Real-time speech interpretation — speak, translate, hear, all in your browser.
Press mic → every 4 seconds, audio is sent to ASR → MT → (optional TTS playback). Results appear in a live scrolling bilingual feed with timestamps.
- 🎙️ Live microphone recording with 4-second chunk processing
- 🔊 Multi-provider ASR — OpenAI Whisper, DashScope Paraformer, Groq Whisper
- 🌐 Multi-provider MT — OpenAI-compatible APIs, DashScope, Anthropic Claude
- 🔈 Optional TTS auto-playback — hear the translation spoken aloud in real time
- 📜 Live bilingual feed — source + translation with timestamps, auto-scrolling
- 📥 Session transcript export as TXT
- 🌍 12 language pairs — ZH, EN, JA, KO, FR, DE, ES, PT, RU, AR, TH, VI
- 🖥️ Runs entirely in-browser — no app install, no server setup beyond
npm run dev
Mic → [4s chunk] → ASR (transcribe) → MT (translate) → Feed
↓
TTS playback (optional)
git clone https://github.com/Grayson0130/liveinterpret.git
cd liveinterpret
npm install
npm run dev
# → http://localhost:3003Note: Requires microphone permission. Works in Chrome, Firefox, Edge, Safari 14.1+.
| Provider | Model | Notes |
|---|---|---|
| OpenAI Whisper | whisper-1 |
Solid accuracy, many languages |
| DashScope | paraformer-realtime-v2 |
Excellent for Chinese |
| Groq Whisper | whisper-large-v3-turbo |
Fastest response |
| Provider | Example model |
|---|---|
| OpenAI / DashScope / Groq / DeepSeek | gpt-4o-mini, qwen-plus |
| Anthropic Claude | claude-3-5-haiku-20241022 (fast) |
| Provider | Voices |
|---|---|
| OpenAI TTS | alloy, echo, fable, onyx, nova, shimmer |
| DashScope CosyVoice | longxiaochun, longwan, longfei, loongstella |
- Multilingual meetings and remote conferences
- Live event interpretation without dedicated hardware
- Language learning through real-time bilingual transcription
- Accessibility tooling for deaf/hard-of-hearing users
Next.js 14 · TypeScript · Web MediaRecorder API · openai SDK · @anthropic-ai/sdk
MIT © Jinhui Zhao