| 309 | return self.audio_path |
| 310 | |
| 311 | def TTS(self, request, context): |
| 312 | try: |
| 313 | from fish_speech.utils.schema import ServeTTSRequest, ServeReferenceAudio |
| 314 | |
| 315 | if not request.dst: |
| 316 | return backend_pb2.Result( |
| 317 | success=False, message="dst (output path) is required" |
| 318 | ) |
| 319 | |
| 320 | text = request.text.strip() |
| 321 | if not text: |
| 322 | return backend_pb2.Result(success=False, message="Text is empty") |
| 323 | |
| 324 | # Get generation parameters from options |
| 325 | top_p = self.options.get("top_p", 0.8) |
| 326 | temperature = self.options.get("temperature", 0.8) |
| 327 | repetition_penalty = self.options.get("repetition_penalty", 1.1) |
| 328 | max_new_tokens = self.options.get("max_new_tokens", 1024) |
| 329 | chunk_length = self.options.get("chunk_length", 200) |
| 330 | |
| 331 | # Build references list for voice cloning |
| 332 | references = [] |
| 333 | voice_name = request.voice if request.voice else None |
| 334 | |
| 335 | if voice_name and voice_name in self.voices: |
| 336 | ref_audio_path = self._get_ref_audio_path(voice_name) |
| 337 | if ref_audio_path and os.path.exists(ref_audio_path): |
| 338 | with open(ref_audio_path, "rb") as f: |
| 339 | audio_bytes = f.read() |
| 340 | ref_text = self.voices[voice_name].get("ref_text", "") |
| 341 | references.append( |
| 342 | ServeReferenceAudio(audio=audio_bytes, text=ref_text) |
| 343 | ) |
| 344 | print( |
| 345 | f"[INFO] Using voice '{voice_name}' with reference audio: {ref_audio_path}", |
| 346 | file=sys.stderr, |
| 347 | ) |
| 348 | elif self.audio_path: |
| 349 | ref_audio_path = self._get_ref_audio_path() |
| 350 | if ref_audio_path and os.path.exists(ref_audio_path): |
| 351 | with open(ref_audio_path, "rb") as f: |
| 352 | audio_bytes = f.read() |
| 353 | ref_text = self.options.get("ref_text", "") |
| 354 | references.append( |
| 355 | ServeReferenceAudio(audio=audio_bytes, text=ref_text) |
| 356 | ) |
| 357 | print( |
| 358 | f"[INFO] Using reference audio: {ref_audio_path}", |
| 359 | file=sys.stderr, |
| 360 | ) |
| 361 | |
| 362 | # Build ServeTTSRequest |
| 363 | tts_request = ServeTTSRequest( |
| 364 | text=text, |
| 365 | references=references, |
| 366 | top_p=top_p, |
| 367 | temperature=temperature, |
| 368 | repetition_penalty=repetition_penalty, |