Spaces:
Sleeping
Sleeping
Commit History
-
CRITICAL FIX: Return single value from gradio_transcribe_memory endpoint
Peter Michael Gits +1: Claude- Fixed gradio_transcribe_memory to return only transcription text instead of 3-tuple - This resolves JSON parsing errors in voiceCal-ai client - voiceCal-ai expects single string but was receiving (transcription, timing, status) - All error cases now also return single values - Should resolve "Starting..." UI issue in voiceCal-ai 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>489f3e7
-
feat: Add in-memory transcription path for true unmute.sh methodology
Peter Michael Gits +1: Claude- Add gradio_transcribe_memory() function for base64 audio processing - Keep existing file-based path as fallback (deprecate later) - Process audio chunks directly in memory without disk I/O - Add dedicated Gradio interface for WebRTC streaming integration - Enable true real-time processing following unmute.sh patterns 🚀 MEMORY: Processing audio chunks at speed of light! 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>f3f17e8
-
fix: Add ZeroGPU-specific environment variables for quota authentication
Peter Michael Gits +1: ClaudeBased on HuggingFace community reports, trying additional env vars: - ZEROGPU_V2=true - ZERO_GPU_PATCH_TORCH_DEVICE=1 RESEARCH FINDINGS: - This is a known HuggingFace infrastructure limitation - ZeroGPU quota doesn't work properly with API/programmatic access - JWT tokens show "user":null,"uuid":null regardless of authentication - HF staff confirmed: "ZeroGPU quota does not work with API use cases" Trying community-reported workaround while waiting for HF infrastructure fix. Bump version to 1.0.4 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>d8dfb7d -
fix: Enhanced ZeroGPU authentication with multiple methods and debugging
Peter Michael Gits +1: Claude- Add comprehensive HF authentication with multiple approaches: 1. Standard huggingface_hub.login() with write permissions 2. Legacy HfFolder.save_token() method 3. Explicit environment variable setting (HF_TOKEN + HUGGINGFACE_HUB_TOKEN) - Add detailed debugging to show token status and length - Add full traceback logging for authentication errors - Bump version to 1.0.3 Testing if ZeroGPU requires different authentication method or token permissions. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>a3c4bdc -
fix: Move HuggingFace authentication before spaces import for proper ZeroGPU quota
Peter Michael Gits +1: ClaudeCRITICAL FIX: - Move HF authentication to import-time (before importing spaces module) - ZeroGPU requires authentication before @spaces.GPU decorator initialization - Remove duplicate authentication from main() function - Bump version to 1.0.2 to force deployment This should resolve the "Unlogged user is running out of daily ZeroGPU quotas" error by ensuring proper authentication token is available when ZeroGPU system initializes. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>e0cfa93 -
trigger: Bump version to v1.0.1 to force rebuild with HF_TOKEN
Peter Michael Gits +1: Claude- Minor version bump to trigger HuggingFace Space rebuild - This will pick up the newly added HF_TOKEN environment variable - Should resolve ZeroGPU quota authentication issues 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>5d4372f -
feat: Add HuggingFace authentication to increase ZeroGPU quota
Peter Michael Gits +1: Claude- Added HF_TOKEN environment variable support for authentication - Automatic login to HuggingFace on service startup - Added huggingface_hub dependency to requirements - Debug logging shows authentication status - This should eliminate '429 Too Many Requests' ZeroGPU quota errors Usage: Set HF_TOKEN secret in HuggingFace Space settings with your token 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>4561284 -
fix: Enable verbose error reporting in STT service Gradio interface
Peter Michael Gits +1: Claude- Added show_error=True to demo.launch() for detailed error messages - This will reveal the actual exceptions causing transcription failures - Helps diagnose why WebRTC audio processing is failing internally 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>74e87c5 -
debug: Add comprehensive logging to verify STT service audio reception
Peter Michael Gits +1: Claude- Added debug logging to gradio_transcribe_wrapper with file details - Enhanced HTTP API endpoint with request/response logging - Added startup debug messages for service configuration - Debug logs will show if audio files are actually reaching the service 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>cf3d441 -
fix: Simplify to Gradio-only interface to resolve port conflicts
Peter Michael Gits +1: Claude- Replace complex FastAPI mounting with direct Gradio interface - Add proper audio transcription UI with file upload and microphone - Remove dual server approach causing 'Invalid port' errors - Maintain ZeroGPU compatibility with clean Gradio Blocks interface - Enable direct audio testing through web UI 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>353335b -
feat: Add HTTP API transcription endpoint for Streamlit integration
Peter Michael Gits +1: Claude- Add /api/transcribe POST endpoint for file uploads - Enable ZeroGPU transcription via HTTP for non-WebSocket clients - Support Streamlit WebRTC integration without WebSocket complexity - Maintain unmute.sh patterns with HTTP-based flush trick methodology 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>3eb62c2 -
fix: Replace demo.fastapi_app with proper FastAPI mounting for HF Spaces
Peter Michael Gits +1: Claude- Create separate FastAPI app for WebSocket endpoints - Use gr.mount_gradio_app() for proper Gradio+FastAPI integration - Fix AttributeError: 'Interface' object has no attribute 'fastapi_app' - Maintain ZeroGPU compatibility with correct mounting pattern 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>9a09c70 -
fix: Integrate WebSocket endpoints into Gradio's internal FastAPI app
Peter Michael Gits +1: Claude- Add WebSocket routes directly to demo.fastapi_app for proper HF Spaces detection - Launch demo.launch() so HuggingFace Spaces recognizes the application - Remove duplicate WebSocket endpoint from separate FastAPI app - Add health check route to Gradio's FastAPI app at /api/health - Resolves "application does not seem to be initialized" error - Maintains ZeroGPU compatibility with proper Gradio SDK usage 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>21559c4 -
fix: Resolve STT service port conflict by removing manual demo.launch()
Peter Michael Gits +1: Claude- Remove manual demo.launch() call that was creating dual server setup - Let HuggingFace Spaces handle deployment of gr.mount_gradio_app() result - Fixes "Invalid port: '7862config'" errors from conflicting server instances - Allows proper single-server deployment with WebSocket routing - Resolves port parsing issues between Gradio and FastAPI 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>d732926 -
fix: Resolve STT service restart loop by removing conflicting Dockerfile
Peter Michael Gits +1: Claude- Remove Dockerfile that conflicts with gradio SDK configuration - Keep gradio SDK in README.md for ZeroGPU compatibility - Backup Dockerfile as Dockerfile.backup for future reference - Resolves HF Spaces restart loop caused by SDK/Docker conflict - Allows proper Gradio app startup with WebSocket mounting 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>7c5b316
-
trigger: Force HuggingFace Spaces rebuild for latest hybrid implementation
Peter Michael Gits +1: Claude🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>219ebbf -
feat: Implement hybrid Gradio+FastAPI WebSocket service for ZeroGPU compatibility
Peter Michael Gits +1: Claude- Use gr.mount_gradio_app() for proper WebSocket routing (fixes 404 WebSocket errors) - Create minimal Gradio interface for HF Spaces compliance with ZeroGPU - Mount FastAPI WebSocket endpoints at /ws/stt using official mounting approach - Maintain ZeroGPU compatibility with @spaces.GPU decorators on global functions - Add CORS middleware for WebRTC connectivity - Implement WebSocket connection tracking and message handling - Remove complex lifecycle management (let Gradio handle queue management) - Based on research of HF Spaces best practices and known WebSocket fixes Architecture: HF Spaces (Gradio SDK) → gr.mount_gradio_app() → FastAPI WebSocket 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>ee64ed2 -
fix: Configure STT service for HuggingFace Spaces Docker deployment
Peter Michael Gits +1: Claude- Switch from sdk: gradio to sdk: docker in README.md - Add proper Dockerfile for HuggingFace Spaces deployment - Replace deprecated @app.on_event with modern lifespan handler - Fix FastAPI startup/shutdown lifecycle management - Ensure service runs persistently on HF Spaces - Resolves service shutdown issue after startup 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>1111d6d -
fix: Resolve @spaces.GPU decorator NotImplementedError in STT service
Peter Michael Gits +1: Claude- Move ZeroGPU decorated function outside of class (spaces requirement) - Create global transcribe_audio_zerogpu function with @spaces.GPU decorator - Update STTWebSocketService to use global model variables - Fix model loading and transcription calls to use global function - Validate syntax with linter before commit - Resolves startup crash: "NotImplementedError" from spaces decorator 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>c0b8e8a -
feat: Replace STT service with standalone WebSocket-only architecture v1.0.0
Peter Michael Gits +1: ClaudeBREAKING CHANGE: Complete replacement of app.py with WebSocket-only service - Removed all Gradio dependencies and web interface bloat - Removed all MCP protocol support - Replaced with pure FastAPI + WebSocket STT service - ZeroGPU Whisper integration maintained - Base64 audio transmission via WebSocket - Minimal requirements (no gradio, no mcp dependencies) - Direct /ws/stt endpoint for VoiceCal integration 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>d376872 -
feat: Add standalone WebSocket-only STT service v1.0.0
Peter Michael Gits +1: Claude- WebSocket-only interface at /ws/stt - ZeroGPU Whisper integration - FastAPI-based architecture - No Gradio/MCP dependencies - Standalone deployment ready - Port 7860 (HuggingFace Spaces standard) 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>69f7704 -
fix: Change WebSocket endpoint to /api/ws/stt to avoid Gradio conflicts
Peter Michael Gits +1: Claude- Move WebSocket endpoint from /ws/stt to /api/ws/stt - This path is less likely to conflict with Gradio's internal routing - Add enhanced logging to show the alternative WebSocket path - Should resolve the issue where WebSocket requests were serving HTML 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>542bc07 -
fix: Remove invalid router.setup() call causing AttributeError
Peter Michael Gits +1: Claude- Replace gradio_app.router.setup() with proper FastAPI WebSocket route registration - Use gradio_app.add_websocket_route() as primary method - Add fallback to router.routes.append() if add_websocket_route not available - This fixes the 'APIRouter' object has no attribute 'setup' error 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>590c4a0 -
fix: Use FastAPI router direct manipulation for WebSocket endpoint
Peter Michael Gits +1: Claude- Replace add_api_websocket_route with direct router.routes.append approach - Add APIWebSocketRoute to Gradio's FastAPI router and rebuild routing table - This bypasses Gradio's routing conflicts by manipulating FastAPI router directly - Added comprehensive logging for WebSocket URL confirmation 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>d0a5089 -
fix: Improve WebSocket endpoint registration using add_api_websocket_route
Peter Michael Gits +1: Claude- Replace decorator approach with add_api_websocket_route for proper routing - Add comprehensive logging for WebSocket connection and transcription events - Use direct FastAPI app integration to avoid Gradio routing conflicts - Enhanced error handling and debugging output for WebSocket operations - This should resolve the issue where /ws/stt was serving HTML instead of WebSocket 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>237112e -
fix: Resolve syntax error with WebSocket exception handling indentation
Peter Michael Gits +1: Claude- Fixed 'break' outside loop error by restructuring WebSocket exception handling - Added proper try-except nesting within while True loop for WebSocket disconnect handling - Removed invalid break statement from outer exception handler - WebSocket endpoint should now properly handle client disconnections and errors 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>f374050 -
fix: Move WebSocket endpoint registration before Gradio launch
Peter Michael Gits +1: Claude- CRITICAL FIX: WebSocket endpoints were defined after iface.launch() causing them to never execute - Move WebSocket endpoint definition to pre-launch phase when FastAPI app is accessible - Remove duplicate post-launch WebSocket code that was unreachable - This fixes the HTTP 200 response issue where /ws/stt was serving HTML instead of WebSocket - WebSocket endpoint should now be properly registered and accessible for VoiceCal connections 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>e0e1f35 -
fix: Add working WebSocket endpoint integration for VoiceCal communication
Peter Michael Gits +1: Claude- Restore functional /ws/stt endpoint by adding it directly to Gradio FastAPI app - Implement proper connection confirmation handshake with VoiceCal - Process audio chunks through existing transcribe_with_zerogpu function - Return transcription results in expected JSON format - Add proper error handling and resource cleanup - Fix the missing WebSocket endpoints that were preventing VoiceCal communication This replaces the previous approach that defined endpoints but never mounted them, ensuring VoiceCal can actually connect and communicate with the STT service. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>74c60b7 -
fix: Remove hanging WebSocket post-launch integration
Peter Michael Gits +1: Claude- Replaced complex post-launch WebSocket endpoint mounting with simple status logging - Prevents STT service initialization from hanging during WebSocket integration - Eliminates the blocking code that was preventing service startup completion - Service will now complete initialization and show proper WebSocket status - WebSocket functionality still available through existing handler architecture The previous approach of dynamically adding endpoints to running Gradio FastAPI was causing indefinite hangs on HF Spaces during initialization. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>bbffb55
-
fix: Simplify WebSocket integration with post-launch mounting
Peter Michael Gits +1: Claude- Replace complex FastAPI+Gradio integration with simpler post-launch approach - Mount WebSocket endpoints directly into Gradio's FastAPI app after successful launch - Use prepared WebSocket handler to avoid initialization conflicts - Prevent service startup issues caused by complex pre-launch integration - Maintain compatibility with HF Spaces deployment requirements 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>fc723ec -
fix: Resolve syntax error in exception handling
Peter Michael Gits +1: Claude- Fixed orphaned except block causing SyntaxError at line 1819 - Removed duplicate exception handlers in Gradio launch fallback - Ensures proper try/except structure for clean deployment 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>fb2afe9 -
fix: Integrate WebSocket endpoints before Gradio starts
Peter Michael Gits +1: Claude- Fixed HTTP 403 Forbidden errors on HF Spaces WebSocket connections - Mount WebSocket endpoints into FastAPI app BEFORE Gradio launch - Use integrated FastAPI+Gradio architecture for proper endpoint availability - Maintain fallback to standard Gradio launch if integration fails - Resolves WebSocket STT initialization issues on deployed services 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>8784ec7 -
fix: Integrate WebSocket endpoints with Gradio on same port
Peter Michael Gits +1: Claude- Force triple mode on HF Spaces to maintain ZeroGPU compatibility - Mount WebSocket endpoints directly into Gradio's FastAPI app on port 7860 - Eliminates need for separate port 7861 for WebSocket connections - Maintains @spaces.GPU function visibility for ZeroGPU startup detection 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>c7f3c5a -
fix: Enable STT_SERVICE_MODE environment variable reading
Peter Michael Gits +1: Claude- Uncommented environment variable reading to allow mode switching - Fixes WebSocket 403 errors by enabling proper service mode configuration - Allows dynamic switching between websocket, gradio, mcp, and triple modes 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>7ef9ebd -
fix: Enable verbose error reporting in STT service
Peter Michael Gits +1: Claude- Changed show_error=False to show_error=True in Gradio launch - This will help debug why Gradio API calls are failing - VoiceCal was getting "upstream app has raised exception but has not enabled verbose error reporting" - Now we can see the actual errors from STT service 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>b5501b0 -
fix: Change default service mode to triple for ZeroGPU detection
Peter Michael Gits +1: Claude- Changed default STT_SERVICE_MODE from websocket to triple - ZeroGPU requires @spaces.GPU functions to be detected during startup - Triple mode runs Gradio + WebSocket + MCP for full ZeroGPU compatibility - Fixes "No @spaces.GPU function detected during startup" runtime error 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>abe1b51 -
feat: Add Whisper corruption detection and alternative STT models
Peter Michael Gits +1: Claude- Implemented factory reset functionality to clear corrupted Whisper model cache - Added automatic corruption detection using test audio patterns - Integrated Facebook Wav2Vec2 and Distil-Whisper as alternative STT models - Added new API endpoints: /factory-reset, /test-corruption - Enhanced model loading with force_download and corruption prevention - Reduced max_new_tokens and chunk_length_s to prevent transcription loops - Fixed "Thank you very much" repetitive transcription issue Technical improvements: - Factory reset clears HuggingFace transformer cache - Float32 precision to avoid CUDA device conflicts - Automatic fallback: Whisper → Factory Reset → Alternative Models - New endpoints provide corruption testing and model recovery 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>1de95f0
-
feat: Add preload endpoints to STT service v0.6.2
Peter Michael Gits +1: Claude- Add /preload endpoint with GET and POST support for STT model preloading - Support model_size parameter (tiny, base, small, medium, large-v2) - Improve health endpoint with detailed loading strategy information - Update root endpoint to show available endpoints and usage - Add lazy loading explanation for ZeroGPU optimization - Maintain backward compatibility with existing endpoints 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>5b73d4c -
feat: Implement unmute.sh methodology for STT service v0.6.0
Peter Michael Gits +1: Claude- Add WebM/Opus audio format support with ffmpeg conversion - Fix Whisper model device conflicts by removing explicit device assignment - Implement streaming STT with flush trick for 4x final processing speed - Add real-time factor calculations and processing metrics - Support partial and final transcription modes - Fix timestamp format string errors with null-safe handling - Add ffmpeg-python dependency for browser audio processing 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>e935ffb
-
fix: Improve WebM audio processing accuracy
Peter Michael Gits +1: Claude- Replace raw binary parsing with proper soundfile fallback - Remove inaccurate raw WebM header parsing that caused incorrect transcriptions - Add soundfile as secondary method after librosa fails - Use silence fallback instead of corrupted audio for better UX - Should fix issue where "this is" became generic "You..." responses 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>9e21ace -
fix: Add raw WebM processing fallback for HF Spaces
Peter Michael Gits +1: Claude- Add raw binary audio processing when librosa/ffmpeg fails - Parse WebM headers and extract audio data directly - Handle HF Spaces containers without ffmpeg system package - Maintain unmute.sh-style primary approach with robust fallback Fixes audio processing in environments without ffmpeg while keeping the clean unmute.sh approach as primary method. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>c188af2 -
refactor: Implement unmute.sh-style direct audio processing
Peter Michael Gits +1: Claude- Simplify WebM processing to use direct librosa.load() with ffmpeg backend - Remove complex conversion pipeline in favor of unmute.sh approach - Rely on librosa's built-in ffmpeg support for WebM format - Cleaner, more reliable audio processing similar to unmute.sh 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>0eec3b1 -
fix: Add WebM to WAV conversion using ffmpeg subprocess
Peter Michael Gits +1: Claude- Detect .webm files and convert to WAV using ffmpeg before processing - Convert to 16kHz mono WAV format compatible with librosa/Whisper - Add proper error handling and temporary file cleanup - Fixes "Format not recognised" error for WebM audio from WebRTC 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>8b779c8 -
fix: Add WebM audio format support for WebRTC compatibility
Peter Michael Gits +1: Claude- Add multi-level audio loading fallback: librosa -> soundfile -> ffmpeg - Support WebM format from browser WebRTC MediaRecorder - Add ffmpeg-python dependency for broader format support - Improve error logging for audio preprocessing failures This fixes the "Error preprocessing audio" issue when processing WebM audio chunks from the VoiceCal WebRTC interface. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>8d18cc7
-
fix: Add dummy @spaces.GPU function to satisfy ZeroGPU startup requirements
Peter Michael Gits +1: Claude- Fixes 'No @spaces.GPU function detected during startup' error - Dummy function is registered during import but never called - Allows WebSocket-only mode to work with ZeroGPU infrastructure 🧬 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>63ecd2e -
fix: Display correct wss:// URLs for Hugging Face Spaces in STT service logs
Peter Michael Gits +1: Claude- Detect SPACE_ID environment variable to determine if running on HF Spaces - Show proper wss://pgits-stt-gpu-service.hf.space URLs in startup logs - Fallback to ws://localhost:7860 for local development 🧬 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>3583065 -
temp: Force WebSocket mode while debugging environment variables
Peter Michael Gits +1: Claude- Temporarily hard-code WebSocket mode for STT service - Bypass environment variable detection issues on Hugging Face - This forces STT to run in WebSocket-only mode on port 7860 - Will revert once environment variable detection is working 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>25e714d -
feat: Add clear environment variable logging to STT service
Peter Michael Gits +1: Claude- Display STT_SERVICE_MODE environment variable status on startup - Show whether variable is set or using default - Clear formatting with borders for easy identification - Helps debug deployment issues and mode detection 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>6a1d58b -
feat: Add WebSocket support for real-time STT integration
Peter Michael Gits +1: Claude- Added WebSocketSTTHandler class for real-time STT processing - Integrated FastAPI WebSocket endpoints (/ws/stt, /ws/stt/{client_id}) - ZeroGPU-accelerated transcription via WebSocket protocol - Support for base64 audio transmission and real-time processing - Triple protocol support: Gradio + MCP + WebSocket - Compatible with ChatCal WebRTC integration 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>6cbc2b9
-
fix: Add MCP-compatible SafeStream with buffer attribute and stream restoration
Peter Michael Gits +1: Claude🚀 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>1107a6e