README.md

metadata

title: Speaker Diarization
emoji: 🔥
colorFrom: blue
colorTo: blue
sdk: docker
pinned: false
license: mit

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Real-Time Speaker Diarization

This project implements real-time speaker diarization using WebRTC, FastAPI, and Gradio. It automatically transcribes speech and identifies different speakers in real-time.

Architecture

The system is split into two components:

1. Model Server (Hugging Face Space): Runs the speech recognition and speaker diarization models
2. Signaling Server (Render): Handles WebRTC signaling for direct audio streaming from browser

Deployment Instructions

Deploy Model Server on Hugging Face Space

1. Create a new Space on Hugging Face (Docker SDK)
2. Upload all files from the Speaker-Diarization directory
3. In Space settings:
   • Set Hardware to CPU (or GPU if available)
   • Set the public visibility
   • Environment: Make sure Docker SDK is selected

Deploy Signaling Server on Render

1. Create a new Render Web Service
2. Connect to your GitHub repo containing the render-signal directory
3. Configure Render service:
   • Set Build Command: cd render-signal && pip install -r requirements.txt
   • Set Start Command: cd render-signal && python backend.py
   • Select Environment: Python 3
   • Set Environment Variables:
     • HF_SPACE_URL: Set to your Hugging Face Space URL (e.g., your-username-speaker-diarization.hf.space)

Update Configuration

After both services are deployed:

1. Update ui.py on your Hugging Face Space:

   • Change RENDER_SIGNALING_URL to your Render app URL (wss://your-app.onrender.com/stream)
   • Make sure HF_SPACE_URL matches your actual Hugging Face Space URL

2. Update backend.py on your Render service:

   • Set API_WS to your Hugging Face Space WebSocket URL (wss://your-username-speaker-diarization.hf.space/ws_inference)

Usage

1. Open your Hugging Face Space URL in a web browser
2. Click "Start Listening" to begin
3. Speak into your microphone
4. The system will transcribe your speech and identify different speakers in real-time

Technology Stack

• Frontend: Gradio UI with WebRTC for audio streaming
• Signaling: FastRTC on Render for WebRTC signaling
• Backend: FastAPI + WebSockets
• Models:
  • SpeechBrain ECAPA-TDNN for speaker embeddings
  • Automatic Speech Recognition for transcription

License

MIT