Spaces:
Running
Running
| # STT Provider Configuration | |
| # Options: "speechmatics", "deepgram", or "deepgram-flux" | |
| STT_PROVIDER=speechmatics | |
| # Speechmatics API Key | |
| # Get your API key from: https://portal.speechmatics.com/ | |
| SPEECHMATICS_API_KEY=your_speechmatics_api_key_here | |
| # Deepgram API Key (only needed if STT_PROVIDER=deepgram or deepgram-flux) | |
| # Get your API key from: https://console.deepgram.com/ | |
| DEEPGRAM_API_KEY=your_deepgram_api_key_here | |
| # ElevenLabs API Key | |
| # Get your API key from: https://elevenlabs.io/app/settings/api-keys | |
| ELEVENLABS_API_KEY=your_elevenlabs_api_key_here | |
| # ElevenLabs Voice ID (optional, defaults to custom voice) | |
| # Find voice IDs at: https://elevenlabs.io/app/voices | |
| ELEVENLABS_VOICE_ID=ry8mpwRw6nugb2qjP0tu | |
| # DeepInfra API Key (for Qwen LLM and Gating Layer) | |
| # Get your API key from: https://deepinfra.com/ | |
| DEEPINFRA_API_KEY=your_deepinfra_api_key_here | |
| # Optional: Override default models | |
| # DEEPINFRA_MODEL=Qwen/Qwen3-235B-A22B-Instruct-2507 # Main LLM (default) | |
| # DEEPINFRA_GATING_MODEL=meta-llama/Llama-3.2-3B-Instruct # Gating Layer (default) | |
| # Pipecat FastAPI service URL (for frontend to connect) | |
| NEXT_PUBLIC_PIPECAT_URL=http://localhost:7860 | |
| # Pipecat FastAPI service configuration | |
| PIPECAT_HOST=localhost | |
| PIPECAT_PORT=7860 | |
| # Mem0 API Key (optional, enables long-term memory) | |
| # Get one from: https://docs.mem0.ai/ | |
| MEM0_API_KEY=your_mem0_api_key_here | |
| # TTS Provider Configuration | |
| # Options: "elevenlabs" (cloud, requires API key) or "qwen3" (local, free) | |
| TTS_PROVIDER=qwen3 | |
| # Qwen3-TTS Configuration (only needed if TTS_PROVIDER=qwen3) | |
| # Model: 0.6B (faster, less memory) or 1.7B (better quality) | |
| QWEN3_TTS_MODEL=Qwen/Qwen3-TTS-12Hz-0.6B-Base | |
| # Device: "mps" for Mac, "cuda" for NVIDIA GPU, "cpu" for CPU | |
| QWEN3_TTS_DEVICE=mps | |
| # Reference audio file for voice cloning (relative to project root) | |
| QWEN3_TTS_REF_AUDIO=assets/audio/tars-clean-compressed.mp3 | |
| # Emotional State Monitoring | |
| # Continuously analyzes video for confusion/hesitation/frustration | |
| # Triggers TARS to offer help proactively | |
| EMOTIONAL_MONITORING_ENABLED=true | |
| # How often to sample video frames (in seconds) | |
| EMOTIONAL_SAMPLING_INTERVAL=3.0 | |
| # How many consecutive negative states before intervention | |
| EMOTIONAL_INTERVENTION_THRESHOLD=2 | |