Sokuji
A live speech translator runs offline on your device, with translated voice and subtitles for desktop calls and supported Chrome and Edge meeting sites.

Sokuji translates speech for people taking part in meetings across languages. Other participants hear your translated voice as ordinary call audio, so they don't need to install anything.
- Voice and captions: Send translated speech through a virtual microphone, read incoming speech as subtitles, or use captions for both sides without synthetic speech.
- Calls and other audio: The desktop app works with applications that accept a microphone and can subtitle system audio, including videos and recordings; the browser extension handles supported sites such as Zoom, Google Meet, Microsoft Teams and Discord.
- On-device models: Run speech recognition, translation and speech output locally with models including Whisper, Opus-MT, Qwen and Piper, using your CPU or integrated graphics.
- Meeting controls: Use movable subtitle overlays, export conversations, and adjust translated speech speed to keep pace; voice cloning depends on the selected provider.
The free, open source app runs on Windows, macOS and Linux, with Chrome and Edge extensions, under AGPL-3.0. After downloading local models, translation works offline without an account or API key, and audio processing stays on your device. Windows audio routing requires VB-CABLE; macOS includes an audio driver.
Optional cloud providers include OpenAI, Google Gemini and Soniox. Cloud audio goes directly to the chosen provider. The separate paid Kizuna AI account service handles credentials and billing rather than running the models. Soniox provides automatic language detection for two-way translation; OpenAI Realtime-compatible endpoints are supported in the desktop app.