← all repositories
kizuna-ai-lab/sokuji

Live voice translation that keeps your audio on-device

Built to make real-time multilingual conversation possible without forcing your audio into the cloud.

sokuji
Not currently ranked — collecting fresh signals.
star history

What it does

Sokuji is a desktop and browser-extension app that listens to your microphone or system audio, translates speech in real time, and pipes the result back out as synthesized voice or subtitles. It targets video calls and meetings, offering both a cloud path through providers like OpenAI and Google Gemini and a fully offline path that runs ASR, translation, and TTS inside your browser using WASM and WebGPU. The desktop build can create a virtual microphone, letting you route translated audio directly into Zoom, Teams, or Discord as if you were speaking the target language natively.

The interesting bit

The real trick is the fully offline pipeline. Sokuji squeezes ASR, translation, and TTS into the browser using sherpa-onnx and Transformers.js via WASM and WebGPU, caching models in IndexedDB so it can run on integrated graphics without an API key. It is essentially trying to replace a cloud interpreter with a browser tab and some local storage.

Key highlights

  • Dual-mode architecture: Choose between cloud providers (OpenAI, Gemini, Palabra.ai, Doubao, etc.) or local inference that requires no network and no API key.
  • Virtual microphone output: On desktop, translated speech can be fed as a native-sounding input into Zoom, Teams, Discord, or OBS.
  • Broad model catalog offline: Claims 50 ASR models, 55+ translation pairs, and 136 TTS voices available for local use after download.
  • Cross-platform surface area: Ships as an Electron app for Linux, Windows, and macOS, plus a Chrome/Edge extension for web-based meetings.
  • Noise suppression and echo cancellation: Uses Web Audio API and AI noise suppression to clean input before translation.

Caveats

  • The OpenAI-compatible provider mode is available only in the desktop Electron build, not the browser extension.
  • Browser support for the extension currently excludes Brave, which is listed as “coming soon”.
  • Anonymous usage analytics are sent via PostHog, even though the privacy pitch emphasizes keeping audio local.

Verdict

Grab it if you need a real-time meeting interpreter that can run entirely offline in a browser tab. Pass if you just need to translate text documents or require a solution with full enterprise browser support today.

Frequently asked

What is kizuna-ai-lab/sokuji?
Built to make real-time multilingual conversation possible without forcing your audio into the cloud.
Is sokuji open source?
Yes — kizuna-ai-lab/sokuji is open source, released under the AGPL-3.0 license.
What language is sokuji written in?
kizuna-ai-lab/sokuji is primarily written in TypeScript.
How popular is sokuji?
kizuna-ai-lab/sokuji has 1k stars on GitHub.
Where can I find sokuji?
kizuna-ai-lab/sokuji is on GitHub at https://github.com/kizuna-ai-lab/sokuji.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.