diff options
Diffstat (limited to 'app/converter/clients/qwen.py')
| -rw-r--r-- | app/converter/clients/qwen.py | 29 |
1 files changed, 17 insertions, 12 deletions
diff --git a/app/converter/clients/qwen.py b/app/converter/clients/qwen.py index ed3149b..17f14c5 100644 --- a/app/converter/clients/qwen.py +++ b/app/converter/clients/qwen.py @@ -15,7 +15,7 @@ from .base import (BaseTTSClient, ConversionCancelled, resolve_request_seed, VOICE_MODE_CLONE, VOICE_MODE_CUSTOM, VOICE_MODE_DESIGN, VOICE_MODES) from .languages import normalize_language -from .speakers import speaker_display_name +from .speakers import QWEN3_TTS_SPEAKERS, speaker_display_name_for logger = logging.getLogger(__name__) @@ -32,7 +32,8 @@ class QwenTTSClient(BaseTTSClient): voice_mode: str = "custom_voice", voice_clone_ref_audio: Optional[str] = None, voice_clone_ref_text: Optional[str] = None, skip_transcription: bool = False, language: Optional[str] = None, api_url: Optional[str] = None, - instructions: Optional[str] = None, quiet: bool = False): + instructions: Optional[str] = None, quiet: bool = False, + voice: Optional[str] = None): super().__init__(chunks_dir, quiet=quiet) if voice_mode not in VOICE_MODES: raise ValueError( @@ -42,11 +43,17 @@ class QwenTTSClient(BaseTTSClient): 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 + # Built-in CustomVoice speaker (VOICE_MODE_CUSTOM): the --voice + # value / the Generate form's Speaker pick. Required there — there + # is no configured default speaker. + self.speaker = (voice or "").strip() or None + if voice_mode == VOICE_MODE_CUSTOM and not self.speaker: + raise ValueError( + "CustomVoice mode requires a speaker: pass --voice SPEAKER " + f"(one of {', '.join(QWEN3_TTS_SPEAKERS)})") # Voice design / style instruction (VoiceDesign mode): describes the - # voice to design. Defaults to the configured CustomVoice INSTRUCT so - # a run never sends an empty design prompt. - self.instructions = (instructions if instructions is not None - else config.INSTRUCT).strip() + # voice to design. Required there (validated by the converter). + self.instructions = (instructions or "").strip() # api_url overrides the configured endpoint for the active voice mode # (used by the hub's "[remote]" backend entries and --api-url). self.api_url = (api_url or "").strip() or None @@ -261,21 +268,19 @@ class QwenTTSClient(BaseTTSClient): # ------------------------------------------------------------------ def _generate_custom_voice(self, text: str) -> Tuple: - """Generate audio using CustomVoice mode.""" + """Generate audio using CustomVoice mode with the run's speaker.""" custom_api = self._resolve_api_name("/run_instruct", "/run_custom_voice", "/generate_custom_voice") if custom_api == "/run_instruct": payload = dict( text=text, lang_disp=self.language, - spk_disp=speaker_display_name(), - instruct=config.INSTRUCT, + spk_disp=speaker_display_name_for(self.speaker), ) else: payload = dict( text=text, language=self.language, - speaker=config.SPEAKER, - instruct=config.INSTRUCT, + speaker=self.speaker, ) if self._endpoint_accepts_param(custom_api, "model_id_cv"): payload["model_id_cv"] = CUSTOM_VOICE_MODEL_ID @@ -323,7 +328,7 @@ class QwenTTSClient(BaseTTSClient): clone_api = self._resolve_api_name("/run_voice_clone", "/generate_voice_clone", api_info=self.clone_api_info) - use_xvector = config.XVECTOR_ONLY or not self.voice_clone_ref_text + use_xvector = not self.voice_clone_ref_text if clone_api == "/run_voice_clone": payload = dict( |