Commit History
-
fix: Configure MediaRecorder with consistent WebM/Opus format
Peter Michael Gits +1: Claude- Fix root cause of "Thank you very much" repetitive transcriptions - Configure MediaRecorder with audio/webm;codecs=opus for consistent encoding - Add 128kbps audio bitrate for good quality - Prevent "EBML header parsing failed" errors in STT service - Follow unmute.sh methodology for WebM format consistency - Add fallback handling for browsers without WebM/Opus support Resolves STT service logs showing corrupted WebM files causing ffmpeg conversion failures and silence fallback transcriptions. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>5d2f2af
-
feat: Implement unmute.sh streaming STT methodology v0.7.0
Peter Michael Gits +1: ClaudeUNMUTE.SH METHODOLOGY IMPLEMENTATION: - Buffer audio chunks during recording (like unmute.sh streaming) - Apply flush trick when recording stops - combine all chunks into complete audio - Process complete audio with is_final=True for proper transcription finalization - Clear buffer after processing to prevent chunk contamination - Add flush trick markers and metadata following unmute.sh patterns Key Changes: ✅ Buffer chunks during recording instead of processing individually ✅ Implement flush trick: combine chunks + process with is_final=True ✅ Send only final transcription after complete audio processing ✅ Clear buffers after each recording session ✅ Add unmute.sh-style metadata (is_final, chunks_processed) This should fix the 'Thank you very much' issue by properly implementing unmute.sh's streaming methodology with flush trick for clean transcriptions. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>7d32abe
-
Disable demo TTS responses due to poor STT accuracy
Peter Michael Gits +1: Claude- Only first 3 words were being transcribed correctly - Demo TTS responses not working properly with incomplete transcriptions - Keep WebRTC demo page functional for STT testing without TTS interference 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>bc9a856 -
fix: Add PyTorch dependencies and fix base64 imports v0.5.7
Peter Michael Gits +1: Claude- 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>e985487
-
temp: Update WebSocket URLs to standard port 7860
Peter Michael Gits +1: Claude- Revert STT back to port 7860 (services now forced to WebSocket mode) - Both STT and TTS should run WebSocket endpoints on port 7860 - Temporary fix while debugging environment variable detection - Should resolve TTS connection issues 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>22227a1 -
feat: Add enhanced TTS connection error handling and diagnostics
Peter Michael Gits +1: Claude- Add timeout handling for TTS WebSocket connections - Improve error messages with specific diagnostic hints - Detect when TTS service is in wrong mode (Gradio vs WebSocket) - Guide users to set TTS_SERVICE_MODE=websocket environment variable - Better debugging for TTS connection failures 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>57bc78b -
fix: Update STT WebSocket URL to use port 7861
Peter Michael Gits +1: Claude- STT service is running in triple mode with WebSocket on port 7861 - Update VoiceCal to connect to correct STT WebSocket endpoint - Fixes STT connection issues while TTS service configuration is pending 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>f118868 -
feat: Deploy complete VoiceCal application with all files v0.5.6
Peter Michael Gits +1: Claude- Add all application files: app.py, requirements.txt, core modules - Include WebRTC integration and voice service components - Complete Gradio application ready for Hugging Face deployment - Comprehensive README with VoiceCal branding and documentation - All dependencies and Docker configuration included 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>5e8a657