Live Translation
Real-time speech transcription, translation and voice, fully offline
- Next.js
- TypeScript
- Tailwind CSS
- shadcn/ui
- Socket.IO
- Python
- Faster Whisper
- Argos Translate
- Kokoro TTS
A web app that listens to your microphone, writes what is said as live captions, translates them into another language and can read the translation aloud. Audio and video files can also be uploaded for timestamped transcripts with SRT subtitle export. Everything runs locally with open-source AI models, so no audio ever leaves the machine and no paid API is needed.
Functionality
- Live captions from the microphone, with a translation under each line
- 17 languages, with automatic detection of the spoken language
- AI voice that reads translations aloud, with several voices and speeds
- File upload for audio and video, with SRT and TXT export
- Caption history, microphone level meter and processing stats
How it works
- The browser streams audio chunks to a Python server over Socket.IO
- Faster Whisper transcribes on CPU, with filters against hallucinated phrases
- Argos Translate translates offline; Kokoro (ONNX) generates the voice
- A 24-second clip is fully processed in about 38 seconds on a 4-core CPU