live-translate
v0.2.0
Published
Real-time English ↔ Mandarin speech translator. Local-first, cloud-optional.
Readme
waxberry
Real-time English ↔ Mandarin speech translator. Local-first, cloud-optional. Your voice never leaves your machine.
Install
npm install -g waxberryQuick start
waxberry doctor # 1. check prerequisites
waxberry config # 2. choose a provider
waxberry start # 3. start the backend
waxberry # 4. start translating (hold SPACE)Provider comparison
| Provider | Quality | Speed | Cost | Privacy | Needs | |----------|---------|-------|------|---------|-------| | Opus-MT | ★★ | ★★★★★ | Free | Fully local | Nothing | | Ollama | ★★★★ | ★★★ | Free | Fully local | Ollama installed | | Claude Haiku | ★★★★ | ★★★★ | ~$0.001/use | Text via API | API key | | Claude Sonnet | ★★★★★ | ★★★ | ~$0.01/use | Text via API | API key | | GPT-4o-mini | ★★★★ | ★★★★ | ~$0.001/use | Text via API | API key | | DeepSeek | ★★★★ | ★★★ | ~$0.0005/use | Text via API | API key |
Privacy model: ASR (speech-to-text) and TTS (text-to-speech) always run locally. Only the translated text is sent to a cloud API when using a cloud provider. Your audio never leaves your machine.
Commands
waxberry doctor
Check Node.js, Docker, Sox, microphone, speaker, and configured provider.
waxberry config
Interactive provider setup. Saves to ~/.waxberry/config.json.
waxberry config --provider anthropic --model claude-haiku-4-5-20241022 --api-key sk-ant-...
Non-interactive setup for scripting.
waxberry start
Start the Docker backend. On first run, downloads ~850 MB of models.
waxberry
Push-to-talk translation mode. Press SPACE to start/stop recording. Q to quit.
waxberry status
Show health of all services and active provider.
waxberry stop
Stop the Docker backend.Privacy
Mic → ASR (local) → text → [Cloud API if configured] → text → TTS (local) → Speaker
↑ only this step is cloud- Opus-MT and Ollama: fully local, nothing leaves your machine.
- Anthropic, OpenAI, DeepSeek: audio stays local; only translated text is sent.
Prerequisites
- Node.js 18+
- Docker (with Docker Compose v2)
- Sox —
brew install soxorapt install sox
Troubleshooting
waxberry: command not found — make sure npm global bin is on your PATH: npm config get prefix/bin.
Sox not found — install with brew install sox (macOS) or apt install sox (Ubuntu).
No microphone detected — check system permissions for terminal microphone access.
Services timeout on start — first run downloads ~850 MB of models. Allow up to 10 minutes.
Ollama model not pulled — waxberry start will offer to pull it automatically, or run ollama pull qwen2.5:7b manually.
License
MIT
