fix: Add PyTorch dependencies and fix base64 imports v0.5.7

- Add missing PyTorch, TorchAudio, and Transformers dependencies to requirements.txt
- Fix missing base64 import statements in WebSocket handler
- Add better HTTP 503 error handling for cold starts
- Update to version 0.5.7 with comprehensive fixes

🤖 Generated with [Claude Code](https://claude.ai/code)

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

requirements.txt

@@ -38,6 +38,11 @@ webrtcvad==2.0.10
 # Audio processing (compatible with existing numpy/librosa)
 librosa>=0.10.1
 
+# ML/AI dependencies for audio processing
+torch>=2.0.0
+torchaudio>=2.0.0
+transformers>=4.30.0
+
 # ASGI server for FastAPI integration
 uvicorn>=0.24.0
 numpy>=1.21.0

version.py

@@ -2,8 +2,8 @@
 Version information for ChatCal Voice-Enabled AI Assistant
 """
 
-__version__ = "0.5.6"
-__build_date__ = "2025-08-20T17:05:00"
+__version__ = "0.5.7"
+__build_date__ = "2025-08-21T00:30:00"
 __description__ = "VoiceCal - Voice-Enabled AI Scheduling Assistant with WebSocket services"
 
 def get_version_info():

webrtc/server/websocket_handler.py

@@ -148,7 +148,11 @@ class WebRTCHandler:
             logger.error(f"❌ STT service connection timeout for {client_id} - service may be cold starting or WebSocket endpoints not available")
             return False
         except websockets.exceptions.WebSocketException as e:
-            logger.error(f"❌ STT WebSocket error for {client_id}: {e}")
+            if "503" in str(e):
+                logger.error(f"❌ STT service unavailable (HTTP 503) for {client_id} - service may be cold starting")
+                logger.info(f"🔄 Try again in a few moments - Hugging Face services need time to start")
+            else:
+                logger.error(f"❌ STT WebSocket error for {client_id}: {e}")
             logger.info(f"🔍 Debug: Attempted connection to {self.stt_websocket_url}")
             return False
         except Exception as e:
@@ -290,6 +294,7 @@ class WebRTCHandler:
             
             if response_data.get("type") == "tts_audio_response":
                 # Decode base64 audio data
+                import base64
                 audio_b64 = response_data.get("audio_data", "")
                 audio_bytes = base64.b64decode(audio_b64)
                 logger.info(f"🔊 TTS audio received: {len(audio_bytes)} bytes")
@@ -323,6 +328,7 @@ class WebRTCHandler:
             
             if audio_data:
                 # Convert audio to base64 for WebSocket transmission
+                import base64
                 audio_b64 = base64.b64encode(audio_data).decode('utf-8')
                 
                 # Send audio playback message to client