fix: make voice-service startup graceful and fix device config

- Wrap model loading in try/except so server starts even if models fail
- Fix device env var mapping (unified 'device' field instead of 'whisper_device')
- Default Whisper model to 'base' instead of 'large-v3' (3GB) for CPU deployment
- Increase healthcheck start_period to 120s for model download time

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

deploy/docker/docker-compose.yml

@@ -300,15 +300,15 @@ services:
     environment:
       - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
       - AGENT_SERVICE_URL=http://agent-service:3002
-      - WHISPER_MODEL=${WHISPER_MODEL:-large-v3}
+      - WHISPER_MODEL=${WHISPER_MODEL:-base}
       - KOKORO_MODEL=${KOKORO_MODEL:-kokoro-82m}
       - DEVICE=${VOICE_DEVICE:-cpu}
     healthcheck:
       test: ["CMD-SHELL", "python3 -c \"import urllib.request; urllib.request.urlopen('http://localhost:3008/docs')\""]
       interval: 30s
-      timeout: 5s
-      retries: 3
-      start_period: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 120s
     depends_on:
       - agent-service
     networks:

packages/services/voice-service/src/api/main.py

@@ -25,25 +25,33 @@ app.include_router(twilio_router, prefix="/api/v1/twilio", tags=["twilio"])
 
 @app.on_event("startup")
 async def startup():
-    """Load models on startup."""
+    """Load models on startup (graceful — server starts even if models fail)."""
     from ..config.settings import settings
-    from ..stt.whisper_service import WhisperSTTService
-    from ..tts.kokoro_service import KokoroTTSService
-    from ..vad.silero_service import SileroVADService
 
     print("Voice service starting up...")
+    print(f"  Device: {settings.device}")
+    print(f"  Whisper model: {settings.whisper_model}")
 
     # Initialize STT service
+    try:
+        from ..stt.whisper_service import WhisperSTTService
+
         stt = WhisperSTTService(
             model=settings.whisper_model,
-        device=settings.whisper_device,
+            device=settings.device,
             language=settings.whisper_language,
         )
         await stt.initialize()
         app.state.stt = stt
         print(f"STT model loaded: {settings.whisper_model}")
+    except Exception as e:
+        app.state.stt = None
+        print(f"WARNING: STT model failed to load: {e}")
 
     # Initialize TTS service
+    try:
+        from ..tts.kokoro_service import KokoroTTSService
+
         tts = KokoroTTSService(
             model=settings.kokoro_model,
             voice=settings.kokoro_voice,
@@ -51,12 +59,23 @@ async def startup():
         await tts.initialize()
         app.state.tts = tts
         print(f"TTS model loaded: {settings.kokoro_model}")
+    except Exception as e:
+        app.state.tts = None
+        print(f"WARNING: TTS model failed to load: {e}")
 
     # Initialize VAD service
+    try:
+        from ..vad.silero_service import SileroVADService
+
         vad = SileroVADService()
         await vad.initialize()
         app.state.vad = vad
         print("VAD model loaded: Silero VAD")
+    except Exception as e:
+        app.state.vad = None
+        print(f"WARNING: VAD model failed to load: {e}")
+
+    print("Voice service startup complete.")
 
 
 @app.on_event("shutdown")

packages/services/voice-service/src/config/settings.py

@@ -17,14 +17,16 @@ class Settings(BaseSettings):
     agent_service_url: str = "http://agent-service:3002"
 
     # STT (faster-whisper)
-    whisper_model: str = "large-v3"
-    whisper_device: str = "cuda"  # "cuda" or "cpu"
+    whisper_model: str = "base"
     whisper_language: str = "zh"
 
     # TTS (Kokoro)
     kokoro_model: str = "kokoro-82m"
     kokoro_voice: str = "zh_female_1"
 
+    # Device (cpu or cuda)
+    device: str = "cpu"
+
     # Twilio
     twilio_account_sid: str = ""
     twilio_auth_token: str = ""

packages/services/voice-service/src/stt/whisper_service.py

@@ -11,7 +11,7 @@ import numpy as np
 class WhisperSTTService:
     """Speech-to-Text service using faster-whisper."""
 
-    def __init__(self, model: str = "large-v3", device: str = "cuda", language: str = "zh"):
+    def __init__(self, model: str = "base", device: str = "cpu", language: str = "zh"):
         self.model_name = model
         self.device = device
         self.language = language
@@ -21,11 +21,14 @@ class WhisperSTTService:
         """Load the Whisper model."""
         from faster_whisper import WhisperModel
 
+        compute_type = "float16" if self.device == "cuda" else "int8"
         try:
             self._model = WhisperModel(
-                self.model_name, device=self.device, compute_type="float16"
+                self.model_name, device=self.device, compute_type=compute_type
             )
-        except Exception:
+        except Exception as e:
+            print(f"Failed to load Whisper model on {self.device}: {e}")
+            if self.device != "cpu":
                 self._model = WhisperModel(
                     self.model_name, device="cpu", compute_type="int8"
                 )