Empy Recorder

Electron + React desktop app that captures microphone + system audio, splits channels, and transcribes both sides with Groq Whisper into a diarized JSON.

Open-source recorder you can extend to build your own note-taker implementation.

Features

• Stereo capture (mic = left, system = right) via MediaRecorder (WebM/Opus)
• Stereo split to mono FLAC (16 kHz) using ffmpeg
• Dual-channel transcription with Groq Whisper and merged speakers/utterances
• Session management: open folder, download transcript, cancel and delete
• Optional basic echo reduction via a WebRTC AEC3 AudioWorklet
• Recordings stored under DATA_DIR/recordings/YYYY-MM-DD/SESSION_ID/
• 100 MB safety limit per recording to avoid long transcriptions

Audio capture

• Microphone: getUserMedia with Web Audio routing
• System audio: getDisplayMedia + electron-audio-loopback
• Recording: ChannelMergerNode -> MediaRecorder (Opus in WebM)

Requirements

• Node.js >= 20.9
• macOS 12.3+ or Windows 10+ (target platforms)
• Microphone + system audio permissions
  • macOS: Screen Recording + Microphone access in System Settings
• Groq API key for transcription

Quick start

1. Install deps: npm install
2. Create a local .env (or .env.prod for packaging):
   • GROQ_API_KEY=... (required)
   • WHISPER_MODEL=whisper-large-v3 (optional)
   • DATA_DIR=.Recordings (optional, overrides data root)
   • FORCE_DEV_DATA_DIR=true (optional, keep data inside project root)
3. Run: npm run start

Data layout

DATA_DIR/
  recordings/
    YYYY-MM-DD/
      SESSION_ID/
        stereo.webm
        mic.flac
        system.flac
        transcript.json

Scripts

• npm run start - dev app (Vite + Electron Forge)
• npm run lint - TypeScript typecheck
• npm run package - package app
• npm run make - build installers
• npm run build:aec - build aec-wasm (Rust + wasm-pack)

Architecture

• src/renderer - React UI, capture pipeline, optional AEC3 worklet
• src/preload.ts - IPC bridge for renderer <-> main
• src/main - session storage, ffmpeg split, Groq transcription, file I/O
• packages/webrtcaec3 - vendored WebRTC AEC3 wasm assets
• aec-wasm - placeholder Rust AEC module (not wired into main pipeline)

Notes

• System audio capture uses getDisplayMedia plus electron-audio-loopback.
• Recordings larger than 100 MB trigger an error to avoid slow transcriptions.

Docs

• docs/init.md - dev spec and JSON format
• docs/feature:denoising.md - echo reduction notes
• docs/features:webrtcaec3.md - AEC3 integration details
• docs/feature:coreaudio.md - CoreAudio loopback research

License

MIT