fixing error stt and tts, empty chunk audio

.gitignore

@@ -27,4 +27,5 @@ API_CONTRACT_VOICE.md
 HIGHLIGHT_VOICE.md
 HIGHLIGHT_STT_TTS.md
 
-credentials/
+credentials/
+scripts/

main.py

@@ -116,6 +116,8 @@ async def text_to_speech(req: TTSRequest) -> StreamingResponse:
     if not req.text.strip():
         raise HTTPException(status_code=400, detail="text must not be empty.")
 
+    logger.info("TTS request: provider=%s, text_len=%d, text=%r", req.provider, len(req.text), req.text)
+
     if req.provider == "gemini":
         if not GOOGLE_API_KEY:
             raise HTTPException(status_code=503, detail="Gemini TTS not configured.")

src/config.py

@@ -38,5 +38,5 @@ DEEPGRAM_LANGUAGE: str = os.getenv("DEEPGRAM_LANGUAGE", "id")
 DEEPGRAM_ENDPOINTING_MS: int = int(os.getenv("DEEPGRAM_ENDPOINTING_MS", "300"))
 DEEPGRAM_UTTERANCE_END_MS: int = int(os.getenv("DEEPGRAM_UTTERANCE_END_MS", "2000"))
 
-CHIRP3_REGION: str = os.getenv("CHIRP3_REGION", "us")
+CHIRP3_REGION: str = os.getenv("CHIRP3_REGION", "us-central1")  # chirp_2 valid regions: us-central1, europe-west4, asia-southeast1
 CHIRP3_LANGUAGE: str = os.getenv("CHIRP3_LANGUAGE", "id-ID")

src/pipeline.py

@@ -94,6 +94,7 @@ class VoicePipeline:
         self._tts_task = asyncio.create_task(self._speak(text))
 
     async def _speak(self, text: str) -> None:
+        logger.info("TTS speak: provider=%s, text_len=%d, text=%r", self._tts_provider, len(text), text)
         async with self._tts_lock:
             try:
                 stream = (

src/stt/chirp3_client.py

@@ -19,34 +19,39 @@ OnTranscriptCallback = Callable[[str], Awaitable[None]]
 
 _CHUNK_SIZE = 24 * 1024  # stay under the 25 KB gRPC streaming limit
 
+def _make_client() -> SpeechClient:
+    return SpeechClient(
+        credentials=_credentials,
+        client_options=ClientOptions(api_endpoint=f"{CHIRP3_REGION}-speech.googleapis.com"),
+    )
+
+
+def _make_config_request() -> cloud_speech_types.StreamingRecognizeRequest:
+    recognition_config = cloud_speech_types.RecognitionConfig(
+        auto_decoding_config=cloud_speech_types.AutoDetectDecodingConfig(),
+        language_codes=[CHIRP3_LANGUAGE],
+        model="chirp_2",
+    )
+    streaming_config = cloud_speech_types.StreamingRecognitionConfig(
+        config=recognition_config
+    )
+    return cloud_speech_types.StreamingRecognizeRequest(
+        recognizer=f"projects/{GOOGLE_PROJECT_ID}/locations/{CHIRP3_REGION}/recognizers/_",
+        streaming_config=streaming_config,
+    )
+
 
 async def transcribe_audio(data: bytes, mimetype: str = "audio/wav") -> dict:
     """Transcribes a full audio file using Chirp 3 (Google Cloud Speech-to-Text V2)."""
     def _run() -> str:
-        logger.info("chirp3 transcribe_audio: data=%d bytes, region=%s, project=%s", len(data), CHIRP3_REGION, GOOGLE_PROJECT_ID)
-        client = SpeechClient(
-            credentials=_credentials,
-            client_options=ClientOptions(
-                api_endpoint=f"{CHIRP3_REGION}-speech.googleapis.com"
-            )
-        )
+        logger.info("chirp3 transcribe_audio: data=%d bytes, project=%s", len(data), GOOGLE_PROJECT_ID)
+        client = _make_client()
         chunks = [data[i: i + _CHUNK_SIZE] for i in range(0, len(data), _CHUNK_SIZE)]
         audio_requests = (
             cloud_speech_types.StreamingRecognizeRequest(audio=chunk)
             for chunk in chunks
         )
-        recognition_config = cloud_speech_types.RecognitionConfig(
-            auto_decoding_config=cloud_speech_types.AutoDetectDecodingConfig(),
-            language_codes=[CHIRP3_LANGUAGE],
-            model="chirp_3",
-        )
-        streaming_config = cloud_speech_types.StreamingRecognitionConfig(
-            config=recognition_config
-        )
-        config_request = cloud_speech_types.StreamingRecognizeRequest(
-            recognizer=f"projects/{GOOGLE_PROJECT_ID}/locations/{CHIRP3_REGION}/recognizers/_",
-            streaming_config=streaming_config,
-        )
+        config_request = _make_config_request()
 
         def requests():
             yield config_request
@@ -118,13 +123,7 @@ class Chirp3STTStreamer:
             logger.exception("Chirp 3 STT transcription failed")
 
     def _recognize(self, wav_data: bytes) -> str:
-        client = SpeechClient(
-            credentials=_credentials,
-            client_options=ClientOptions(
-                api_endpoint=f"{CHIRP3_REGION}-speech.googleapis.com"
-            )
-        )
-
+        client = _make_client()
         chunks = [
             wav_data[i : i + _CHUNK_SIZE]
             for i in range(0, len(wav_data), _CHUNK_SIZE)
@@ -133,19 +132,7 @@ class Chirp3STTStreamer:
             cloud_speech_types.StreamingRecognizeRequest(audio=chunk)
             for chunk in chunks
         )
-
-        recognition_config = cloud_speech_types.RecognitionConfig(
-            auto_decoding_config=cloud_speech_types.AutoDetectDecodingConfig(),
-            language_codes=[CHIRP3_LANGUAGE],
-            model="chirp_3",
-        )
-        streaming_config = cloud_speech_types.StreamingRecognitionConfig(
-            config=recognition_config
-        )
-        config_request = cloud_speech_types.StreamingRecognizeRequest(
-            recognizer=f"projects/{GOOGLE_PROJECT_ID}/locations/{CHIRP3_REGION}/recognizers/_",
-            streaming_config=streaming_config,
-        )
+        config_request = _make_config_request()
 
         def requests():
             yield config_request

src/tts/cartesia_client.py

@@ -32,7 +32,7 @@ async def synthesize_stream(text: str) -> AsyncIterator[bytes]:
         "Content-Type": "application/json",
     }
 
-    logger.info("Cartesia TTS: synthesizing '%s'", text[:60])
+    logger.info("Cartesia TTS: text_len=%d, text=%r", len(text), text)
     async with httpx.AsyncClient(timeout=30) as client:
         async with client.stream("POST", CARTESIA_TTS_URL, json=payload, headers=headers) as response:
             if response.status_code >= 400:

src/tts/gemini_client.py

@@ -44,7 +44,7 @@ async def synthesize_stream(text: str) -> AsyncIterator[bytes]:
     last_exc: Exception | None = None
 
     for model in models_to_try:
-        logger.info("Gemini TTS [%s]: synthesizing '%s'", model, text[:60])
+        logger.info("Gemini TTS [%s]: text_len=%d, text=%r", model, len(text), text)
 
         def _collect(model_name: str = model) -> list[bytes]:
             client = texttospeech.TextToSpeechClient(credentials=_credentials)
@@ -64,10 +64,17 @@ async def synthesize_stream(text: str) -> AsyncIterator[bytes]:
                     input=texttospeech.StreamingSynthesisInput(text=text)
                 )
 
-            chunks = [r.audio_content for r in client.streaming_synthesize(_gen())]
+            all_chunks = [r.audio_content for r in client.streaming_synthesize(_gen())]
+            chunks = [c for c in all_chunks if c]
+            total_bytes = sum(len(c) for c in chunks)
             logger.info(
-                "Gemini TTS [%s]: received %d chunks, %d bytes total",
-                model_name, len(chunks), sum(len(c) for c in chunks),
+                "Gemini TTS [%s]: received %d chunks (%d empty skipped), %d bytes total, aligned=%s, first32=%s",
+                model_name,
+                len(chunks),
+                len(all_chunks) - len(chunks),
+                total_bytes,
+                total_bytes % 2 == 0,
+                chunks[0][:32].hex() if chunks else "N/A",
             )
             return chunks