Voice Together

Voice Together is a local language-learning player.

Current flow:

Upload video/audio
  -> Go backend saves it
  -> ffmpeg extracts browser audio and transcription WAV
  -> Whisper-compatible tool generates timed transcript segments
  -> React player syncs audio with subtitles
  -> learner loops any dialogue segment

Requirements

• Go
• Node.js + npm
• Python 3

The project installs ffmpeg and faster-whisper into a local .venv with scripts/setup-tools.sh, so it does not require system-wide ffmpeg.

Run

cd ~/myproject/voice-together
scripts/setup-tools.sh
./start.sh

Then open:

http://192.168.1.30:8788

For the existing sample file, click 导入 demo.mp4.

Useful commands:

./start.sh status
./start.sh logs
./start.sh restart
./start.sh stop

Development Mode

Start the Go backend with Air reload and the React frontend with Vite hot reload:

./start.sh dev

Open the development frontend at:

http://192.168.1.30:5173

Development lifecycle commands:

./start.sh dev-status
./start.sh dev-logs
./start.sh dev-restart
./start.sh dev-stop

Air uses backend/.air.toml. The backend listens on port 8788, and Vite proxies /api requests to it.

The React frontend is embedded into the Go binary during ./start.sh build, so the built server does not need frontend/dist at runtime:

backend/bin/voice-together -root "$PWD"

Development Notes

• Backend: backend/cmd/server
• Frontend: frontend/src
• Embedded frontend source: backend/internal/web
• Runtime data: backend/data
• Extracted browser audio: backend/data/clips/<clip-id>/audio.mp3
• Transcription WAV: backend/data/clips/<clip-id>/transcribe.wav

If faster-whisper is not installed, the app still starts but returns an explicit placeholder subtitle telling you to run scripts/setup-tools.sh.