Add creds, Update used TTS model

.env.example

@@ -11,10 +11,13 @@ WAKE_WORD_ENABLED=false
 
 # Gemini TTS (optional — hanya dibutuhkan jika provider=gemini)
 GOOGLE_API_KEY=
-GEMINI_TTS_MODEL=gemini-2.5-flash-preview-tts
+# Comma-separated list — dirotasi round-robin per request untuk distribusi quota RPM
+GEMINI_TTS_MODELS=gemini-2.5-flash-tts
+# GEMINI_TTS_MODELS=gemini-2.5-flash-tts,gemini-3.1-flash-tts-preview,gemini-2.5-pro-tts
 GEMINI_TTS_VOICE=Autonoe
 GEMINI_TTS_LANGUAGE=id-ID
 
+GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
 
 # Google Cloud Project
 GOOGLE_PROJECT_NAME=

.gitignore

@@ -25,4 +25,6 @@ convert_audio.py
 API_CONTRACT_CHATBOT.md
 API_CONTRACT_VOICE.md
 HIGHLIGHT_VOICE.md
-HIGHLIGHT_STT_TTS.md
+HIGHLIGHT_STT_TTS.md
+
+credentials/

Dockerfile

@@ -12,7 +12,10 @@ RUN uv sync --no-dev --no-install-project
 # Copy source
 COPY . .
 
+COPY credentials/creds.json /app/credentials/creds.json
+
 ENV PATH="/app/.venv/bin:$PATH"
+ENV GOOGLE_APPLICATION_CREDENTIALS=/app/credentials/creds.json
 
 EXPOSE 7860
 

src/config.py

@@ -19,6 +19,13 @@ GOOGLE_PROJECT_NUMBER: str = os.getenv("GOOGLE_PROJECT_NUMBER", "")
 GOOGLE_CLOUD_LOCATION: str = os.getenv("GOOGLE_CLOUD_LOCATION", "us-central1")
 
 GEMINI_TTS_MODEL: str = os.getenv("GEMINI_TTS_MODEL", "gemini-2.5-flash-tts")
+GEMINI_TTS_MODELS: list[str] = [
+    m.strip() for m in os.getenv(
+        "GEMINI_TTS_MODELS",
+        "gemini-2.5-flash-tts",
+        # "gemini-2.5-flash-tts,gemini-3.1-flash-tts-preview,gemini-2.5-pro-tts",
+    ).split(",") if m.strip()
+]
 GEMINI_TTS_VOICE: str = os.getenv("GEMINI_TTS_VOICE", "Autonoe")
 GEMINI_TTS_LANGUAGE: str = os.getenv("GEMINI_TTS_LANGUAGE", "id-ID")
 

src/google_auth.py

@@ -0,0 +1,34 @@
+import os
+import logging
+from typing import Optional
+
+import google.auth
+import google.auth.credentials
+from google.oauth2 import service_account
+
+logger = logging.getLogger(__name__)
+
+_SCOPES = ["https://www.googleapis.com/auth/cloud-platform"]
+
+
+def get_google_credentials() -> google.auth.credentials.Credentials:
+    """
+    Resolution order:
+    1. GOOGLE_APPLICATION_CREDENTIALS env var set & file exists → service account JSON
+    2. Fallback → ADC (gcloud auth application-default login)
+    """
+    creds_path: Optional[str] = os.environ.get("GOOGLE_APPLICATION_CREDENTIALS")
+
+    if creds_path:
+        if not os.path.isfile(creds_path):
+            raise FileNotFoundError(
+                f"GOOGLE_APPLICATION_CREDENTIALS='{creds_path}' tapi file tidak ditemukan."
+            )
+        logger.info("Google auth: service account dari %s", creds_path)
+        return service_account.Credentials.from_service_account_file(
+            creds_path, scopes=_SCOPES
+        )
+
+    logger.info("Google auth: menggunakan ADC (gcloud login)")
+    credentials, _ = google.auth.default(scopes=_SCOPES)
+    return credentials

src/stt/chirp3_client.py

@@ -9,9 +9,12 @@ from google.cloud.speech_v2.types import cloud_speech as cloud_speech_types
 from google.api_core.client_options import ClientOptions
 
 from src.config import GOOGLE_PROJECT_ID, CHIRP3_REGION, CHIRP3_LANGUAGE
+from src.google_auth import get_google_credentials
 
 logger = logging.getLogger(__name__)
 
+_credentials = get_google_credentials()
+
 OnTranscriptCallback = Callable[[str], Awaitable[None]]
 
 _CHUNK_SIZE = 24 * 1024  # stay under the 25 KB gRPC streaming limit
@@ -20,7 +23,9 @@ _CHUNK_SIZE = 24 * 1024  # stay under the 25 KB gRPC streaming limit
 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"
             )
@@ -48,10 +53,13 @@ async def transcribe_audio(data: bytes, mimetype: str = "audio/wav") -> dict:
             yield from audio_requests
 
         transcripts: list[str] = []
+        response_count = 0
         for response in client.streaming_recognize(requests=requests()):
+            response_count += 1
             for result in response.results:
                 if result.alternatives:
                     transcripts.append(result.alternatives[0].transcript)
+        logger.info("chirp3 transcribe_audio: %d responses, %d transcripts", response_count, len(transcripts))
 
         transcript = " ".join(transcripts).strip()
         if transcript:
@@ -111,6 +119,7 @@ class Chirp3STTStreamer:
 
     def _recognize(self, wav_data: bytes) -> str:
         client = SpeechClient(
+            credentials=_credentials,
             client_options=ClientOptions(
                 api_endpoint=f"{CHIRP3_REGION}-speech.googleapis.com"
             )

src/tts/gemini_client.py

@@ -1,39 +1,86 @@
 import asyncio
 import logging
+import threading
 from typing import AsyncIterator
 
+from google.api_core import exceptions as gcp_exceptions
 from google.cloud import texttospeech
 
-from src.config import GEMINI_TTS_MODEL, GEMINI_TTS_VOICE, GEMINI_TTS_LANGUAGE
+from src.config import GEMINI_TTS_MODELS, GEMINI_TTS_VOICE, GEMINI_TTS_LANGUAGE
+from src.google_auth import get_google_credentials
 
 logger = logging.getLogger(__name__)
 
+_credentials = get_google_credentials()
+
 GEMINI_SAMPLE_RATE = 24000  # Cloud TTS Gemini model outputs PCM Linear16 at 24kHz
 
 
+class _ModelRotator:
+    def __init__(self, models: list[str]) -> None:
+        self._models = models
+        self._index = 0
+        self._lock = threading.Lock()
+
+    def get_rotation(self) -> list[str]:
+        """Returns models starting from current index, then advances for next call."""
+        with self._lock:
+            n = len(self._models)
+            start = self._index % n
+            self._index += 1
+            return self._models[start:] + self._models[:start]
+
+
+_rotator = _ModelRotator(GEMINI_TTS_MODELS)
+
+
 async def synthesize_stream(text: str) -> AsyncIterator[bytes]:
-    """Calls Cloud TTS Gemini model and yields raw PCM Linear16 chunks at 24kHz."""
-    logger.info("Gemini TTS: synthesizing '%s'", text[:60])
-
-    def _collect() -> list[bytes]:
-        client = texttospeech.TextToSpeechClient()
-        config_req = texttospeech.StreamingSynthesizeRequest(
-            streaming_config=texttospeech.StreamingSynthesizeConfig(
-                voice=texttospeech.VoiceSelectionParams(
-                    name=GEMINI_TTS_VOICE,
-                    language_code=GEMINI_TTS_LANGUAGE,
-                    model_name=GEMINI_TTS_MODEL,
+    """Calls Cloud TTS Gemini model and yields raw PCM Linear16 chunks at 24kHz.
+
+    Rotates across GEMINI_TTS_MODELS round-robin and falls back to the next model
+    on ResourceExhausted (quota exceeded).
+    """
+    models_to_try = _rotator.get_rotation()
+    last_exc: Exception | None = None
+
+    for model in models_to_try:
+        logger.info("Gemini TTS [%s]: synthesizing '%s'", model, text[:60])
+
+        def _collect(model_name: str = model) -> list[bytes]:
+            client = texttospeech.TextToSpeechClient(credentials=_credentials)
+            config_req = texttospeech.StreamingSynthesizeRequest(
+                streaming_config=texttospeech.StreamingSynthesizeConfig(
+                    voice=texttospeech.VoiceSelectionParams(
+                        name=GEMINI_TTS_VOICE,
+                        language_code=GEMINI_TTS_LANGUAGE,
+                        model_name=model_name,
+                    )
                 )
             )
-        )
 
-        def _gen():
-            yield config_req
-            yield texttospeech.StreamingSynthesizeRequest(
-                input=texttospeech.StreamingSynthesisInput(text=text)
+            def _gen():
+                yield config_req
+                yield texttospeech.StreamingSynthesizeRequest(
+                    input=texttospeech.StreamingSynthesisInput(text=text)
+                )
+
+            chunks = [r.audio_content for r in client.streaming_synthesize(_gen())]
+            logger.info(
+                "Gemini TTS [%s]: received %d chunks, %d bytes total",
+                model_name, len(chunks), sum(len(c) for c in chunks),
             )
+            return chunks
 
-        return [r.audio_content for r in client.streaming_synthesize(_gen())]
+        try:
+            chunks = await asyncio.to_thread(_collect)
+            for chunk in chunks:
+                yield chunk
+            return
+        except gcp_exceptions.ResourceExhausted:
+            logger.warning("Gemini TTS quota exhausted for model '%s', trying next model", model)
+            last_exc = gcp_exceptions.ResourceExhausted(f"Quota exhausted on {model}")
+            continue
+        except Exception:
+            raise
 
-    for chunk in await asyncio.to_thread(_collect):
-        yield chunk
+    raise last_exc  # type: ignore[misc]