)
+
+    def dispose(self) -> None:
+        if self._session is not None:
+            try:
+                self._session.dispose()
+            except Exception:
+                pass
+            self._session = None
+
+    def __del__(self) -> None:
+        self.dispose()
+
+
+def create_pipeline(
+    plugin_id: str,
+    *,
+    lang_code: str = "a",
+    device: str = "cpu",
+) -> Pipeline:
+    """Create a callable TTS pipeline via the Plugin Architecture.
+
+    Builds a proper HostContext and EngineConfig, then delegates to the
+    PluginManager to create the engine. Returns a :class:`Pipeline` whose
+    ``__call__`` interface matches the callable protocol used by consumers.
+
+    Args:
+        plugin_id: Plugin identifier (e.g., "kokoro", "supertonic").
+        lang_code: Language code for the engine.
+        device: Device to use (e.g., "cpu", "cuda:0").
+
+    Returns:
+        A callable Pipeline instance.
+    """
+    import logging
+    import tempfile
+    from pathlib import Path
+
+    from abogen.tts_plugin.host_context import HostContext
+    from abogen.tts_plugin.types import EngineConfig
+
+    manager = get_plugin_manager()
+
+    ctx = HostContext(
+        config_dir=Path(tempfile.gettempdir()),
+        logger=logging.getLogger(f"abogen.pipeline.{plugin_id}"),
+        http_client=type("_StubHttpClient", (), {
+            "get": staticmethod(lambda url, **kw: None),
+            "post": staticmethod(lambda url, **kw: None),
+        })(),
+    )
+
+    config = EngineConfig(device=device, lang_code=lang_code)
+
+    engine = manager.create_engine(plugin_id, context=ctx, model_path=None, config=config)
+    return Pipeline(engine)

abogen/utils.py

@@ -529,21 +529,20 @@ 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):
+    def __init__(self, callback, lang_code="a", device="cpu"):
         super().__init__()
         self.callback = callback
+        self.lang_code = lang_code
+        self.device = device
 
     def run(self):
         try:
-            np_module, kpipeline_class = load_numpy_kpipeline()
-            self.callback(np_module, kpipeline_class, None)
+            from abogen.tts_plugin.utils import create_pipeline
+
+            backend = create_pipeline(
+                "kokoro", lang_code=self.lang_code, device=self.device
+            )
+            self.callback(backend, None)
         except Exception as e:
-            self.callback(None, None, str(e))
+            self.callback(None, str(e))

abogen/voice_cache.py

@@ -17,7 +17,7 @@ if LocalEntryNotFoundError is None:  # pragma: no cover - fallback for tests
         pass
 
 
-from abogen.constants import VOICES_INTERNAL
+from abogen.tts_plugin.utils import get_voices
 
 _CACHE_LOCK = threading.Lock()
 _CACHED_VOICES: Set[str] = set()
@@ -26,8 +26,9 @@ _BOOTSTRAPPED = False
 
 
 def _normalize_targets(voices: Optional[Iterable[str]]) -> Set[str]:
+    kokoro_voices = get_voices("kokoro")
     if not voices:
-        return set(VOICES_INTERNAL)
+        return set(kokoro_voices)
     normalized: Set[str] = set()
     for voice in voices:
         if not voice:
@@ -35,7 +36,7 @@ def _normalize_targets(voices: Optional[Iterable[str]]) -> Set[str]:
         voice_id = str(voice).strip()
         if not voice_id:
             continue
-        if voice_id in VOICES_INTERNAL:
+        if voice_id in kokoro_voices:
             normalized.add(voice_id)
     return normalized
 
@@ -143,3 +144,11 @@ def _ensure_single_voice_asset(
 
     hf_hub_download(resume_download=True, **common_kwargs)
     return True
+
+
+def clear_voice_cache() -> None:
+    """Clear the in‑process voice cache (used during shutdown)."""
+    with _CACHE_LOCK:
+        _CACHED_VOICES.clear()
+    global _BOOTSTRAPPED
+    _BOOTSTRAPPED = False

abogen/voice_formulas.py

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

abogen/voice_metadata.py

@@ -0,0 +1,33 @@
+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

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

abogen/webui/Dockerfile

@@ -2,7 +2,6 @@ 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
 
@@ -27,22 +26,22 @@ RUN python3 -m venv "$VIRTUAL_ENV"
 WORKDIR /app
 
 COPY pyproject.toml README.md ./
-COPY abogen ./abogen
-
-RUN pip install --upgrade pip \
+RUN pip install uv \
   && if [ -n "$TORCH_VERSION" ]; then \
-       pip install torch=="$TORCH_VERSION" torchvision=="$TORCH_VERSION" torchaudio=="$TORCH_VERSION" --index-url "$TORCH_INDEX_URL"; \
+       uv pip install --system torch=="$TORCH_VERSION" torchvision=="$TORCH_VERSION" torchaudio=="$TORCH_VERSION" --index-url "$TORCH_INDEX_URL"; \
      else \
-       pip install torch torchvision torchaudio --index-url "$TORCH_INDEX_URL"; \
+       uv pip install --system torch torchvision torchaudio --index-url "$TORCH_INDEX_URL"; \
      fi \
-  && pip install --no-cache-dir . \
+  && 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 \
-  && pip install --no-cache-dir "mutagen>=1.47.0"
+  && uv pip install --system "mutagen>=1.47.0"
+
+COPY abogen ./abogen
 
 # 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 \
-      pip install --no-cache-dir onnxruntime-gpu; \
+      uv pip install --system onnxruntime-gpu; \
     fi
 
 ENV ABOGEN_HOST=0.0.0.0 \

abogen/webui/app.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import atexit
 import logging
 import os
 from pathlib import Path
@@ -8,6 +7,8 @@ from typing import Any, Optional
 
 from flask import Flask
 
+from abogen import shutdown  # noqa: F401
+shutdown.register_shutdown()
 from abogen.utils import get_user_cache_path, get_user_output_path, get_user_settings_dir
 
 from .conversion_runner import run_conversion_job
@@ -83,6 +84,12 @@ def create_app(config: Optional[dict[str, Any]] = None) -> Flask:
         "UPLOAD_FOLDER": str(uploads_dir),
         "OUTPUT_FOLDER": str(outputs_dir),
         "MAX_CONTENT_LENGTH": 1024 * 1024 * 400,  # 400 MB uploads
+        # Large books can submit four form fields per chapter. Werkzeug's
+        # defaults reject those requests before the wizard route can process
+        # them, even though the encoded payload is much smaller than the upload
+        # limit above.
+        "MAX_FORM_MEMORY_SIZE": 10 * 1024 * 1024,
+        "MAX_FORM_PARTS": 10_000,
     }
     if config:
         base_config.update(config)
@@ -113,8 +120,6 @@ def create_app(config: Optional[dict[str, Any]] = None) -> Flask:
     app.register_blueprint(books_bp, url_prefix="/find-books")
     app.register_blueprint(api_bp, url_prefix="/api")
 
-    atexit.register(service.shutdown)
-
     global _access_log_filter_attached
     if not _access_log_filter_attached:
         logging.getLogger("werkzeug").addFilter(_SuppressSuccessfulAccessFilter())

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.utils import load_numpy_kpipeline
+from abogen.tts_plugin.utils import create_pipeline
 
 
 _MARKER_RE = re.compile(re.escape(MARKER_PREFIX) + r"(?P<code>[A-Z0-9_]+)" + re.escape(MARKER_SUFFIX))
@@ -45,8 +45,7 @@ def _load_pipeline(language: str, use_gpu: bool) -> Any:
     device = "cpu"
     if use_gpu:
         device = _select_device()
-    _np, KPipeline = load_numpy_kpipeline()
-    return KPipeline(lang_code=language, repo_id="hexgrad/Kokoro-82M", device=device)
+    return create_pipeline("kokoro", lang_code=language, device=device)
 
 
 def _extract_cases_from_text(text: str) -> List[Tuple[str, str]]:
@@ -248,4 +247,8 @@ def run_debug_tts_wavs(
         "sample_rate": SAMPLE_RATE,
     }
     (run_dir / "manifest.json").write_text(json.dumps(manifest, indent=2), encoding="utf-8")
+    try:
+        pipeline.dispose()
+    except Exception:
+        pass
     return manifest

abogen/webui/routes/api.py

@@ -34,6 +34,7 @@ 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_plugin.utils import is_plugin_registered
 from abogen.integrations.audiobookshelf import AudiobookshelfClient, AudiobookshelfConfig
 from abogen.integrations.calibre_opds import (
     CalibreOPDSClient,
@@ -63,7 +64,7 @@ def api_save_voice_profile() -> ResponseReturnValue:
     if profile is None:
         # Speaker Studio payload format
         provider = str(payload.get("provider") or "kokoro").strip().lower()
-        if provider not in {"kokoro", "supertonic"}:
+        if not is_plugin_registered(provider):
             provider = "kokoro"
         if provider == "supertonic":
             profile = {
@@ -162,7 +163,7 @@ def api_voice_profiles_preview() -> ResponseReturnValue:
     formula = str(payload.get("formula") or "").strip()
     profile_name = str(payload.get("profile") or "").strip()
     provider = str(payload.get("tts_provider") or payload.get("provider") or "").strip().lower() or None
-    supertonic_total_steps = int(payload.get("supertonic_total_steps") or payload.get("total_steps") or settings.get("supertonic_total_steps") or 8)
+    supertonic_total_steps = int(payload.get("supertonic_total_steps") or payload.get("total_steps") or settings.get("supertonic_total_steps") or 5)
 
     voice_spec = ""
     resolved_provider = provider or "kokoro"
@@ -224,13 +225,13 @@ def api_speaker_preview() -> ResponseReturnValue:
     speed_value = payload.get("speed")
     speed = coerce_float(speed_value, 1.0)
     tts_provider = str(payload.get("tts_provider") or "").strip().lower()
-    supertonic_total_steps = int(payload.get("supertonic_total_steps") or 8)
+    supertonic_total_steps = int(payload.get("supertonic_total_steps") or 5)
     
     settings = load_settings()
     use_gpu = settings.get("use_gpu", False)
 
     base_spec, speaker_name = split_profile_spec(voice)
-    resolved_provider = tts_provider if tts_provider in {"kokoro", "supertonic"} else ""
+    resolved_provider = tts_provider if is_plugin_registered(tts_provider) else ""
 
     if speaker_name:
         entry = normalize_profile_entry(load_profiles().get(speaker_name))
@@ -269,7 +270,7 @@ def api_speaker_preview() -> ResponseReturnValue:
             use_gpu=use_gpu
             ,
             tts_provider=resolved_provider,
-            supertonic_total_steps=supertonic_total_steps or int(settings.get("supertonic_total_steps") or 8),
+            supertonic_total_steps=supertonic_total_steps or int(settings.get("supertonic_total_steps") or 5),
             pronunciation_overrides=pronunciation_overrides,
             manual_overrides=manual_overrides,
             speakers=speakers,

abogen/webui/routes/settings.py

@@ -43,12 +43,8 @@ def update_settings() -> ResponseReturnValue:
     current["language"] = (form.get("language") or "en").strip()
     current["default_speaker"] = (form.get("default_speaker") or "").strip()
     current["default_voice"] = (form.get("default_voice") or "").strip()
-    provider = str(form.get("tts_provider") or "kokoro").strip().lower()
-    if provider in {"kokoro", "supertonic"}:
-        current["tts_provider"] = provider
     try:
-        total_steps = int(form.get("supertonic_total_steps", current.get("supertonic_total_steps", 8)))
-        current["supertonic_total_steps"] = max(2, min(15, total_steps))
+        current["supertonic_total_steps"] = max(2, min(15, int(form.get("supertonic_total_steps", current.get("supertonic_total_steps", 5)))))
     except (TypeError, ValueError):
         pass
     try:

abogen/webui/routes/utils/form.py

@@ -7,6 +7,7 @@ from flask.typing import ResponseReturnValue
 
 from abogen.webui.service import PendingJob, JobStatus
 from abogen.webui.routes.utils.service import get_service
+from abogen.tts_plugin.utils import is_plugin_registered
 from abogen.webui.routes.utils.settings import (
     load_settings,
     coerce_bool,
@@ -32,7 +33,7 @@ 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.constants import VOICES_INTERNAL
+from abogen.tts_plugin.utils import get_default_voice
 from abogen.speaker_configs import get_config
 from abogen.kokoro_text_normalization import normalize_roman_numeral_titles
 from dataclasses import dataclass
@@ -577,9 +578,9 @@ def apply_book_step_form(
     # NOTE: Do not auto-set a global TTS provider at the book level based on the
     # narrator defaults. Provider is resolved per-speaker/per-chunk from the voice
     # spec (e.g. "speaker:Name" for saved speakers, or a Kokoro mix formula).
-    # This enables mixed-provider conversions (e.g. narrator=Supertonic, characters=Kokoro).
+    # This enables mixed-provider conversions (e.g. narrator=SuperTonic, characters=Kokoro).
     provider_value = str(form.get("tts_provider") or "").strip().lower()
-    if provider_value in {"kokoro", "supertonic"}:
+    if is_plugin_registered(provider_value):
         pending.tts_provider = provider_value
 
     # Determine the base speaker selection (saved speaker ref or raw voice).
@@ -616,8 +617,8 @@ def apply_book_step_form(
         custom_formula = ""
 
     base_voice_spec = resolved_default_voice or narrator_voice_raw
-    if not base_voice_spec and VOICES_INTERNAL:
-        base_voice_spec = VOICES_INTERNAL[0]
+    if not base_voice_spec:
+        base_voice_spec = get_default_voice("kokoro")
 
     voice_choice, resolved_language, selected_profile = resolve_voice_choice(
         pending.language,
@@ -796,8 +797,8 @@ 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 and VOICES_INTERNAL:
-        base_voice = VOICES_INTERNAL[0]
+    if not base_voice:
+        base_voice = get_default_voice("kokoro")
     selected_speaker_config = (form.get("speaker_config") or "").strip()
     speaker_config_payload = get_config(selected_speaker_config) if selected_speaker_config else None
 
@@ -913,15 +914,6 @@ def build_pending_job_from_extraction(
         else:
             normalization_overrides[key] = default_val
 
-    provider_value = str(form.get("tts_provider") or "").strip().lower()
-    if provider_value not in {"kokoro", "supertonic"}:
-        provider_value = settings.get("tts_provider", "kokoro")
-    try:
-        total_steps = int(form.get("supertonic_total_steps", settings.get("supertonic_total_steps", 8)))
-        supertonic_steps = max(2, min(15, total_steps))
-    except (TypeError, ValueError):
-        supertonic_steps = int(settings.get("supertonic_total_steps", 8))
-
     pending = PendingJob(
         id=uuid.uuid4().hex,
         original_filename=original_name,
@@ -937,8 +929,6 @@ def build_pending_job_from_extraction(
         replace_single_newlines=replace_single_newlines,
         subtitle_format=subtitle_format,
         total_characters=total_chars,
-        tts_provider=provider_value,
-        supertonic_total_steps=supertonic_steps,
         save_chapters_separately=save_chapters_separately,
         merge_chapters_at_end=merge_chapters_at_end,
         separate_chapters_format=separate_chapters_format,