feat: language parameter for potential accent tuning

1 files changed, 38 insertions, 6 deletions

diff --git a/converter/tts.py b/converter/tts.py
index 83c7330..db5de9b 100644
--- a/converter/tts.py
+++ b/converter/tts.py
@@ -15,11 +15,38 @@ from . import config
 logger = logging.getLogger(__name__)
 
 
+def normalize_language(value: Optional[str]) -> str:
+    """Normalize a user-provided language name to a Qwen3-TTS display name.
+
+    Accepts the display names in config.TTS_LANGUAGES case-insensitively as
+    well as the short aliases in config.TTS_LANGUAGE_ALIASES (ISO 639-1 codes
+    and common shorthands). Raises ValueError for anything else, since the
+    Qwen3-TTS demo silently falls back to "Auto" for unrecognized languages.
+    """
+    if value is None:
+        raise ValueError("Language must not be None")
+    candidate = value.strip()
+    if not candidate:
+        raise ValueError("Language must not be empty")
+    for name in config.TTS_LANGUAGES:
+        if candidate.lower() == name.lower():
+            return name
+    alias = config.TTS_LANGUAGE_ALIASES.get(candidate.lower())
+    if alias:
+        return alias
+    raise ValueError(
+        f"Unknown language: {value!r}. Expected one of "
+        f"{', '.join(config.TTS_LANGUAGES)} (or an alias: "
+        f"{', '.join(sorted(config.TTS_LANGUAGE_ALIASES))})."
+    )
+
+
 class QwenTTSClient:
     """Generates audio chunks through a Qwen3-TTS Gradio server."""
 
     def __init__(self, voice_mode: str = "custom_voice", voice_clone_ref_audio: Optional[str] = None,
-                 voice_clone_ref_text: Optional[str] = None, skip_transcription: bool = False):
+                 voice_clone_ref_text: Optional[str] = None, skip_transcription: bool = False,
+                 language: Optional[str] = None):
         if voice_mode not in config.VOICE_MODES:
             raise ValueError(
                 f"Unknown voice mode: {voice_mode!r} (expected one of {config.VOICE_MODES})"
@@ -28,6 +55,11 @@ class QwenTTSClient:
         self.voice_clone_ref_audio = voice_clone_ref_audio
         self.voice_clone_ref_text = (voice_clone_ref_text or "").strip()
         self.skip_transcription = skip_transcription
+        if language is None:
+            language = (config.VOICE_CLONE_LANGUAGE if voice_mode == config.VOICE_MODE_CLONE
+                        else config.CUSTOM_VOICE_LANGUAGE)
+        # Validate before connecting so bad values fail fast without a server.
+        self.language = normalize_language(language)
         self.client = None
         self.api_info: Dict[str, Any] = {}
         self.clone_client = None
@@ -70,7 +102,7 @@ class QwenTTSClient:
                 self.voice_clone_ref_text = self.transcribe_audio(self.voice_clone_ref_audio) or ""
         if not self.voice_clone_ref_text:
             print("[WARNING] No reference text available; using x-vector-only clone mode (lower quality).")
-            print('          Pass --voice-sample-text "..." for higher-quality in-context cloning.')
+            print('          Pass --transcription "..." for higher-quality in-context cloning.')
         else:
             print(f"[OK] Reference text: {self.voice_clone_ref_text[:100]}...")
 
@@ -258,7 +290,7 @@ class QwenTTSClient:
         if custom_api == "/run_instruct":
             payload = dict(
                 text=text,
-                lang_disp=config.CUSTOM_VOICE_LANGUAGE,
+                lang_disp=self.language,
                 spk_disp=config.SPEAKER_DISPLAY_NAMES.get(
                     config.CUSTOM_VOICE_SPEAKER.lower(), config.CUSTOM_VOICE_SPEAKER),
                 instruct=config.CUSTOM_VOICE_INSTRUCT,
@@ -266,7 +298,7 @@ class QwenTTSClient:
         else:
             payload = dict(
                 text=text,
-                language=config.CUSTOM_VOICE_LANGUAGE,
+                language=self.language,
                 speaker=config.CUSTOM_VOICE_SPEAKER,
                 instruct=config.CUSTOM_VOICE_INSTRUCT,
             )
@@ -305,14 +337,14 @@ class QwenTTSClient:
                 ref_txt=self.voice_clone_ref_text,
                 use_xvec=use_xvector,
                 text=text,
-                lang_disp=config.VOICE_CLONE_LANGUAGE,
+                lang_disp=self.language,
             )
         else:
             payload = dict(
                 ref_audio=self._ref_audio_payload(),
                 ref_text=self.voice_clone_ref_text,
                 target_text=text,
-                language=config.VOICE_CLONE_LANGUAGE,
+                language=self.language,
                 use_xvector_only=use_xvector,
             )
             optional_params = {