Real-time speech-to-text with Voice Activity Detection (VAD), powered by JigsawStack's STT API.

• Voice Activity Detection: Automatically detects when you start/stop speaking
• Audio Accumulation: Sends growing audio clips every 1s for evolving transcriptions
• Real-time Waveform: Visual feedback showing speech activity with gradient animation
• Segment Management: Finalizes segments after 3 seconds of silence
• Modern UI: Clean interface with final (black) and interim (gray italic) text display

1. Install dependencies:

npm install or yarn install

2. Set your JigsawStack API key in .env:

JIGSAWSTACK_API_KEY=your_api_key_here

3. Run both servers in separate terminals:

Terminal 1: WebSocket Server (Port 8080)

npm start or yarn start

Terminal 2: HTTP Server (Port 3000)

npm run start:http or yarn start:http

4. Open http://localhost:3000/index.html and allow microphone access

1. VAD monitors audio energy levels to detect speech
2. When speech starts, audio is accumulated in a buffer
3. Every 1 second, accumulated audio is sent for transcription
4. Transcriptions evolve and improve as more audio context is added
5. After 3 seconds of silence, the segment is finalized
6. Process restarts fresh for the next speech segment

Edit the config in index.html (line 382):

{
  language: "en",        // Language code
  encoding: "wav",       // Audio format (wav, webm, pcm16)
  interimResults: true,  // Show evolving transcriptions
  format: "text"         // Output format (text or json)
}

• server.ts - WebSocket server handling STT requests
• http-server.ts - Serves the web interface
• index-vad.html - VAD-based client with waveform visualization
• index.html - Simple interval-based client (no VAD)

• JigsawStack STT API
• Web Audio API for VAD and PCM capture
• WebSockets for real-time communication
• TypeScript + Node.js