Update README.md to include Supertonic TTS in PyQt6 Desktop GUI features - Supertonic 3 support

Fixes #163

.github/FUNDING.yml

@@ -0,0 +1,15 @@
+# These are supported funding model platforms
+
+github: [jborza, jeremiahsb, mohangk]
+patreon: # Replace with a single Patreon username
+open_collective: # Replace with a single Open Collective username
+ko_fi: # Replace with a single Ko-fi username
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: # Replace with a single Liberapay username
+issuehunt: # Replace with a single IssueHunt username
+lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
+polar: # Replace with a single Polar username
+buy_me_a_coffee: # Replace with a single Buy Me a Coffee username
+thanks_dev: # Replace with a single thanks.dev username
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

.gitignore

@@ -38,3 +38,4 @@ dist/
 .old/
 test_assets/
 dev_notes/
+.claude/

README.md

@@ -38,6 +38,8 @@ 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
@@ -65,6 +67,9 @@ 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.
 
@@ -173,7 +178,7 @@ Abogen offers **two interfaces**, but currently they have different feature sets
 
 | Command | Interface | Features |
 |---------|-----------|----------|
-| `abogen` | PyQt6 Desktop GUI | Stable core features |
+| `abogen` | PyQt6 Desktop GUI | Stable core features + **Supertonic TTS**|
 | `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.
@@ -407,18 +412,18 @@ 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
+```nginx
-  proxy_set_header Host $host;
+proxy_set_header Host $host;
-  proxy_set_header X-Real-IP $remote_addr;
+proxy_set_header X-Real-IP $remote_addr;
-  proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
-  proxy_set_header X-Forwarded-Proto $scheme;
+proxy_set_header X-Forwarded-Proto $scheme;
-  proxy_set_header X-Forwarded-Host $host;
+proxy_set_header X-Forwarded-Host $host;
-  proxy_set_header X-Forwarded-Port $server_port;
+proxy_set_header X-Forwarded-Port $server_port;
-  proxy_set_header Authorization $http_authorization;
+proxy_set_header Authorization $http_authorization;
-  client_max_body_size 5g;
+client_max_body_size 5g;
-  proxy_read_timeout 300s;
+proxy_read_timeout 300s;
-  proxy_connect_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).
 6. If you publish Audiobookshelf under a path prefix (for example `/abs`), add a **Custom Location** with `Location: /abs/` and set the **Forward Path** to `/`. That rewrite strips the `/abs` prefix before traffic reaches Audiobookshelf so `/abs/api/...` on the internet becomes `/api/...` on the backend. Use the same prefixed URL in Abogen’s “Base URL” field.

WINDOWS_INSTALL.bat

@@ -323,6 +323,13 @@ 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.
     )
@@ -348,6 +355,13 @@ 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/VERSION

@@ -1 +1 @@
-1.3.0
+1.3.1

abogen/book_parser.py

@@ -915,7 +915,11 @@ class EpubParser(BaseBookParser):
 
             if slice_html.strip():
                 slice_soup = BeautifulSoup(slice_html, "html.parser")
-                for tag in slice_soup.find_all(["p", "div"]):
+
+                # Add line breaks after block-level elements to ensure pauses in speech
+                for tag in slice_soup.find_all(
+                    ["p", "div", "h1", "h2", "h3", "h4", "h5", "h6", "li", "blockquote"]
+                ):
                     tag.append("\n\n")
 
                 for ol in slice_soup.find_all("ol"):

abogen/constants.py

@@ -29,6 +29,57 @@ 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",

abogen/conversion.py

@@ -1,16 +0,0 @@
-"""Backwards-compatible re-export of conversion module.
-
-The PyQt-based implementation lives in abogen.pyqt.conversion.
-The web-based implementation is in abogen.webui.conversion_runner.
-"""
-
-from __future__ import annotations
-
-# Re-export PyQt conversion classes for backwards compatibility
-from abogen.pyqt.conversion import (  # noqa: F401
-    ConversionThread,
-    VoicePreviewThread,
-    PlayAudioThread,
-)
-
-__all__ = ["ConversionThread", "VoicePreviewThread", "PlayAudioThread"]

abogen/pyqt/conversion.py

@@ -20,6 +20,7 @@ from abogen.constants import (
     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
 import abogen.hf_tracker as hf_tracker
 import static_ffmpeg
@@ -42,6 +43,10 @@ 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,
 )
 
 class CountdownDialog(QDialog):
@@ -262,7 +267,10 @@ class ConversionThread(QThread):
         use_gpu=True,
         from_queue=False,
         save_base_path=None,
-    ):  # Add use_gpu parameter
+        tts_provider="kokoro",
+        supertonic_language="en",
+        supertonic_total_steps=8,
+    ):
         super().__init__()
         self._chapter_options_event = threading.Event()
         self._timestamp_response_event = threading.Event()
@@ -272,6 +280,10 @@ class ConversionThread(QThread):
         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
@@ -296,6 +308,31 @@ class ConversionThread(QThread):
         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.voice_cache = {}  # Cache for loaded voices
+
+    def load_voice_cached(self, voice_name, tts):
+        """Load voice with caching to avoid reloading same voice.
+
+        Args:
+            voice_name: Voice name or formula string
+            tts: TTS pipeline instance
+
+        Returns:
+            Loaded voice tensor or voice name string
+        """
+        # Check cache first
+        if voice_name in self.voice_cache:
+            return self.voice_cache[voice_name]
+
+        # Load voice
+        if "*" in voice_name:
+            loaded_voice = get_new_voice(tts, voice_name, self.use_gpu)
+        else:
+            loaded_voice = voice_name
+
+        # Cache it
+        self.voice_cache[voice_name] = loaded_voice
+        return loaded_voice
 
     def _stream_audio_in_chunks(
         self, segments, process_func, progress_prefix="Processing"
@@ -398,6 +435,10 @@ 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(
@@ -470,9 +511,16 @@ class ConversionThread(QThread):
             else:
                 device = "cpu"
 
-            tts = self.KPipeline(
+            if self.tts_provider == "supertonic":
-                lang_code=self.lang_code, repo_id="hexgrad/Kokoro-82M", device=device
+                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
@@ -524,6 +572,26 @@ class ConversionThread(QThread):
             # Clean up text using utility function
             text = clean_text(text)
 
+            # Apply word substitutions if enabled
+            if getattr(self, "word_substitutions_enabled", False):
+                from abogen.word_substitution import apply_word_substitutions
+
+                self.log_updated.emit("Applying word substitutions...")
+
+                substitutions_list = getattr(self, "word_substitutions_list", "")
+                case_sensitive = getattr(self, "case_sensitive_substitutions", False)
+                replace_caps = getattr(self, "replace_all_caps", False)
+                replace_nums = getattr(self, "replace_numerals", False)
+                fix_punct = getattr(self, "fix_nonstandard_punctuation", False)
+
+                text = apply_word_substitutions(
+                    text,
+                    substitutions_list,
+                    case_sensitive,
+                    replace_caps,
+                    replace_nums,
+                    fix_punct,
+                )
 
             # --- Chapter splitting logic ---
             # Use pre-compiled pattern for better performance
@@ -550,6 +618,42 @@ class ConversionThread(QThread):
                 chapters = [("text", text)]
             total_chapters = len(chapters)
 
+            # --- Voice marker splitting logic ---
+            # Split each chapter by voice markers, preserving voice state across chapters
+            chapters_with_voices = []
+            current_voice = self.voice  # Start with default voice
+            total_valid_markers = 0
+            total_invalid_markers = 0
+
+            for chapter_name, chapter_text in chapters:
+                # Use current_voice as the starting voice for this chapter
+                voice_segments, last_voice, valid_count, invalid_count = split_text_by_voice_markers(chapter_text, current_voice)
+                chapters_with_voices.append((chapter_name, voice_segments))
+
+                # Update current_voice so next chapter continues with this voice
+                current_voice = last_voice
+
+                # Track total valid/invalid markers
+                total_valid_markers += valid_count
+                total_invalid_markers += invalid_count
+
+            # Log voice marker information with accurate counts
+            total_markers = total_valid_markers + total_invalid_markers
+            if total_markers > 0:
+                if total_invalid_markers == 0:
+                    # All markers were valid
+                    self.log_updated.emit(
+                        (f"\nDetected {total_markers} voice marker(s) - all valid", "grey")
+                    )
+                else:
+                    # Some markers were invalid
+                    self.log_updated.emit(
+                        (f"\nDetected {total_markers} voice marker(s) - {total_valid_markers} valid, {total_invalid_markers} invalid (using previous voice)", "orange")
+                    )
+
+            # Replace chapters with the new structure
+            chapters = chapters_with_voices
+
             # For text files with chapters, prompt user for options if not already set
             is_txt_file = not self.is_direct_text and (
                 self.file_name.lower().endswith(".txt")
@@ -662,7 +766,7 @@ class ConversionThread(QThread):
                 merged_out_path = f"{base_filepath_no_ext}.{self.output_format}"
                 subtitle_entries = []
                 current_time = 0.0
-                rate = 24000
+                rate = self.sample_rate
                 subtitle_mode = self.subtitle_mode
                 self.etr_start_time = time.time()
                 self.processed_char_count = 0
@@ -678,7 +782,7 @@ class ConversionThread(QThread):
                     merged_out_file = sf.SoundFile(
                         merged_out_path,
                         "w",
-                        samplerate=24000,
+                        samplerate=self.sample_rate,
                         channels=1,
                         format=self.output_format,
                     )
@@ -694,62 +798,18 @@ class ConversionThread(QThread):
                     # Prepare ffmpeg command for m4b output
                     cmd = [
                         "ffmpeg",
-                        "-y",
+                            "-y",
-                        "-thread_queue_size",
+                            "-thread_queue_size",
-                        "32768",
+                            "32768",
-                        "-f",
+                            "-f",
-                        "f32le",
+                            "f32le",
-                        "-ar",
+                            "-ar",
-                        "24000",
+                            str(self.sample_rate),
-                        "-ac",
+                            "-ac",
-                        "1",
+                            "1",
-                        "-i",
+                            "-i",
-                        "pipe:0",
+                            "pipe:0",
-                    ]
-                    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",
                         ]
-                    )
-                    cmd += metadata_options
-                    cmd.append(merged_out_path)
-                    ffmpeg_proc = create_process(cmd, stdin=subprocess.PIPE, text=False)
-                elif self.output_format == "opus":
-                    static_ffmpeg.add_paths()
-                    cmd = [
-                        "ffmpeg",
-                        "-y",
-                        "-thread_queue_size",
-                        "32768",
-                        "-f",
-                        "f32le",
-                        "-ar",
-                        "24000",
-                        "-ac",
-                        "1",
-                        "-i",
-                        "pipe:0",
-                    ]
                     cmd.extend(["-c:a", "libopus", "-b:a", "24000"])
                     cmd.append(merged_out_path)
                     ffmpeg_proc = create_process(cmd, stdin=subprocess.PIPE, text=False)
@@ -831,7 +891,7 @@ class ConversionThread(QThread):
                 merged_out_path = None
                 subtitle_entries = []
                 current_time = 0.0
-                rate = 24000
+                rate = self.sample_rate
                 subtitle_mode = self.subtitle_mode
                 self.etr_start_time = time.time()
                 self.processed_char_count = 0
@@ -842,7 +902,7 @@ class ConversionThread(QThread):
                 ]
                 srt_index = 1  # SRT numbering fix for chapter-only mode
             # Instead of processing the whole text, process by chapter
-            for chapter_idx, (chapter_name, chapter_text) in enumerate(chapters, 1):
+            for chapter_idx, (chapter_name, voice_segments) in enumerate(chapters, 1):
                 chapter_out_path = None
                 chapter_out_file = None
                 chapter_ffmpeg_proc = None
@@ -862,11 +922,6 @@ class ConversionThread(QThread):
                 if merge_chapters_at_end:
                     chapter_time["start"] = current_time
 
-                # Check if the voice is a formula and load it if necessary
-                if "*" in self.voice:
-                    loaded_voice = get_new_voice(tts, self.voice, self.use_gpu)
-                else:
-                    loaded_voice = self.voice
                 # Prepare per-chapter output file if needed
                 if save_chapters_separately and total_chapters > 1:
                     # First pass: keep alphanumeric, spaces, hyphens, and underscores
@@ -889,7 +944,7 @@ class ConversionThread(QThread):
                         chapter_out_file = sf.SoundFile(
                             chapter_out_path,
                             "w",
-                            samplerate=24000,
+                            samplerate=self.sample_rate,
                             channels=1,
                             format=separate_chapters_format,
                         )
@@ -904,7 +959,7 @@ class ConversionThread(QThread):
                             "-f",
                             "f32le",
                             "-ar",
-                            "24000",
+                            str(self.sample_rate),
                             "-ac",
                             "1",
                             "-i",
@@ -986,291 +1041,307 @@ class ConversionThread(QThread):
                     chapter_subtitle_path = None
                     chapter_subtitle_file = None
 
-                # Determine if spaCy segmentation should be used for PRE-TTS segmentation
-                # Only non-English languages use spaCy for pre-segmentation
-                # English uses spaCy only for subtitle generation (post-TTS)
-                # spaCy is disabled when subtitle mode is "Disabled" or "Line"
-                # spaCy is also disabled when input is a subtitle file
-                is_subtitle_input = (
-                    not self.is_direct_text
-                    and self.file_name
-                    and os.path.splitext(self.file_name)[1].lower()
-                    in [".srt", ".ass", ".vtt"]
-                )
-                use_spacy = (
-                    getattr(self, "use_spacy_segmentation", False)
-                    and self.subtitle_mode not in ["Disabled", "Line"]
-                    and not is_subtitle_input
-                )
-                spacy_sentences = None
-                active_split_pattern = self.split_pattern
-                spacing_pattern = r"\s*" if self.lang_code in ["z", "j"] else r"\s+"
 
-                # Pre-load spaCy model for English if it will be needed for subtitle generation
+                # Process each voice segment within the chapter
-                if (
+                for segment_idx, (voice_name, segment_text) in enumerate(voice_segments):
-                    use_spacy
+                    # Load voice for this segment (with caching)
-                    and self.lang_code in ["a", "b"]
+                    try:
-                    and self.subtitle_mode in ["Sentence", "Sentence + Comma"]
+                        loaded_voice = self.load_voice_cached(voice_name, tts)
-                ):
+                        if segment_idx > 0:
-                    from abogen.spacy_utils import get_spacy_model
+                            voice_display = voice_name if len(voice_name) < 50 else voice_name[:47] + "..."
-
+                            self.log_updated.emit((f"  → Voice: {voice_display}", "grey"))
-                    nlp = get_spacy_model(
+                    except Exception:
-                        self.lang_code,
-                        log_callback=lambda msg: self.log_updated.emit(msg),
-                    )
-                    if nlp:
                         self.log_updated.emit(
-                            (
+                            (f"⚠ Voice loading error for '{voice_name}', continuing with previous", "orange")
-                                "\nUsing spaCy for sentence segmentation (only for subtitles)...",
-                                "grey",
-                            )
                         )
+                        if segment_idx == 0:
+                            loaded_voice = self.load_voice_cached(self.voice, tts)
 
-                if use_spacy and self.lang_code not in ["a", "b"]:
+                    # Determine if spaCy segmentation should be used for PRE-TTS segmentation
-                    # Non-English: use spaCy for pre-TTS segmentation
+                    # Only non-English languages use spaCy for pre-segmentation
-                    self.log_updated.emit(
+                    # English uses spaCy only for subtitle generation (post-TTS)
-                        ("\nUsing spaCy for sentence segmentation (pre-TTS)...", "grey")
+                    # spaCy is disabled when subtitle mode is "Disabled" or "Line"
+                    # spaCy is also disabled when input is a subtitle file
+                    is_subtitle_input = (
+                        not self.is_direct_text
+                        and self.file_name
+                        and os.path.splitext(self.file_name)[1].lower()
+                        in [".srt", ".ass", ".vtt"]
                     )
-                    from abogen.spacy_utils import segment_sentences
+                    use_spacy = (
-
+                        getattr(self, "use_spacy_segmentation", False)
-                    spacy_sentences = segment_sentences(
+                        and self.subtitle_mode not in ["Disabled", "Line"]
-                        chapter_text,
+                        and not is_subtitle_input
-                        self.lang_code,
-                        log_callback=lambda msg: self.log_updated.emit(msg),
                     )
-                    if spacy_sentences:
+                    spacy_sentences = None
-                        self.log_updated.emit(
+                    active_split_pattern = self.split_pattern
-                            (
+                    spacing_pattern = r"\s*" if self.lang_code in ["z", "j"] else r"\s+"
-                                f"\nspaCy: Text segmented into {len(spacy_sentences)} sentences...",
-                                "grey",
-                            )
-                        )
-                        # For Sentence + Comma mode, still split on commas within spaCy sentences
-                        if self.subtitle_mode == "Sentence + Comma":
-                            active_split_pattern = r"(?<=[{}]){}|\n+".format(
-                                self.PUNCTUATION_COMMAS, spacing_pattern
-                            )
-                        else:
-                            active_split_pattern = (
-                                "\n"  # Use newline splitting for Sentence mode
-                            )
-                    else:
-                        self.log_updated.emit(
-                            ("\nspaCy: Fallback to default segmentation...", "grey")
-                        )
 
-                # Process text - either as spaCy sentences or as single text
+                    # Pre-load spaCy model for English if it will be needed for subtitle generation
-                text_segments = spacy_sentences if spacy_sentences else [chapter_text]
+                    if (
-
+                        use_spacy
-                # Print active split pattern used by the TTS engine once for this batch
+                        and self.lang_code in ["a", "b"]
-                try:
+                        and self.subtitle_mode in ["Sentence", "Sentence + Comma"]
-                    print(f"Using split pattern: {active_split_pattern!r}")
-                except Exception:
-                    # Print must never break processing
-                    print("Using split pattern: (unprintable)")
-
-                for text_segment in text_segments:
-                    for result in tts(
-                        text_segment,
-                        voice=loaded_voice,
-                        speed=self.speed,
-                        split_pattern=active_split_pattern,
                     ):
-                        # Print the result for debugging
+                        from abogen.spacy_utils import get_spacy_model
-                        # print(f"Result: {result}")
+
-                        if self.cancel_requested:
+                        nlp = get_spacy_model(
-                            if chapter_out_file:
+                            self.lang_code,
-                                chapter_out_file.close()
+                            log_callback=lambda msg: self.log_updated.emit(msg),
-                            if merged_out_file:
-                                merged_out_file.close()
-                            self.conversion_finished.emit("Cancelled", None)
-                            return
-                        current_segment += 1
-                        grapheme_len = len(result.graphemes)
-                        self.processed_char_count += grapheme_len
-                        # Log progress with both character counts and the graphemes content
-                        self.log_updated.emit(
-                            f"\n{self.processed_char_count:,}/{self.total_char_count:,}: {result.graphemes}"
                         )
+                        if nlp:
+                            self.log_updated.emit(
+                                (
+                                    "\nUsing spaCy for sentence segmentation (only for subtitles)...",
+                                    "grey",
+                                )
+                            )
 
-                        chunk_dur = len(result.audio) / rate
+                    if use_spacy and self.lang_code not in ["a", "b"]:
-                        chunk_start = current_time
+                        # Non-English: use spaCy for pre-TTS segmentation
-                        # Write audio directly to merged file ONLY if merging
+                        self.log_updated.emit(
-                        if merge_chapters_at_end and merged_out_file:
+                            ("\nUsing spaCy for sentence segmentation (pre-TTS)...", "grey")
-                            merged_out_file.write(result.audio)
+                        )
-                        elif merge_chapters_at_end and ffmpeg_proc:
+                        from abogen.spacy_utils import segment_sentences
-                            if hasattr(result.audio, "numpy"):
+
-                                audio_bytes = (
+                        spacy_sentences = segment_sentences(
-                                    result.audio.numpy().astype("float32").tobytes()
+                            segment_text,
+                            self.lang_code,
+                            log_callback=lambda msg: self.log_updated.emit(msg),
+                        )
+                        if spacy_sentences:
+                            self.log_updated.emit(
+                                (
+                                    f"\nspaCy: Text segmented into {len(spacy_sentences)} sentences...",
+                                    "grey",
+                                )
+                            )
+                            # For Sentence + Comma mode, still split on commas within spaCy sentences
+                            if self.subtitle_mode == "Sentence + Comma":
+                                active_split_pattern = r"(?<=[{}]){}|\n+".format(
+                                    self.PUNCTUATION_COMMAS, spacing_pattern
                                 )
                             else:
-                                audio_bytes = result.audio.astype("float32").tobytes()
+                                active_split_pattern = (
-                            ffmpeg_proc.stdin.write(audio_bytes)
+                                    "\n"  # Use newline splitting for Sentence mode
-                        if chapter_out_file:
-                            chapter_out_file.write(result.audio)
-                        elif chapter_ffmpeg_proc:
-                            if hasattr(result.audio, "numpy"):
-                                audio_bytes = (
-                                    result.audio.numpy().astype("float32").tobytes()
                                 )
-                            else:
+                        else:
-                                audio_bytes = result.audio.astype("float32").tobytes()
+                            self.log_updated.emit(
-                            chapter_ffmpeg_proc.stdin.write(audio_bytes)
+                                ("\nspaCy: Fallback to default segmentation...", "grey")
-                        # Subtitle logic
+                            )
-                        if self.subtitle_mode != "Disabled":
-                            tokens_list = getattr(result, "tokens", [])
 
-                            # Fallback for languages without token support (non-English)
+                    # Process text - either as spaCy sentences or as single text
-                            # Create a single token representing the entire segment duration
+                    text_segments = spacy_sentences if spacy_sentences else [segment_text]
-                            if not tokens_list and result.graphemes:
 
-                                class FakeToken:
+                    # Print active split pattern used by the TTS engine once for this batch
-                                    def __init__(self, text, start, end):
+                    try:
-                                        self.text = text
+                        print(f"Using split pattern: {active_split_pattern!r}")
-                                        self.start_ts = start
+                    except Exception:
-                                        self.end_ts = end
+                        # Print must never break processing
-                                        self.whitespace = ""
+                        print("Using split pattern: (unprintable)")
 
-                                tokens_list = [
+                    for text_segment in text_segments:
-                                    FakeToken(result.graphemes, 0, chunk_dur)
+                        for result in tts(
-                                ]
+                            text_segment,
+                            voice=loaded_voice,
+                            speed=self.speed,
+                            split_pattern=active_split_pattern,
+                        ):
+                            # Print the result for debugging
+                            # print(f"Result: {result}")
+                            if self.cancel_requested:
+                                if chapter_out_file:
+                                    chapter_out_file.close()
+                                if merged_out_file:
+                                    merged_out_file.close()
+                                self.conversion_finished.emit("Cancelled", None)
+                                return
+                            current_segment += 1
+                            grapheme_len = len(result.graphemes)
+                            self.processed_char_count += grapheme_len
+                            # Log progress with both character counts and the graphemes content
+                            self.log_updated.emit(
+                                f"\n{self.processed_char_count:,}/{self.total_char_count:,}: {result.graphemes}"
+                            )
 
-                            tokens_with_timestamps = []
+                            chunk_dur = len(result.audio) / rate
-                            chapter_tokens_with_timestamps = []
+                            chunk_start = current_time
+                            # Write audio directly to merged file ONLY if merging
+                            if merge_chapters_at_end and merged_out_file:
+                                merged_out_file.write(result.audio)
+                            elif merge_chapters_at_end and ffmpeg_proc:
+                                if hasattr(result.audio, "numpy"):
+                                    audio_bytes = (
+                                        result.audio.numpy().astype("float32").tobytes()
+                                    )
+                                else:
+                                    audio_bytes = result.audio.astype("float32").tobytes()
+                                ffmpeg_proc.stdin.write(audio_bytes)
+                            if chapter_out_file:
+                                chapter_out_file.write(result.audio)
+                            elif chapter_ffmpeg_proc:
+                                if hasattr(result.audio, "numpy"):
+                                    audio_bytes = (
+                                        result.audio.numpy().astype("float32").tobytes()
+                                    )
+                                else:
+                                    audio_bytes = result.audio.astype("float32").tobytes()
+                                chapter_ffmpeg_proc.stdin.write(audio_bytes)
+                            # Subtitle logic
+                            if self.subtitle_mode != "Disabled":
+                                tokens_list = getattr(result, "tokens", [])
 
-                            # Process every token, regardless of text or timestamps
+                                # Fallback for languages without token support (non-English)
-                            for tok in tokens_list:
+                                # Create a single token representing the entire segment duration
-                                tokens_with_timestamps.append(
+                                if not tokens_list and result.graphemes:
-                                    {
+
-                                        "start": chunk_start + (tok.start_ts or 0),
+                                    class FakeToken:
-                                        "end": chunk_start + (tok.end_ts or 0),
+                                        def __init__(self, text, start, end):
-                                        "text": tok.text,
+                                            self.text = text
-                                        "whitespace": tok.whitespace,
+                                            self.start_ts = start
-                                    }
+                                            self.end_ts = end
-                                )
+                                            self.whitespace = ""
-                                if chapter_out_file or chapter_ffmpeg_proc:
+
-                                    chapter_tokens_with_timestamps.append(
+                                    tokens_list = [
+                                        FakeToken(result.graphemes, 0, chunk_dur)
+                                    ]
+
+                                tokens_with_timestamps = []
+                                chapter_tokens_with_timestamps = []
+
+                                # Process every token, regardless of text or timestamps
+                                for tok in tokens_list:
+                                    tokens_with_timestamps.append(
                                         {
-                                            "start": chapter_current_time
+                                            "start": chunk_start + (tok.start_ts or 0),
-                                            + (tok.start_ts or 0),
+                                            "end": chunk_start + (tok.end_ts or 0),
-                                            "end": chapter_current_time
-                                            + (tok.end_ts or 0),
                                             "text": tok.text,
                                             "whitespace": tok.whitespace,
                                         }
                                     )
-                            # Process tokens according to subtitle mode
+                                    if chapter_out_file or chapter_ffmpeg_proc:
-                            # Global subtitle processing ONLY if merging
+                                        chapter_tokens_with_timestamps.append(
+                                            {
+                                                "start": chapter_current_time
+                                                + (tok.start_ts or 0),
+                                                "end": chapter_current_time
+                                                + (tok.end_ts or 0),
+                                                "text": tok.text,
+                                                "whitespace": tok.whitespace,
+                                            }
+                                        )
+                                # Process tokens according to subtitle mode
+                                # Global subtitle processing ONLY if merging
+                                if merge_chapters_at_end:
+                                    # Incremental subtitle writing for merged output
+                                    new_entries = []
+                                    self._process_subtitle_tokens(
+                                        tokens_with_timestamps,
+                                        new_entries,
+                                        self.max_subtitle_words,
+                                        fallback_end_time=chunk_start + chunk_dur,
+                                    )
+                                    if merged_subtitle_file:
+                                        subtitle_format = getattr(
+                                            self, "subtitle_format", "srt"
+                                        )
+                                        if "ass" in subtitle_format:
+                                            for start, end, text in new_entries:
+                                                start_time = self._ass_time(start)
+                                                end_time = self._ass_time(end)
+                                                # Use karaoke effect for highlighting mode
+                                                effect = (
+                                                    "karaoke"
+                                                    if self.subtitle_mode
+                                                    == "Sentence + Highlighting"
+                                                    else ""
+                                                )
+                                                merged_subtitle_file.write(
+                                                    f"Dialogue: 0,{start_time},{end_time},Default,,{merged_subtitle_margin},{merged_subtitle_margin},0,{effect},{merged_subtitle_alignment_tag}{text}\n"
+                                                )
+                                        else:
+                                            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"
+                                                )
+                                                merged_srt_index += 1
+                                # Per-chapter subtitle processing for both file and ffmpeg_proc
+                                if chapter_out_file or chapter_ffmpeg_proc:
+                                    new_chapter_entries = []
+                                    self._process_subtitle_tokens(
+                                        chapter_tokens_with_timestamps,
+                                        new_chapter_entries,
+                                        self.max_subtitle_words,
+                                        fallback_end_time=chapter_current_time + chunk_dur,
+                                    )
+                                    if chapter_subtitle_file:
+                                        subtitle_format = getattr(
+                                            self, "subtitle_format", "srt"
+                                        )
+                                        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)
+                                                # Use karaoke effect for highlighting mode
+                                                effect = (
+                                                    "karaoke"
+                                                    if self.subtitle_mode
+                                                    == "Sentence + Highlighting"
+                                                    else ""
+                                                )
+                                                chapter_subtitle_file.write(
+                                                    f"Dialogue: 0,{start_time},{end_time},Default,,{chapter_subtitle_margin},{chapter_subtitle_margin},0,{effect},{chapter_subtitle_alignment_tag}{text}\n"
+                                                )
+                                        else:
+                                            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"
+                                                )
+                                                chapter_srt_index += 1
                             if merge_chapters_at_end:
-                                # Incremental subtitle writing for merged output
+                                current_time += chunk_dur
-                                new_entries = []
+                                if chapter_out_file or chapter_ffmpeg_proc:
-                                self._process_subtitle_tokens(
+                                    chapter_current_time += chunk_dur
-                                    tokens_with_timestamps,
+                            else:
-                                    new_entries,
+                                if chapter_out_file or chapter_ffmpeg_proc:
-                                    self.max_subtitle_words,
+                                    chapter_current_time += chunk_dur
-                                    fallback_end_time=chunk_start + chunk_dur,
+                            # Calculate percentage based on characters processed
-                                )
+                            percent = min(
-                                if merged_subtitle_file:
+                                int(
-                                    subtitle_format = getattr(
+                                    self.processed_char_count / self.total_char_count * 100
-                                        self, "subtitle_format", "srt"
+                                ),
-                                    )
+                                99,
-                                    if "ass" in subtitle_format:
-                                        for start, end, text in new_entries:
-                                            start_time = self._ass_time(start)
-                                            end_time = self._ass_time(end)
-                                            # Use karaoke effect for highlighting mode
-                                            effect = (
-                                                "karaoke"
-                                                if self.subtitle_mode
-                                                == "Sentence + Highlighting"
-                                                else ""
-                                            )
-                                            merged_subtitle_file.write(
-                                                f"Dialogue: 0,{start_time},{end_time},Default,,{merged_subtitle_margin},{merged_subtitle_margin},0,{effect},{merged_subtitle_alignment_tag}{text}\n"
-                                            )
-                                    else:
-                                        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"
-                                            )
-                                            merged_srt_index += 1
-                            # Per-chapter subtitle processing for both file and ffmpeg_proc
-                            if chapter_out_file or chapter_ffmpeg_proc:
-                                new_chapter_entries = []
-                                self._process_subtitle_tokens(
-                                    chapter_tokens_with_timestamps,
-                                    new_chapter_entries,
-                                    self.max_subtitle_words,
-                                    fallback_end_time=chapter_current_time + chunk_dur,
-                                )
-                                if chapter_subtitle_file:
-                                    subtitle_format = getattr(
-                                        self, "subtitle_format", "srt"
-                                    )
-                                    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)
-                                            # Use karaoke effect for highlighting mode
-                                            effect = (
-                                                "karaoke"
-                                                if self.subtitle_mode
-                                                == "Sentence + Highlighting"
-                                                else ""
-                                            )
-                                            chapter_subtitle_file.write(
-                                                f"Dialogue: 0,{start_time},{end_time},Default,,{chapter_subtitle_margin},{chapter_subtitle_margin},0,{effect},{chapter_subtitle_alignment_tag}{text}\n"
-                                            )
-                                    else:
-                                        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"
-                                            )
-                                            chapter_srt_index += 1
-                        if merge_chapters_at_end:
-                            current_time += chunk_dur
-                            if chapter_out_file or chapter_ffmpeg_proc:
-                                chapter_current_time += chunk_dur
-                        else:
-                            if chapter_out_file or chapter_ffmpeg_proc:
-                                chapter_current_time += chunk_dur
-                        # Calculate percentage based on characters processed
-                        percent = min(
-                            int(
-                                self.processed_char_count / self.total_char_count * 100
-                            ),
-                            99,
-                        )
-
-                        # Calculate ETR based on characters processed
-                        etr_str = "Processing..."
-                        chars_done = self.processed_char_count
-                        elapsed = time.time() - self.etr_start_time
-
-                        # Calculate ETR if enough data is available
-                        if (
-                            chars_done > 0 and elapsed > 0.5
-                        ):  # Check elapsed > 0.5 to avoid instability
-                            avg_time_per_char = elapsed / chars_done
-                            remaining = (
-                                self.total_char_count - self.processed_char_count
                             )
-                            if remaining > 0:
-                                secs = avg_time_per_char * remaining
-                                h = int(secs // 3600)
-                                m = int((secs % 3600) // 60)
-                                s = int(secs % 60)
-                                etr_str = f"{h:02d}:{m:02d}:{s:02d}"
 
-                        # Update progress more frequently (after each result)
+                            # Calculate ETR based on characters processed
-                        self.progress_updated.emit(percent, etr_str)
+                            etr_str = "Processing..."
+                            chars_done = self.processed_char_count
+                            elapsed = time.time() - self.etr_start_time
+
+                            # Calculate ETR if enough data is available
+                            if (
+                                chars_done > 0 and elapsed > 0.5
+                            ):  # Check elapsed > 0.5 to avoid instability
+                                avg_time_per_char = elapsed / chars_done
+                                remaining = (
+                                    self.total_char_count - self.processed_char_count
+                                )
+                                if remaining > 0:
+                                    secs = avg_time_per_char * remaining
+                                    h = int(secs // 3600)
+                                    m = int((secs % 3600) // 60)
+                                    s = int(secs % 60)
+                                    etr_str = f"{h:02d}:{m:02d}:{s:02d}"
+
+                            # Update progress more frequently (after each result)
+                            self.progress_updated.emit(percent, etr_str)
 
                 # 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 * 24000
+                        self.silence_duration * self.sample_rate
                     )  # Silence duration at 24,000 Hz
                     silence_audio = self.np.zeros(silence_samples, dtype="float32")
                     silence_bytes = silence_audio.tobytes()
@@ -1501,7 +1572,7 @@ class ConversionThread(QThread):
                 parent_dir, f"{sanitized_base_name}{suffix}"
             )
             merged_out_path = f"{base_filepath_no_ext}.{self.output_format}"
-            rate = 24000
+            rate = self.sample_rate
 
             # Setup audio output
             merged_out_file, ffmpeg_proc = None, None
@@ -2351,6 +2422,9 @@ class VoicePreviewThread(QThread):
         speed,
         use_gpu=False,
         parent=None,
+        tts_provider="kokoro",
+        supertonic_language="en",
+        supertonic_total_steps=8,
     ):
         super().__init__(parent)
         self.np_module = np_module
@@ -2359,6 +2433,10 @@ class VoicePreviewThread(QThread):
         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")
@@ -2368,6 +2446,11 @@ 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 = (
@@ -2382,39 +2465,56 @@ class VoicePreviewThread(QThread):
 
     def run(self):
         print(
-            f"\nVoice: {self.voice}\nLanguage: {self.lang_code}\nSpeed: {self.speed}\nGPU: {self.use_gpu}\n"
+            f"\nVoice: {self.voice}\nLanguage: {self.lang_code}\nSpeed: {self.speed}\nGPU: {self.use_gpu}\nTTS Provider: {self.tts_provider}\n"
         )
 
         # Generate the preview and save to cache
         try:
+            if self.tts_provider == "supertonic":
+                from abogen.tts_supertonic import SupertonicPipeline
 
-            # Set device based on use_gpu setting and platform
+                tts = SupertonicPipeline(
-            if self.use_gpu:
+                    sample_rate=self.sample_rate,
-                if platform.system() == "Darwin" and platform.processor() == "arm":
+                    lang=self.supertonic_language,
-                    device = "mps"  # Use MPS for Apple Silicon
+                    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 = "cuda"  # Use CUDA for other platforms
+                    device = "cpu"
-            else:
+
-                device = "cpu"
+                tts = self.kpipeline_class(
+                    lang_code=self.lang_code, repo_id="hexgrad/Kokoro-82M", device=device
+                )
+                if "*" in self.voice:
+                    loaded_voice = get_new_voice(tts, self.voice, self.use_gpu)
+                else:
+                    loaded_voice = self.voice
+                sample_text = get_sample_voice_text(self.lang_code)
+                audio_segments = []
+                for result in tts(
+                    sample_text, voice=loaded_voice, speed=self.speed, split_pattern=None
+                ):
+                    audio_segments.append(result.audio)
 
-            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)
-            else:
-                loaded_voice = self.voice
-            sample_text = get_sample_voice_text(self.lang_code)
-            audio_segments = []
-            for result in tts(
-                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)
-                # Save directly to the cache path
+                sf.write(self.cache_path, audio, self.sample_rate)
-                sf.write(self.cache_path, audio, 24000)
                 self.temp_wav = self.cache_path
             self.finished.emit()
         except Exception as e:

abogen/pyqt/gui.py

@@ -9,6 +9,7 @@ from abogen.pyqt.queue_manager_gui import QueueManager
 from abogen.pyqt.queued_item import QueuedItem
 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,
@@ -74,7 +75,7 @@ from abogen.subtitle_utils import (
     calculate_text_length,
 )
 
-from abogen.conversion import ConversionThread, VoicePreviewThread, PlayAudioThread
+from abogen.pyqt.conversion import ConversionThread, VoicePreviewThread, PlayAudioThread, ChapterOptionsDialog, TimestampDetectionDialog
 from abogen.pyqt.book_handler import HandlerDialog
 from abogen.constants import (
     PROGRAM_NAME,
@@ -83,6 +84,8 @@ from abogen.constants import (
     PROGRAM_DESCRIPTION,
     LANGUAGE_DESCRIPTIONS,
     VOICES_INTERNAL,
+    KOKORO_LANG_TO_COUNTRY,
+    SUPERTONIC_LANG_TO_COUNTRY,
     SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
     COLORS,
     SUBTITLE_FORMATS,
@@ -665,6 +668,11 @@ class TextboxDialog(QDialog):
         self.insert_chapter_btn.clicked.connect(self.insert_chapter_marker)
         button_layout.addWidget(self.insert_chapter_btn)
 
+        self.insert_voice_btn = QPushButton("Insert Voice Marker", self)
+        self.insert_voice_btn.setToolTip("Insert a voice change marker at the cursor position")
+        self.insert_voice_btn.clicked.connect(self.insert_voice_marker)
+        button_layout.addWidget(self.insert_voice_btn)
+
         self.cancel_button = QPushButton("Cancel", self)
         self.cancel_button.clicked.connect(self.reject)
 
@@ -767,6 +775,23 @@ class TextboxDialog(QDialog):
         self.update_char_count()
         self.text_edit.setFocus()
 
+    def insert_voice_marker(self):
+        """Insert a voice marker template at cursor position."""
+        cursor = self.text_edit.textCursor()
+        # Use the currently selected voice as the default
+        try:
+            parent_window = self.parent()
+            if parent_window and hasattr(parent_window, 'selected_voice'):
+                default_voice = parent_window.selected_voice or "af_heart"
+            else:
+                default_voice = "af_heart"
+        except Exception:
+            default_voice = "af_heart"
+        cursor.insertText(f"\n<<VOICE:{default_voice}>>\n")
+        self.text_edit.setTextCursor(cursor)
+        self.update_char_count()
+        self.text_edit.setFocus()
+
 
 def migrate_subtitle_format(config):
     """Convert old subtitle_format values to new internal keys."""
@@ -783,6 +808,108 @@ def migrate_subtitle_format(config):
         save_config(config)
 
 
+class WordSubstitutionsDialog(QDialog):
+    """Dialog for configuring word substitutions and text preprocessing options."""
+
+    def __init__(
+        self,
+        parent=None,
+        initial_list="",
+        initial_case_sensitive=False,
+        initial_caps=False,
+        initial_numerals=False,
+        initial_punctuation=False,
+    ):
+        super().__init__(parent)
+        self.setWindowTitle("Word Substitutions Settings")
+        self.setWindowFlags(
+            Qt.WindowType.Window
+            | Qt.WindowType.WindowCloseButtonHint
+            | Qt.WindowType.WindowMaximizeButtonHint
+        )
+        self.resize(600, 500)
+
+        layout = QVBoxLayout(self)
+
+        # Instructions
+        instructions = QLabel(
+            "Enter word substitutions (one per line) in format: Word|NewWord\n"
+            "  - If nothing after |, the word will be erased completely\n"
+            "  - Substitutions match whole words only (e.g., \"tree\" won't match \"trees\" but will match \"tree's\")\n"
+            "  - By default, matching is case-insensitive (e.g., \"gonna\" matches \"Gonna\", \"GONNA\", etc.)",
+            self,
+        )
+        instructions.setStyleSheet(
+            "padding: 10px; background-color: #f0f0f0; border-radius: 5px;"
+        )
+        instructions.setWordWrap(True)
+        layout.addWidget(instructions)
+
+        # Text edit area
+        self.text_edit = QTextEdit(self)
+        self.text_edit.setAcceptRichText(False)
+        self.text_edit.setPlaceholderText("Word|NewWord")
+        self.text_edit.setPlainText(initial_list)
+        layout.addWidget(self.text_edit)
+
+        # Checkboxes
+        self.case_sensitive_checkbox = QCheckBox(
+            "Case-sensitive word matching", self
+        )
+        self.case_sensitive_checkbox.setChecked(initial_case_sensitive)
+        layout.addWidget(self.case_sensitive_checkbox)
+
+        self.caps_checkbox = QCheckBox("Replace ALL CAPS with lowercase", self)
+        self.caps_checkbox.setChecked(initial_caps)
+        layout.addWidget(self.caps_checkbox)
+
+        self.numerals_checkbox = QCheckBox(
+            "Replace Numerals with Words (e.g., 309 \u2192 three hundred and nine)", self
+        )
+        self.numerals_checkbox.setChecked(initial_numerals)
+        layout.addWidget(self.numerals_checkbox)
+
+        self.punctuation_checkbox = QCheckBox(
+            "Fix Nonstandard Punctuation (curly quotes and other Unicode punctuation that may affect how words sound)",
+            self,
+        )
+        self.punctuation_checkbox.setChecked(initial_punctuation)
+        layout.addWidget(self.punctuation_checkbox)
+
+        # Buttons
+        button_layout = QHBoxLayout()
+        self.cancel_button = QPushButton("Cancel", self)
+        self.cancel_button.clicked.connect(self.reject)
+        self.ok_button = QPushButton("OK", self)
+        self.ok_button.setDefault(True)
+        self.ok_button.clicked.connect(self.accept)
+
+        button_layout.addStretch()
+        button_layout.addWidget(self.cancel_button)
+        button_layout.addWidget(self.ok_button)
+        layout.addLayout(button_layout)
+
+    def get_substitutions_list(self):
+        """Get the substitutions list as plain text."""
+        return self.text_edit.toPlainText()
+
+    def get_case_sensitive(self):
+        """Get whether case-sensitive matching is enabled."""
+        return self.case_sensitive_checkbox.isChecked()
+
+    def get_replace_all_caps(self):
+        """Get whether ALL CAPS replacement is enabled."""
+        return self.caps_checkbox.isChecked()
+
+    def get_replace_numerals(self):
+        """Get whether numeral-to-word conversion is enabled."""
+        return self.numerals_checkbox.isChecked()
+
+    def get_fix_nonstandard_punctuation(self):
+        """Get whether nonstandard punctuation fixing is enabled."""
+        return self.punctuation_checkbox.isChecked()
+
+
 class abogen(QWidget):
     def __init__(self):
         super().__init__()
@@ -833,6 +960,22 @@ class abogen(QWidget):
         self.use_silent_gaps = self.config.get("use_silent_gaps", True)
         self.subtitle_speed_method = self.config.get("subtitle_speed_method", "tts")
         self.use_spacy_segmentation = self.config.get("use_spacy_segmentation", True)
+        # Word substitution settings
+        self.word_substitutions_enabled = self.config.get(
+            "word_substitutions_enabled", False
+        )
+        self.word_substitutions_list = self.config.get("word_substitutions_list", "")
+        self.case_sensitive_substitutions = self.config.get(
+            "case_sensitive_substitutions", False
+        )
+        self.replace_all_caps = self.config.get("replace_all_caps", False)
+        self.replace_numerals = self.config.get("replace_numerals", False)
+        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
 
@@ -881,6 +1024,16 @@ 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()
@@ -970,6 +1123,53 @@ 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)
@@ -1071,6 +1271,35 @@ class abogen(QWidget):
         subtitle_layout.addWidget(self.subtitle_combo)
         controls_layout.addLayout(subtitle_layout)
 
+        # Word Substitutions section
+        word_sub_layout = QHBoxLayout()
+        word_sub_layout.setSpacing(7)
+        word_sub_label = QLabel("Word Substitutions:", self)
+        word_sub_layout.addWidget(word_sub_label)
+
+        self.word_sub_combo = QComboBox(self)
+        self.word_sub_combo.addItems(["Disabled", "Enabled"])
+        self.word_sub_combo.setStyleSheet(
+            "QComboBox { min-height: 20px; padding: 6px 12px; }"
+        )
+        self.word_sub_combo.setSizePolicy(
+            QSizePolicy.Policy.Expanding, QSizePolicy.Policy.Fixed
+        )
+        self.word_sub_combo.setCurrentText(
+            "Enabled" if self.word_substitutions_enabled else "Disabled"
+        )
+        self.word_sub_combo.currentTextChanged.connect(self.on_word_sub_changed)
+        word_sub_layout.addWidget(self.word_sub_combo)
+
+        self.btn_word_sub_settings = QPushButton("Settings", self)
+        self.btn_word_sub_settings.setFixedSize(80, 36)
+        self.btn_word_sub_settings.setStyleSheet("QPushButton { padding: 6px 12px; }")
+        self.btn_word_sub_settings.clicked.connect(self.show_word_sub_dialog)
+        self.btn_word_sub_settings.setEnabled(self.word_substitutions_enabled)
+        word_sub_layout.addWidget(self.btn_word_sub_settings)
+
+        controls_layout.addLayout(word_sub_layout)
+
         # Output voice format
         format_layout = QHBoxLayout()
         format_layout.setSpacing(7)
@@ -1598,6 +1827,12 @@ 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(
@@ -1657,6 +1892,48 @@ 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:"):
@@ -1665,10 +1942,26 @@ class abogen(QWidget):
             from abogen.voice_profiles import load_profiles
 
             entry = load_profiles().get(pname, {})
-            # set mixed voices and language
             if isinstance(entry, dict):
-                self.mixed_voice_state = entry.get("voices", [])
+                entry_provider = str(entry.get("provider", "")).strip().lower()
-                self.selected_lang = entry.get("language")
+                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:
                 self.mixed_voice_state = entry
                 self.selected_lang = entry[0][0] if entry and entry[0] else None
@@ -1681,7 +1974,12 @@ class abogen(QWidget):
         else:
             self.mixed_voice_state = None
             self.selected_profile_name = None
-            self.selected_voice, self.selected_lang = data, data[0]
+            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.config["selected_voice"] = data
             if "selected_profile_name" in self.config:
                 del self.config["selected_profile_name"]
@@ -1700,19 +1998,40 @@ 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
+        # re-add profiles matching current provider
         profile_icon = QIcon(get_resource_path("abogen.assets", "profile.png"))
-        for pname in load_profiles().keys():
+        for pname, entry in load_profiles().items():
-            self.voice_combo.addItem(profile_icon, pname, f"profile:{pname}")
+            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":
+                    self.voice_combo.addItem(profile_icon, pname, f"profile:{pname}")
         # re-add voices
-        for v in VOICES_INTERNAL:
+        if provider == "supertonic":
-            icon = QIcon()
+            for v in DEFAULT_SUPERTONIC_VOICES:
-            flag_path = get_resource_path("abogen.assets.flags", f"{v[0]}.png")
+                icon = QIcon()
-            if flag_path and os.path.exists(flag_path):
+                if v.startswith("F"):
-                icon = QIcon(flag_path)
+                    icon_path = get_resource_path("abogen.assets", "female.png")
-            self.voice_combo.addItem(icon, f"{v}", v)
+                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")
+                if flag_path and os.path.exists(flag_path):
+                    icon = QIcon(flag_path)
+                self.voice_combo.addItem(icon, f"{v}", v)
         # restore selection
         idx = -1
         if self.selected_profile_name:
@@ -1903,6 +2222,9 @@ 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
@@ -2015,15 +2337,46 @@ class abogen(QWidget):
                 self.subtitle_speed_method = getattr(
                     queued_item, "subtitle_speed_method", "tts"
                 )
+                # Word substitution settings
+                self.word_substitutions_enabled = getattr(
+                    queued_item, "word_substitutions_enabled", False
+                )
+                self.word_substitutions_list = getattr(
+                    queued_item, "word_substitutions_list", ""
+                )
+                self.case_sensitive_substitutions = getattr(
+                    queued_item, "case_sensitive_substitutions", False
+                )
+                self.replace_all_caps = getattr(queued_item, "replace_all_caps", False)
+                self.replace_numerals = getattr(queued_item, "replace_numerals", False)
+                self.fix_nonstandard_punctuation = getattr(
+                    queued_item, "fix_nonstandard_punctuation", False
+                )
 
-                # This ensures that if conversion.py (or utils) reads from config/disk 
+                # This ensures that if conversion.py (or utils) reads from config/disk
                 # instead of using passed arguments, it sees the correct queue values.
                 self.config["replace_single_newlines"] = self.replace_single_newlines
                 self.config["subtitle_mode"] = self.subtitle_mode
                 self.config["selected_format"] = self.selected_format
                 self.config["use_silent_gaps"] = self.use_silent_gaps
                 self.config["subtitle_speed_method"] = self.subtitle_speed_method
-                
+                # Word substitution settings
+                self.config["word_substitutions_enabled"] = self.word_substitutions_enabled
+                self.config["word_substitutions_list"] = self.word_substitutions_list
+                self.config["case_sensitive_substitutions"] = self.case_sensitive_substitutions
+                self.config["replace_all_caps"] = self.replace_all_caps
+                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:
@@ -2046,6 +2399,8 @@ 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
@@ -2055,6 +2410,8 @@ 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
 
@@ -2144,6 +2501,10 @@ 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,
@@ -2159,8 +2520,11 @@ class abogen(QWidget):
                 total_char_count=self.char_count,
                 use_gpu=self.gpu_ok,
                 from_queue=from_queue,
-                save_base_path=self.displayed_file_path,  # Pass the save base path (original file for EPUB)
+                save_base_path=self.displayed_file_path,
-            )  # Use gpu_ok status
+                tts_provider=tts_provider,
+                supertonic_language=supertonic_language,
+                supertonic_total_steps=supertonic_total_steps,
+            )
             # 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
@@ -2179,6 +2543,21 @@ class abogen(QWidget):
             self.conversion_thread.subtitle_speed_method = self.subtitle_speed_method
             # Pass use_spacy_segmentation setting
             self.conversion_thread.use_spacy_segmentation = self.use_spacy_segmentation
+            # Pass word substitution settings
+            self.conversion_thread.word_substitutions_enabled = (
+                self.word_substitutions_enabled
+            )
+            self.conversion_thread.word_substitutions_list = (
+                self.word_substitutions_list
+            )
+            self.conversion_thread.case_sensitive_substitutions = (
+                self.case_sensitive_substitutions
+            )
+            self.conversion_thread.replace_all_caps = self.replace_all_caps
+            self.conversion_thread.replace_numerals = self.replace_numerals
+            self.conversion_thread.fix_nonstandard_punctuation = (
+                self.fix_nonstandard_punctuation
+            )
             # Pass separate_chapters_format setting
             self.conversion_thread.separate_chapters_format = (
                 self.separate_chapters_format
@@ -2222,9 +2601,15 @@ 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))
-            self.update_log("Loading modules...")
+            tts_provider = self.provider_combo.currentData()
-            load_thread = LoadPipelineThread(pipeline_loaded_callback)
+            if tts_provider == "supertonic":
-            load_thread.start()
+                # 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)
+                load_thread.start()
 
         threading.Thread(target=gpu_and_load, daemon=True).start()
 
@@ -2537,9 +2922,32 @@ 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 = ""
 
@@ -2644,6 +3052,13 @@ 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
@@ -2702,14 +3117,28 @@ class abogen(QWidget):
                     else None
                 )
         else:
-            lang = self.selected_voice[0]
+            if self.provider_combo.currentData() == "supertonic":
-            voice = self.selected_voice
+                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 ""
 
         # 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
+            np_module, kpipeline_class, lang, voice, speed, gpu_ok,
+            tts_provider=tts_provider,
+            supertonic_language=supertonic_language,
+            supertonic_total_steps=supertonic_total_steps,
         )
         self.preview_thread.finished.connect(self._play_preview_audio)
         self.preview_thread.error.connect(self._preview_error)
@@ -2927,6 +3356,41 @@ class abogen(QWidget):
         self.config["use_gpu"] = self.use_gpu
         save_config(self.config)
 
+    def on_word_sub_changed(self, text):
+        """Handle word substitution dropdown change."""
+        self.word_substitutions_enabled = text == "Enabled"
+        self.btn_word_sub_settings.setEnabled(self.word_substitutions_enabled)
+
+        # Save to config
+        self.config["word_substitutions_enabled"] = self.word_substitutions_enabled
+        save_config(self.config)
+
+    def show_word_sub_dialog(self):
+        """Show word substitutions settings dialog."""
+        dialog = WordSubstitutionsDialog(
+            self,
+            initial_list=self.word_substitutions_list,
+            initial_case_sensitive=self.case_sensitive_substitutions,
+            initial_caps=self.replace_all_caps,
+            initial_numerals=self.replace_numerals,
+            initial_punctuation=self.fix_nonstandard_punctuation,
+        )
+
+        if dialog.exec() == QDialog.DialogCode.Accepted:
+            self.word_substitutions_list = dialog.get_substitutions_list()
+            self.case_sensitive_substitutions = dialog.get_case_sensitive()
+            self.replace_all_caps = dialog.get_replace_all_caps()
+            self.replace_numerals = dialog.get_replace_numerals()
+            self.fix_nonstandard_punctuation = dialog.get_fix_nonstandard_punctuation()
+
+            # Save all settings to config
+            self.config["word_substitutions_list"] = self.word_substitutions_list
+            self.config["case_sensitive_substitutions"] = self.case_sensitive_substitutions
+            self.config["replace_all_caps"] = self.replace_all_caps
+            self.config["replace_numerals"] = self.replace_numerals
+            self.config["fix_nonstandard_punctuation"] = self.fix_nonstandard_punctuation
+            save_config(self.config)
+
     def cleanup_conversion_thread(self):
         # Stop conversion thread
         if (
@@ -2991,8 +3455,6 @@ class abogen(QWidget):
         """Show dialog to ask user about chapter processing options when chapters are detected in a .txt file"""
         # Check if this is a timestamp detection (-1) or chapter detection
         if chapter_count == -1:
-            from abogen.conversion import TimestampDetectionDialog
-
             dialog = TimestampDetectionDialog(parent=self)
             dialog.setWindowModality(Qt.WindowModality.ApplicationModal)
 
@@ -3007,8 +3469,6 @@ class abogen(QWidget):
             return
 
         # Normal chapter detection
-        from abogen.conversion import ChapterOptionsDialog
-
         dialog = ChapterOptionsDialog(chapter_count, parent=self)
         dialog.setWindowModality(Qt.WindowModality.ApplicationModal)
 

abogen/pyqt/queue_manager_gui.py

@@ -35,6 +35,12 @@ OVERRIDE_FIELDS = [
     "replace_single_newlines",
     "use_silent_gaps",
     "subtitle_speed_method",
+    "word_substitutions_enabled",
+    "word_substitutions_list",
+    "case_sensitive_substitutions",
+    "replace_all_caps",
+    "replace_numerals",
+    "fix_nonstandard_punctuation",
 ]
 
 
@@ -474,6 +480,21 @@ class QueueManager(QDialog):
             attrs["subtitle_speed_method"] = getattr(
                 parent, "subtitle_speed_method", "tts"
             )
+            # word substitutions
+            attrs["word_substitutions_enabled"] = getattr(
+                parent, "word_substitutions_enabled", False
+            )
+            attrs["word_substitutions_list"] = getattr(
+                parent, "word_substitutions_list", ""
+            )
+            attrs["case_sensitive_substitutions"] = getattr(
+                parent, "case_sensitive_substitutions", False
+            )
+            attrs["replace_all_caps"] = getattr(parent, "replace_all_caps", False)
+            attrs["replace_numerals"] = getattr(parent, "replace_numerals", False)
+            attrs["fix_nonstandard_punctuation"] = getattr(
+                parent, "fix_nonstandard_punctuation", False
+            )
             # book handler options
             attrs["save_chapters_separately"] = getattr(
                 parent, "save_chapters_separately", None

abogen/pyqt/queued_item.py

@@ -19,3 +19,14 @@ class QueuedItem:
     save_base_path: str = None
     save_chapters_separately: bool = None
     merge_chapters_at_end: bool = None
+    # Word Substitution fields
+    word_substitutions_enabled: bool = False
+    word_substitutions_list: str = ""
+    case_sensitive_substitutions: bool = False
+    replace_all_caps: bool = False
+    replace_numerals: bool = False
+    fix_nonstandard_punctuation: bool = False
+    # TTS Provider fields
+    tts_provider: str = "kokoro"
+    supertonic_language: str = "en"
+    supertonic_total_steps: int = 8

abogen/pyqt/voice_formula_gui.py

@@ -31,6 +31,7 @@ from abogen.constants import (
     VOICES_INTERNAL,
     SUPPORTED_LANGUAGES_FOR_SUBTITLE_GENERATION,
     LANGUAGE_DESCRIPTIONS,
+    KOKORO_LANG_TO_COUNTRY,
     COLORS,
 )
 import re
@@ -189,8 +190,9 @@ 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"{language_code}.png"
+            "abogen.assets.flags", f"{country_code}.png"
         )
         gender_icon_path = get_resource_path(
             "abogen.assets", "female.png" if is_female else "male.png"
@@ -512,7 +514,8 @@ class VoiceFormulaDialog(QDialog):
         header_row.addWidget(QLabel("Language:"))
         self.language_combo = QComboBox()
         for code, desc in LANGUAGE_OPTIONS:
-            flag = get_resource_path("abogen.assets.flags", f"{code}.png")
+            country_code = KOKORO_LANG_TO_COUNTRY.get(code, code)
+            flag = get_resource_path("abogen.assets.flags", f"{country_code}.png")
             if flag and os.path.exists(flag):
                 self.language_combo.addItem(QIcon(flag), desc, code)
             else:

abogen/subtitle_utils.py

@@ -15,6 +15,8 @@ _ASS_STYLING_PATTERN = re.compile(r"\{[^}]+\}")
 _ASS_NEWLINE_N_PATTERN = re.compile(r"\\N")
 _ASS_NEWLINE_LOWER_N_PATTERN = re.compile(r"\\n")
 _CHAPTER_MARKER_SEARCH_PATTERN = re.compile(r"<<CHAPTER_MARKER:(.*?)>>")
+_VOICE_MARKER_PATTERN = re.compile(r"<<VOICE:[^>]*>>")
+_VOICE_MARKER_SEARCH_PATTERN = re.compile(r"<<VOICE:(.*?)>>")
 _WEBVTT_HEADER_PATTERN = re.compile(r"^WEBVTT.*?\n", re.MULTILINE)
 _VTT_STYLE_PATTERN = re.compile(r"STYLE\s*\n.*?(?=\n\n|$)", re.DOTALL)
 _VTT_NOTE_PATTERN = re.compile(r"NOTE\s*\n.*?(?=\n\n|$)", re.DOTALL)
@@ -31,17 +33,19 @@ _LINUX_ILLEGAL_CHARS_PATTERN = re.compile(r"[/\x00]")
 
 
 def clean_subtitle_text(text):
-    """Remove chapter markers and metadata tags from subtitle text."""
+    """Remove chapter markers, voice markers, and metadata tags from subtitle text."""
     # Use pre-compiled patterns for better performance
     text = _METADATA_TAG_PATTERN.sub("", text)
     text = _CHAPTER_MARKER_PATTERN.sub("", text)
+    text = _VOICE_MARKER_PATTERN.sub("", text)
     return text.strip()
 
 
 def calculate_text_length(text):
     # Use pre-compiled patterns for better performance
-    # Ignore chapter markers and metadata patterns in a single pass
+    # Ignore chapter markers, voice markers, and metadata patterns in a single pass
     text = _CHAPTER_MARKER_PATTERN.sub("", text)
+    text = _VOICE_MARKER_PATTERN.sub("", text)
     text = _METADATA_TAG_PATTERN.sub("", text)
     # Ignore newlines and leading/trailing spaces
     text = text.replace("\n", "").strip()
@@ -459,3 +463,122 @@ def sanitize_name_for_os(name, is_folder=True):
         sanitized = sanitized[:255].rstrip(". ")
 
     return sanitized
+
+
+def validate_voice_name(voice_name):
+    """Validate voice name against VOICES_INTERNAL list (case-insensitive).
+    Handles both single voices and formulas like 'af_heart*0.5 + am_echo*0.5'.
+
+    Args:
+        voice_name: Voice name or formula string to validate
+
+    Returns:
+        Tuple of (is_valid, invalid_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
+
+    # Create case-insensitive lookup set (done once per call)
+    voice_lookup_lower = {v.lower() for v in VOICES_INTERNAL}
+    voice_name = voice_name.strip()
+
+    # Check if it's a formula (contains *)
+    if "*" in voice_name:
+        # Extract voice names from formula
+        voices = voice_name.split("+")
+        for term in voices:
+            if "*" in term:
+                base_voice = term.split("*")[0].strip()
+                # Case-insensitive comparison
+                if base_voice.lower() not in voice_lookup_lower:
+                    return False, base_voice
+        return True, None
+    else:
+        # Single voice - case-insensitive comparison
+        if voice_name.lower() not in voice_lookup_lower:
+            return False, voice_name
+        return True, None
+
+
+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.
+
+    Args:
+        text: Text potentially containing <<VOICE:name>> markers
+        default_voice: Voice to use if no markers found or before first marker
+
+    Returns:
+        Tuple of (segments_list, last_voice_used, valid_count, invalid_count):
+            - segments_list: List of (voice_name, segment_text) tuples
+            - last_voice_used: The voice that should continue into next chapter
+            - valid_count: Number of valid voice markers processed
+            - invalid_count: Number of invalid voice markers skipped
+    """
+    from abogen.constants import VOICES_INTERNAL
+
+    voice_splits = list(_VOICE_MARKER_SEARCH_PATTERN.finditer(text))
+
+    if not voice_splits:
+        # No voice markers, return entire text with default voice
+        return [(default_voice, text)], default_voice, 0, 0
+
+    segments = []
+    current_voice = default_voice
+    valid_markers = 0
+    invalid_markers = 0
+
+    # Text before first marker uses default voice
+    first_start = voice_splits[0].start()
+    if first_start > 0:
+        intro_text = text[:first_start].strip()
+        if intro_text:
+            segments.append((current_voice, intro_text))
+
+    # Process each voice marker
+    for idx, match in enumerate(voice_splits):
+        voice_name = match.group(1).strip()
+        start = match.end()
+        end = voice_splits[idx + 1].start() if idx + 1 < len(voice_splits) else len(text)
+        segment_text = text[start:end].strip()
+
+        # Validate voice name
+        is_valid, invalid_voice = validate_voice_name(voice_name)
+        if is_valid:
+            # Normalize to lowercase to match canonical form
+            # Handle both single voices and formulas
+            if "*" in voice_name:
+                # Normalize each voice in the formula
+                normalized_parts = []
+                for part in voice_name.split("+"):
+                    part = part.strip()
+                    if "*" in part:
+                        voice_part, weight = part.split("*", 1)
+                        # 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),
+                            voice_part.strip()
+                        )
+                        normalized_parts.append(f"{canonical_voice}*{weight.strip()}")
+                current_voice = " + ".join(normalized_parts)
+            else:
+                # 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),
+                    voice_name
+                )
+            valid_markers += 1
+        else:
+            # Invalid voice - stay with previous voice
+            invalid_markers += 1
+
+        if segment_text:
+            segments.append((current_voice, segment_text))
+
+    # Return segments, last voice, and counts
+    return segments, current_voice, valid_markers, invalid_markers

abogen/text_extractor.py

@@ -1023,8 +1023,13 @@ class EpubExtractor:
         if not html:
             return ""
         soup = BeautifulSoup(html, "html.parser")
-        for tag in soup.find_all(["p", "div"]):
+
+        # Add line breaks after block-level elements to ensure pauses in speech
+        for tag in soup.find_all(
+            ["p", "div", "h1", "h2", "h3", "h4", "h5", "h6", "li", "blockquote"]
+        ):
             tag.append("\n\n")
+
         for ol in soup.find_all("ol"):
             start_attr = ol.get("start")
             try:

abogen/tts_supertonic.py

@@ -4,6 +4,7 @@ import ast
 from dataclasses import dataclass
 import logging
 import math
+import os
 import re
 from typing import Any, Iterable, Iterator, Optional
 
@@ -15,6 +16,13 @@ logger = logging.getLogger(__name__)
 
 DEFAULT_SUPERTONIC_VOICES = ("M1", "M2", "M3", "M4", "M5", "F1", "F2", "F3", "F4", "F5")
 
+SUPERTONIC_AVAILABLE_LANGS = [
+    "en", "ko", "ja", "ar", "bg", "cs", "da", "de", "el",
+    "es", "et", "fi", "fr", "hi", "hr", "hu", "id", "it",
+    "lt", "lv", "nl", "pl", "pt", "ro", "ru", "sk", "sl",
+    "sv", "tr", "uk", "vi", "na",
+]
+
 
 @dataclass
 class SupertonicSegment:
@@ -89,7 +97,7 @@ _UNSUPPORTED_CHARS_RE = re.compile(
 
 
 def _parse_unsupported_characters(error: BaseException) -> list[str]:
-    """Best-effort extraction of unsupported characters from SuperTonic errors."""
+    """Best-effort extraction of unsupported characters from Supertonic errors."""
 
     message = " ".join(
         str(part) for part in getattr(error, "args", ()) if part is not None
@@ -155,6 +163,7 @@ def _configure_supertonic_gpu() -> None:
     except Exception as exc:
         logger.warning("Could not configure supertonic GPU providers: %s", exc)
 
+SUPERTONIC_MAX_CHUNK_LENGTH = 500
 
 class SupertonicPipeline:
     """Minimal adapter that mimics Kokoro's pipeline iteration interface."""
@@ -165,11 +174,14 @@ class SupertonicPipeline:
         sample_rate: int,
         auto_download: bool = True,
         total_steps: int = 5,
-        max_chunk_length: int = 300,
+        max_chunk_length: int = SUPERTONIC_MAX_CHUNK_LENGTH,
+        lang: str = "en",
+        intra_op_num_threads: Optional[int] = None,
     ) -> None:
         self.sample_rate = int(sample_rate)
         self.total_steps = int(total_steps)
         self.max_chunk_length = int(max_chunk_length)
+        self.lang = str(lang or "en")
 
         # Configure GPU providers before importing TTS
         _configure_supertonic_gpu()
@@ -181,7 +193,8 @@ class SupertonicPipeline:
                 "Supertonic is not installed. Install it with `pip install supertonic`."
             ) from exc
 
-        self._tts = TTS(auto_download=auto_download)
+        threads = intra_op_num_threads if intra_op_num_threads is not None else os.cpu_count()
+        self._tts = TTS(auto_download=auto_download, intra_op_num_threads=threads)
 
     def __call__(
         self,
@@ -191,12 +204,14 @@ class SupertonicPipeline:
         speed: float,
         split_pattern: Optional[str] = None,
         total_steps: Optional[int] = None,
+        lang: Optional[str] = None,
     ) -> Iterator[SupertonicSegment]:
         voice_name = (voice or "").strip() or "M1"
         steps = int(total_steps) if total_steps is not None else self.total_steps
         steps = max(2, min(15, steps))
         speed_value = float(speed) if speed is not None else 1.0
         speed_value = max(0.7, min(2.0, speed_value))
+        language = str(lang or self.lang or "en")
 
         style = self._tts.get_voice_style(voice_name=voice_name)
         chunks = _split_text(
@@ -207,12 +222,13 @@ class SupertonicPipeline:
             removed: set[str] = set()
             last_exc: Exception | None = None
 
-            # SuperTonic can raise ValueError for unsupported characters; strip and retry.
+            # Supertonic can raise ValueError for unsupported characters; strip and retry.
             for attempt in range(3):
                 try:
                     wav, duration = self._tts.synthesize(
                         text=chunk_to_speak,
                         voice_style=style,
+                        lang=language,
                         total_steps=steps,
                         speed=speed_value,
                         max_chunk_length=self.max_chunk_length,
@@ -238,14 +254,14 @@ class SupertonicPipeline:
                     chunk_to_speak = sanitized
                     if not chunk_to_speak:
                         logger.warning(
-                            "SuperTonic: dropped a chunk after removing unsupported characters: %s",
+                            "Supertonic: dropped a chunk after removing unsupported characters: %s",
                             sorted(removed),
                         )
                         break
 
                     if attempt == 0:
                         logger.warning(
-                            "SuperTonic: removed unsupported characters %s and retried.",
+                            "Supertonic: removed unsupported characters %s and retried.",
                             sorted(removed),
                         )
             else:

abogen/webui/conversion_runner.py

@@ -59,7 +59,7 @@ def _supertonic_voice_from_spec(spec: Any, fallback: str) -> str:
     raw = str(spec or "").strip()
     fallback_raw = str(fallback or "").strip()
 
-    # SuperTonic voices are discrete IDs (M1/F3/...). If we see a Kokoro mix
+    # Supertonic voices are discrete IDs (M1/F3/...). If we see a Kokoro mix
     # formula (contains '*' or '+'), ignore it and fall back to a safe voice.
     if not raw or "*" in raw or "+" in raw:
         raw = fallback_raw
@@ -1584,7 +1584,7 @@ def run_conversion_job(job: Job) -> None:
                 pipelines[provider_norm] = SupertonicPipeline(
                     sample_rate=SAMPLE_RATE,
                     auto_download=True,
-                    total_steps=int(getattr(job, "supertonic_total_steps", 5) or 5),
+                    total_steps=int(getattr(job, "supertonic_total_steps", 8) or 8),
                 )
                 return pipelines[provider_norm]
 
@@ -1618,7 +1618,7 @@ def run_conversion_job(job: Job) -> None:
                 provider = str(entry.get("provider") or "kokoro").strip().lower() or "kokoro"
                 if provider == "supertonic":
                     voice = str(entry.get("voice") or getattr(job, "voice", "M1") or "M1").strip() or "M1"
-                    steps = int(entry.get("total_steps") or getattr(job, "supertonic_total_steps", 5) or 5)
+                    steps = int(entry.get("total_steps") or getattr(job, "supertonic_total_steps", 8) or 8)
                     speed = float(entry.get("speed") or getattr(job, "speed", 1.0) or 1.0)
                     return "supertonic", _supertonic_voice_from_spec(voice, getattr(job, "voice", "M1")), speed, steps
                 formula = _formula_from_kokoro_entry(entry)
@@ -1634,7 +1634,7 @@ def run_conversion_job(job: Job) -> None:
             """Resolve a raw voice spec into (provider, resolved_spec, choice, speed, steps).
 
             For Kokoro formulas, `choice` will be a resolved voice tensor (via `voice_formulas`).
-            For SuperTonic, `choice` will be a valid SuperTonic voice id.
+            For Supertonic, `choice` will be a valid Supertonic voice id.
             """
 
             provider, resolved, speed, steps = resolve_voice_target(raw_spec)
@@ -1857,7 +1857,7 @@ def run_conversion_job(job: Job) -> None:
                     voice=voice_name,
                     speed=float(speed_override if speed_override is not None else job.speed),
                     split_pattern=split_pattern,
-                    total_steps=int(supertonic_steps_override if supertonic_steps_override is not None else getattr(job, "supertonic_total_steps", 5)),
+                    total_steps=int(supertonic_steps_override if supertonic_steps_override is not None else getattr(job, "supertonic_total_steps", 8)),
                 )
             else:
                 kokoro_pipeline = get_pipeline("kokoro")
@@ -2448,7 +2448,7 @@ def _load_pipeline(job: Job):
         return SupertonicPipeline(
             sample_rate=SAMPLE_RATE,
             auto_download=True,
-            total_steps=int(getattr(job, "supertonic_total_steps", 5) or 5),
+            total_steps=int(getattr(job, "supertonic_total_steps", 8) or 8),
         )
 
     device = "cpu"
@@ -2610,7 +2610,7 @@ def _build_ffmpeg_command(path: Path, fmt: str, metadata: Optional[Dict[str, str
 def _resolve_voice(pipeline, voice_spec: str, use_gpu: bool):
     if "*" in voice_spec:
         # Voice formulas are a Kokoro-only feature (they require a pipeline that can
-        # load individual Kokoro voices). When running with SuperTonic (or when the
+        # load individual Kokoro voices). When running with Supertonic (or when the
         # pipeline is otherwise unavailable), treat the spec as a plain string and
         # allow downstream provider-specific resolution to choose a safe fallback.
         if pipeline is None or not hasattr(pipeline, "load_single_voice"):

abogen/webui/routes/api.py

@@ -162,7 +162,7 @@ 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 5)
+    supertonic_total_steps = int(payload.get("supertonic_total_steps") or payload.get("total_steps") or settings.get("supertonic_total_steps") or 8)
 
     voice_spec = ""
     resolved_provider = provider or "kokoro"
@@ -224,7 +224,7 @@ 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 5)
+    supertonic_total_steps = int(payload.get("supertonic_total_steps") or 8)
     
     settings = load_settings()
     use_gpu = settings.get("use_gpu", False)
@@ -269,7 +269,7 @@ 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 5),
+            supertonic_total_steps=supertonic_total_steps or int(settings.get("supertonic_total_steps") or 8),
             pronunciation_overrides=pronunciation_overrides,
             manual_overrides=manual_overrides,
             speakers=speakers,

abogen/webui/routes/settings.py

@@ -43,8 +43,12 @@ 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:
-        current["supertonic_total_steps"] = max(2, min(15, int(form.get("supertonic_total_steps", current.get("supertonic_total_steps", 5)))))
+        total_steps = int(form.get("supertonic_total_steps", current.get("supertonic_total_steps", 8)))
+        current["supertonic_total_steps"] = max(2, min(15, total_steps))
     except (TypeError, ValueError):
         pass
     try:

abogen/webui/routes/utils/form.py

@@ -577,7 +577,7 @@ 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"}:
         pending.tts_provider = provider_value
@@ -913,6 +913,15 @@ 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,
@@ -928,6 +937,8 @@ 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,

abogen/webui/routes/utils/preview.py

@@ -17,10 +17,43 @@ _preview_pipeline_lock = threading.Lock()
 def _select_device() -> str:
     import platform
 
+    try:
+        import torch  # type: ignore[import-not-found]
+    except Exception:
+        return "cpu"
+
     system = platform.system()
     if system == "Darwin" and platform.processor() == "arm":
-        return "mps"
+        try:
-    return "cuda"
+            if torch.backends.mps.is_available():
+                return "mps"
+        except Exception:
+            pass
+        return "cpu"
+
+    try:
+        if torch.cuda.is_available():
+            return "cuda"
+    except Exception:
+        pass
+    return "cpu"
+
+
+def _resolve_pipeline(language: str, use_gpu: bool) -> Tuple[Any, bool]:
+    devices: List[str] = ["cpu"]
+    if use_gpu:
+        preferred = _select_device()
+        if preferred != "cpu":
+            devices.insert(0, preferred)
+
+    last_error: Optional[Exception] = None
+    for device in devices:
+        try:
+            return get_preview_pipeline(language, device), device != "cpu"
+        except Exception as exc:
+            last_error = exc
+
+    raise RuntimeError("Preview pipeline is unavailable") from last_error
 
 
 def _to_float32(audio_segment) -> np.ndarray:
@@ -59,7 +92,7 @@ def generate_preview_audio(
     speed: float,
     use_gpu: bool,
     tts_provider: str = "kokoro",
-    supertonic_total_steps: int = 5,
+    supertonic_total_steps: int = 8,
     max_seconds: float = 8.0,
     pronunciation_overrides: Optional[Iterable[Mapping[str, Any]]] = None,
     manual_overrides: Optional[Iterable[Mapping[str, Any]]] = None,
@@ -115,15 +148,7 @@ def generate_preview_audio(
             total_steps=supertonic_total_steps,
         )
     else:
-        device = "cpu"
+        pipeline, pipeline_uses_gpu = _resolve_pipeline(language, use_gpu)
-        if use_gpu:
-            try:
-                device = _select_device()
-            except Exception:
-                device = "cpu"
-                use_gpu = False
-
-        pipeline = get_preview_pipeline(language, device)
         if pipeline is None:
             raise RuntimeError("Preview pipeline is unavailable")
 
@@ -131,7 +156,7 @@ def generate_preview_audio(
         if voice_spec and "*" in voice_spec:
             from abogen.voice_formulas import get_new_voice
 
-            voice_choice = get_new_voice(pipeline, voice_spec, use_gpu)
+            voice_choice = get_new_voice(pipeline, voice_spec, pipeline_uses_gpu)
 
         segments = pipeline(
             normalized_text,
@@ -176,7 +201,7 @@ def synthesize_preview(
     speed: float,
     use_gpu: bool,
     tts_provider: str = "kokoro",
-    supertonic_total_steps: int = 5,
+    supertonic_total_steps: int = 8,
     max_seconds: float = 8.0,
     pronunciation_overrides: Optional[Iterable[Mapping[str, Any]]] = None,
     manual_overrides: Optional[Iterable[Mapping[str, Any]]] = None,

abogen/webui/routes/utils/service.py

@@ -25,7 +25,7 @@ def submit_job(pending: PendingJob) -> str:
         tts_provider=getattr(pending, "tts_provider", "kokoro"),
         voice=pending.voice,
         speed=pending.speed,
-        supertonic_total_steps=getattr(pending, "supertonic_total_steps", 5),
+        supertonic_total_steps=getattr(pending, "supertonic_total_steps", 8),
         use_gpu=pending.use_gpu,
         subtitle_mode=pending.subtitle_mode,
         output_format=pending.output_format,

abogen/webui/routes/utils/settings.py

@@ -175,7 +175,8 @@ def settings_defaults() -> Dict[str, Any]:
         "save_mode": "default_output" if has_output_override() else "save_next_to_input",
         "default_speaker": "",
         "default_voice": VOICES_INTERNAL[0] if VOICES_INTERNAL else "",
-        "supertonic_total_steps": 5,
+        "tts_provider": "kokoro",
+        "supertonic_total_steps": 8,
         "supertonic_speed": 1.0,
         "replace_single_newlines": False,
         "use_gpu": True,

abogen/webui/routes/utils/voice.py

@@ -666,7 +666,7 @@ def resolve_voice_choice(
 
         # Provider-aware behavior:
         # - Kokoro profiles typically represent mixes (formula strings).
-        # - SuperTonic profiles represent a discrete voice id + settings.
+        # - Supertonic profiles represent a discrete voice id + settings.
         #   In that case, we return a speaker reference so downstream can
         #   resolve provider per-speaker and allow mixed-provider casting.
         if provider == "supertonic":

abogen/webui/service.py

@@ -111,7 +111,7 @@ class Job:
     subtitle_format: str
     created_at: float
     tts_provider: str = "kokoro"
-    supertonic_total_steps: int = 5
+    supertonic_total_steps: int = 8
     save_chapters_separately: bool = False
     merge_chapters_at_end: bool = True
     separate_chapters_format: str = "wav"
@@ -204,7 +204,7 @@ class Job:
             "queue_position": self.queue_position,
             "options": {
                 "tts_provider": getattr(self, "tts_provider", "kokoro"),
-                "supertonic_total_steps": getattr(self, "supertonic_total_steps", 5),
+                "supertonic_total_steps": getattr(self, "supertonic_total_steps", 8),
                 "save_chapters_separately": self.save_chapters_separately,
                 "merge_chapters_at_end": self.merge_chapters_at_end,
                 "separate_chapters_format": self.separate_chapters_format,
@@ -552,7 +552,7 @@ class PendingJob:
     normalization_overrides: Dict[str, Any]
     created_at: float
     tts_provider: str = "kokoro"
-    supertonic_total_steps: int = 5
+    supertonic_total_steps: int = 8
     cover_image_path: Optional[Path] = None
     cover_image_mime: Optional[str] = None
     chapter_intro_delay: float = 0.5
@@ -621,7 +621,7 @@ class ConversionService:
         voice: str,
         speed: float,
         tts_provider: str = "kokoro",
-        supertonic_total_steps: int = 5,
+        supertonic_total_steps: int = 8,
         use_gpu: bool,
         subtitle_mode: str,
         output_format: str,
@@ -674,7 +674,7 @@ class ConversionService:
             voice=voice,
             speed=speed,
             tts_provider=tts_provider,
-            supertonic_total_steps=int(supertonic_total_steps or 5),
+            supertonic_total_steps=int(supertonic_total_steps or 8),
             use_gpu=use_gpu,
             subtitle_mode=subtitle_mode,
             output_format=output_format,
@@ -1147,7 +1147,7 @@ class ConversionService:
             "tts_provider": getattr(job, "tts_provider", "kokoro"),
             "voice": job.voice,
             "speed": job.speed,
-            "supertonic_total_steps": getattr(job, "supertonic_total_steps", 5),
+            "supertonic_total_steps": getattr(job, "supertonic_total_steps", 8),
             "use_gpu": job.use_gpu,
             "subtitle_mode": job.subtitle_mode,
             "output_format": job.output_format,
@@ -1275,7 +1275,7 @@ class ConversionService:
             replace_single_newlines=bool(payload.get("replace_single_newlines", False)),
             subtitle_format=payload.get("subtitle_format", "srt"),
             created_at=float(payload.get("created_at", time.time())),
-            supertonic_total_steps=int(payload.get("supertonic_total_steps", 5)),
+            supertonic_total_steps=int(payload.get("supertonic_total_steps", 8)),
             save_chapters_separately=bool(payload.get("save_chapters_separately", False)),
             merge_chapters_at_end=bool(payload.get("merge_chapters_at_end", True)),
             separate_chapters_format=payload.get("separate_chapters_format", "wav"),

abogen/webui/templates/partials/new_job_step_book.html

@@ -26,6 +26,26 @@
     {% set subtitle_value = settings_dict.get('subtitle_mode', 'Disabled') %}
   {% endif %}
 {% endif %}
+{% set tts_provider_value = form_values.get('tts_provider') if form_values else None %}
+{% if not tts_provider_value %}
+  {% if pending and pending.tts_provider %}
+    {% set tts_provider_value = pending.tts_provider %}
+  {% else %}
+    {% set tts_provider_value = settings_dict.get('tts_provider', 'kokoro') %}
+  {% endif %}
+{% endif %}
+{% set supertonic_steps_value = form_values.get('supertonic_total_steps') if form_values else None %}
+{% if supertonic_steps_value is none %}
+  {% if pending and pending.supertonic_total_steps is not none %}
+    {% set supertonic_steps_value = pending.supertonic_total_steps %}
+  {% else %}
+    {% set supertonic_steps_value = settings_dict.get('supertonic_total_steps', 8) %}
+  {% endif %}
+{% endif %}
+{% if supertonic_steps_value is not none and supertonic_steps_value is string %}
+  {% set supertonic_steps_value = supertonic_steps_value|int %}
+{% endif %}
+{% set is_supertonic = tts_provider_value == 'supertonic' %}
 {% set generate_flag = form_values.get('generate_epub3') if form_values else None %}
 {% if generate_flag is not none %}
   {% set generate_epub3 = True %}
@@ -273,6 +293,13 @@
       <div class="form-section__layout form-section__layout--split">
         <div class="form-section__group">
           <div class="field">
+            <label for="tts_provider">TTS Engine</label>
+            <select id="tts_provider" name="tts_provider" data-role="tts-provider" {{ 'disabled' if readonly else '' }}>
+              <option value="kokoro" {% if tts_provider_value == 'kokoro' %}selected{% endif %}>Kokoro</option>
+              <option value="supertonic" {% if tts_provider_value == 'supertonic' %}selected{% endif %}>Supertonic</option>
+            </select>
+          </div>
+          <div class="field" data-role="voice-profile-field">
             <label for="voice_profile">Voice profile</label>
             <select id="voice_profile" name="voice_profile" data-role="voice-profile" {{ 'disabled' if readonly else '' }}>
               <option value="__standard" {% if profile_value == '__standard' %}selected{% endif %}>Standard voice</option>
@@ -280,13 +307,13 @@
               {% if options.voice_profile_options %}
               <optgroup label="Saved mixes">
                 {% for profile in options.voice_profile_options %}
-                <option value="{{ profile.name }}" data-language="{{ profile.language }}" data-formula="{{ profile.formula|e }}" {% if profile_value == profile.name %}selected{% endif %}>{{ profile.name }}{% if profile.language %} ({{ profile.language|upper }}){% endif %}</option>
+                <option value="{{ profile.name }}" data-language="{{ profile.language }}" data-formula="{{ profile.formula|e }}" data-provider="{{ profile.provider|default('kokoro')|lower }}" {% if profile_value == profile.name %}selected{% endif %}>{{ profile.name }}{% if profile.language %} ({{ profile.language|upper }}){% endif %}{% if profile.provider and profile.provider|lower != 'kokoro' %} · {{ profile.provider|capitalize }}{% endif %}</option>
                 {% endfor %}
               </optgroup>
               {% endif %}
             </select>
           </div>
-          <div class="field" data-role="voice-field" {% if profile_value != '__standard' %}hidden aria-hidden="true"{% endif %}>
+          <div class="field" data-role="voice-field" data-provider="kokoro" {% if profile_value != '__standard' or is_supertonic %}hidden aria-hidden="true"{% endif %}>
             <label for="voice">Voice</label>
             <select id="voice" name="voice" data-role="voice-select" data-default="{{ narrator_voice or settings_dict.get('default_voice', '') }}" {{ 'disabled' if readonly else '' }}>
               {% for voice in options.voices %}
@@ -294,10 +321,23 @@
               {% endfor %}
             </select>
           </div>
-          <div class="field" data-conditional="formula" data-role="formula-field" {% if profile_value != '__formula' %}hidden aria-hidden="true"{% endif %}>
+          <div class="field" data-role="voice-field" data-provider="supertonic" {% if profile_value != '__standard' or not is_supertonic %}hidden aria-hidden="true"{% endif %}>
+            <label for="voice_st">Supertonic voice</label>
+            <select id="voice_st" name="voice" data-role="voice-select" data-default="{{ narrator_voice or 'M1' }}" {{ 'disabled' if readonly else '' }}>
+              {% for voice in ['M1','M2','M3','M4','M5','F1','F2','F3','F4','F5'] %}
+              <option value="{{ voice }}" {% if narrator_voice == voice and profile_value == '__standard' %}selected{% endif %}>{{ voice }}</option>
+              {% endfor %}
+            </select>
+          </div>
+          <div class="field" data-conditional="formula" data-role="formula-field" data-provider="kokoro" {% if profile_value != '__formula' or is_supertonic %}hidden aria-hidden="true"{% endif %}>
             <label for="voice_formula">Custom voice formula</label>
             <input type="text" id="voice_formula" name="voice_formula" placeholder="af_nova*0.4+am_liam*0.6" data-role="voice-formula" value="{{ voice_formula_value }}" {{ 'disabled' if readonly else '' }}>
           </div>
+          <div class="field" data-role="supertonic-steps-field" {% if not is_supertonic %}hidden aria-hidden="true"{% endif %}>
+            <label for="supertonic_total_steps">Supertonic quality (total steps)</label>
+            <input type="number" id="supertonic_total_steps" name="supertonic_total_steps" min="2" max="15" value="{{ supertonic_steps_value }}" {{ 'disabled' if readonly else '' }}>
+            <p class="hint">2 = fastest/lowest quality, 15 = slowest/highest quality.</p>
+          </div>
         </div>
         <div class="form-section__group">
           <div class="field field--slider">
@@ -423,3 +463,54 @@
     </div>
   </footer>
 </form>
+
+<script nonce="{{ csp_nonce() if csp_nonce else '' }}">
+(function() {
+  const form = document.querySelector('[data-wizard-form="true"][data-step="book"]');
+  if (!form) return;
+  const providerSelect = form.querySelector('[data-role="tts-provider"]');
+  if (!providerSelect) return;
+
+  function filterProfilesByProvider(provider) {
+    const profileSelect = form.querySelector('[data-role="voice-profile"]');
+    if (!profileSelect) return;
+    const options = profileSelect.querySelectorAll('option[data-provider]');
+    options.forEach(function(opt) {
+      const matches = !opt.dataset.provider || opt.dataset.provider === provider;
+      opt.hidden = !matches;
+      if (opt.selected && opt.hidden) {
+        opt.selected = false;
+      }
+    });
+    if (!profileSelect.value || profileSelect.selectedOptions[0]?.hidden) {
+      const firstVisible = profileSelect.querySelector('option:not([hidden])');
+      if (firstVisible) profileSelect.value = firstVisible.value;
+    }
+    profileSelect.dispatchEvent(new Event('change', { bubbles: true }));
+  }
+
+  function syncProviderUI(provider) {
+    var isSupertonic = provider === 'supertonic';
+    form.querySelectorAll('[data-role="voice-field"]').forEach(function(el) {
+      el.hidden = el.dataset.provider !== provider;
+      el.setAttribute('aria-hidden', el.hidden ? 'true' : 'false');
+    });
+    var formulaField = form.querySelector('[data-role="formula-field"]');
+    if (formulaField) {
+      formulaField.hidden = isSupertonic;
+      formulaField.setAttribute('aria-hidden', isSupertonic ? 'true' : 'false');
+    }
+    var stepsField = form.querySelector('[data-role="supertonic-steps-field"]');
+    if (stepsField) {
+      stepsField.hidden = !isSupertonic;
+      stepsField.setAttribute('aria-hidden', isSupertonic ? 'false' : 'true');
+    }
+    filterProfilesByProvider(provider);
+  }
+
+  providerSelect.addEventListener('change', function() {
+    syncProviderUI(providerSelect.value);
+  });
+  syncProviderUI(providerSelect.value);
+})();
+</script>

abogen/webui/templates/settings.html

@@ -61,6 +61,15 @@
               <p class="hint">Pick a saved speaker from Speaker Studio to use by default for new jobs.</p>
             </div>
 
+            <div class="field">
+              <label for="tts_provider">Default TTS Engine</label>
+              <select id="tts_provider" name="tts_provider">
+                <option value="kokoro" {% if settings.tts_provider == 'kokoro' %}selected{% endif %}>Kokoro</option>
+                <option value="supertonic" {% if settings.tts_provider == 'supertonic' %}selected{% endif %}>Supertonic</option>
+              </select>
+              <p class="hint">Select the default TTS engine for new jobs.</p>
+            </div>
+
             <div class="field field--wide">
               <p class="tag">Kokoro settings</p>
             </div>

abogen/word_substitution.py

@@ -0,0 +1,254 @@
+"""
+Word substitution module for text-to-speech preprocessing.
+
+This module provides functionality to:
+- Replace words/phrases with custom text
+- Convert ALL CAPS to lowercase
+- Convert numerals to words
+- Fix nonstandard punctuation for TTS compatibility
+
+All substitutions preserve special markers (chapter, voice, metadata, timestamps).
+"""
+
+import re
+
+from abogen.subtitle_utils import (
+    _CHAPTER_MARKER_PATTERN,
+    _VOICE_MARKER_PATTERN,
+    _METADATA_TAG_PATTERN,
+    _TIMESTAMP_ONLY_PATTERN,
+)
+
+
+def apply_word_substitutions(
+    text,
+    substitutions_list_str,
+    case_sensitive=False,
+    replace_all_caps=False,
+    replace_numerals=False,
+    fix_nonstandard_punctuation=False,
+):
+    """
+    Apply word substitutions to text while preserving markers.
+
+    Args:
+        text: Input text
+        substitutions_list_str: Newline-separated "Word|NewWord" pairs
+        case_sensitive: If True, match words case-sensitively
+        replace_all_caps: Convert ALL CAPS words to lowercase
+        replace_numerals: Convert numbers to words
+        fix_nonstandard_punctuation: Fix curly quotes, em/en dashes, etc.
+
+    Returns:
+        Modified text
+    """
+    # Apply nonstandard punctuation fixes FIRST (if enabled)
+    if fix_nonstandard_punctuation:
+        text = fix_punctuation(text)
+
+    # Parse substitutions list
+    substitutions = parse_substitutions_list(substitutions_list_str)
+
+    # Split text into segments (markers vs content)
+    segments = split_text_preserving_markers(text)
+
+    # Process each segment
+    processed_segments = []
+    for segment_type, segment_text in segments:
+        if segment_type == "marker":
+            # Preserve markers unchanged
+            processed_segments.append(segment_text)
+        else:
+            # Apply substitutions to content
+            processed_text = segment_text
+
+            # Apply word substitutions
+            if substitutions:
+                processed_text = apply_word_replacements(
+                    processed_text, substitutions, case_sensitive
+                )
+
+            # Apply ALL CAPS conversion
+            if replace_all_caps:
+                processed_text = convert_all_caps_to_lowercase(processed_text)
+
+            # Apply numeral conversion
+            if replace_numerals:
+                processed_text = convert_numerals_to_words(processed_text)
+
+            processed_segments.append(processed_text)
+
+    return "".join(processed_segments)
+
+
+def parse_substitutions_list(substitutions_str):
+    """
+    Parse newline-separated "Word|NewWord" format.
+
+    Args:
+        substitutions_str: String with substitutions, one per line
+
+    Returns:
+        List of tuples: [(word, replacement), ...]
+    """
+    substitutions = []
+    for line in substitutions_str.strip().split("\n"):
+        line = line.strip()
+        if not line or "|" not in line:
+            continue
+
+        parts = line.split("|", 1)
+        if len(parts) == 2:
+            word = parts[0].strip()
+            replacement = parts[1].strip()
+            if word:  # Only add if word is not empty
+                substitutions.append((word, replacement))
+
+    return substitutions
+
+
+def split_text_preserving_markers(text):
+    """
+    Split text into segments alternating between markers and content.
+
+    Args:
+        text: Input text with potential markers
+
+    Returns:
+        List of tuples: [("marker"|"content", text), ...]
+    """
+    # Combined pattern for all markers and timestamps
+    marker_pattern = re.compile(
+        r"(<<CHAPTER_MARKER:[^>]*>>|<<VOICE:[^>]*>>|<<METADATA_[^:]+:[^>]*>>|\d{1,2}:\d{2}:\d{2}(?:[.,]\d{1,3})?)"
+    )
+
+    segments = []
+    last_end = 0
+
+    for match in marker_pattern.finditer(text):
+        # Content before marker
+        if match.start() > last_end:
+            segments.append(("content", text[last_end : match.start()]))
+
+        # Marker itself
+        segments.append(("marker", match.group(0)))
+        last_end = match.end()
+
+    # Remaining content after last marker
+    if last_end < len(text):
+        segments.append(("content", text[last_end:]))
+
+    return segments
+
+
+def apply_word_replacements(text, substitutions, case_sensitive=False):
+    """
+    Apply word substitutions using whole-word matching.
+
+    Args:
+        text: Input text
+        substitutions: List of (word, replacement) tuples
+        case_sensitive: If True, match case-sensitively
+
+    Returns:
+        Text with substitutions applied
+    """
+    for word, replacement in substitutions:
+        # Use word boundaries for exact matching
+        # Escape special regex characters
+        escaped_word = re.escape(word)
+        pattern = re.compile(
+            r"\b" + escaped_word + r"\b",
+            0 if case_sensitive else re.IGNORECASE,
+        )
+        text = pattern.sub(replacement, text)
+
+    return text
+
+
+def convert_all_caps_to_lowercase(text):
+    """
+    Convert ALL CAPS words to lowercase.
+
+    Args:
+        text: Input text
+
+    Returns:
+        Text with ALL CAPS converted to lowercase
+    """
+
+    def replace_caps(match):
+        word = match.group(0)
+        # Convert to lowercase
+        return word.lower()
+
+    # Match words that are ALL CAPS (2+ letters)
+    pattern = re.compile(r"\b[A-Z]{2,}\b")
+    return pattern.sub(replace_caps, text)
+
+
+def convert_numerals_to_words(text):
+    """
+    Convert numerals to words using num2words library.
+
+    Args:
+        text: Input text
+
+    Returns:
+        Text with numerals converted to words
+    """
+    try:
+        from num2words import num2words
+    except ImportError:
+        # If num2words not available, return unchanged
+        return text
+
+    def replace_number(match):
+        try:
+            number = int(match.group(0))
+            # Convert to words in English
+            return num2words(number)
+        except Exception:
+            # If conversion fails, return original
+            return match.group(0)
+
+    # Match integers (but not timestamps or other patterns)
+    # Negative lookbehind/ahead to avoid timestamps
+    pattern = re.compile(r"(?<!\d:)\b\d+\b(?!:\d)")
+    return pattern.sub(replace_number, text)
+
+
+def fix_punctuation(text):
+    """
+    Convert nonstandard punctuation to standard equivalents.
+
+    This helps TTS engines pronounce words correctly by converting:
+    - Curly quotes to straight quotes
+    - Ellipsis to three periods
+
+    Args:
+        text: Input text
+
+    Returns:
+        Text with nonstandard punctuation fixed
+    """
+    # Define replacements
+    replacements = {
+        # Curly double quotes
+        "\u201c": '"',  # Left double quotation mark
+        "\u201d": '"',  # Right double quotation mark
+        "\u201e": '"',  # Double low-9 quotation mark
+        # Curly single quotes
+        "\u2018": "'",  # Left single quotation mark
+        "\u2019": "'",  # Right single quotation mark
+        "\u201a": "'",  # Single low-9 quotation mark
+        "\u201b": "'",  # Single high-reversed-9 quotation mark
+        # Other punctuation
+        "\u2026": "...",  # Ellipsis
+    }
+
+    # Apply all replacements
+    for old_char, new_char in replacements.items():
+        text = text.replace(old_char, new_char)
+
+    return text

pyproject.toml

@@ -30,7 +30,7 @@ dependencies = [
   "pip",
   "kokoro>=0.9.4",
   "misaki[zh]>=0.9.4",
-  "supertonic>=0.1.0",
+  "supertonic>=1.3.1",
   "ebooklib>=0.19",
   "beautifulsoup4>=4.13.4",
   "spacy>=3.8.7,<4.0",
@@ -96,8 +96,6 @@ exclude = [
 [tool.hatch.build.targets.wheel]
 packages = ["abogen"]
 
-[tool.hatch.build]
-include = ["abogen/webui/templates/**", "abogen/webui/static/**"]
 
 [tool.hatch.version]
 path = "abogen/VERSION"
@@ -113,11 +111,11 @@ filterwarnings = [
 
 [project.optional-dependencies]
 # NVIDIA GPU (Windows) (CUDA 12.6)       # uv tool install abogen[cuda126]
-cuda126 = ["torch"]
+cuda126 = ["torch", "onnxruntime-gpu>=1.26.0"]
 # NVIDIA GPU (Windows) (CUDA 12.8)       # uv tool install abogen[cuda]
-cuda = ["torch"]
+cuda = ["torch", "onnxruntime-gpu>=1.26.0"]
 # NVIDIA GPU (Windows) (CUDA 13.0)       # uv tool install abogen[cuda130]
-cuda130 = ["torch"]
+cuda130 = ["torch", "onnxruntime-gpu>=1.26.0"]
 # AMD GPU (Linux) (ROCm 6.4)             # uv tool install abogen[rocm]
 rocm = ["torch", "pytorch-triton-rocm"]
 # Development dependencies               # uv tool install abogen[dev]

tests/test_voice_formula_resolution.py

@@ -6,13 +6,13 @@ from abogen.tts_supertonic import DEFAULT_SUPERTONIC_VOICES
 
 def test_resolve_voice_formula_without_pipeline_does_not_crash() -> None:
     # This can happen when a previously-saved Kokoro mix formula is present
-    # but the active provider is SuperTonic (no Kokoro pipeline object).
+    # but the active provider is Supertonic (no Kokoro pipeline object).
     formula = "af_heart*0.5+af_sky*0.5"
     resolved = _resolve_voice(None, formula, use_gpu=False)
     assert resolved == formula
 
 
 def test_supertonic_voice_from_formula_falls_back_to_valid_voice() -> None:
-    # When a stale Kokoro mix formula is present, SuperTonic should not receive it.
+    # When a stale Kokoro mix formula is present, Supertonic should not receive it.
     chosen = _supertonic_voice_from_spec("af_heart*0.5+af_sky*0.5", "af_heart*1.0")
     assert chosen in DEFAULT_SUPERTONIC_VOICES