Sokuji, an open-source real-time voice translation app built by Kizuna AI Lab, is available with support for both cloud APIs and fully local inference. Aimed at breaking down language barriers in video meetings such as Zoom, Microsoft Teams and Google Meet, its source code is published on GitHub .
Open Source · Kizuna AI Lab
Sokuji: Real-Time Voice Translation That Runs in the Cloud — or Entirely on Your Device
An open-source app that converts a speaker's voice into another language in near real time, plugging straight into Google Meet, Microsoft Teams and Zoom — with a fully offline local mode that keeps every byte on your machine.
2
Processing modes — cloud APIs & on-device local inference
3
Conferencing tools: Meet, Teams & Zoom
0
API keys or internet needed in local mode
Cloud mode vs. Local mode — pick where translation happens
Cloud Mode
Runs on external providers
OpenAI, Gemini, Palabra.ai, Volcengine, Kizuna AI
User-owned API key required
Internet required
Benefit: accuracy & naturalness
Local Mode
Runs on device, in the browser
WASM + WebGPU on CPU / integrated GPU
No API key required
No internet required
Benefit: privacy & offline use
The on-device pipeline
In local mode the full chain completes in the browser — no data ever leaves the device.
→
Translation
Language conversion
→
What works well
Push-to-Talk preserves context and proves practical
Local model eases privacy concerns
Open source + offline broadens cost/privacy options
Current limits
Weaker speech generation for incoming English
Latency appears on long utterances
Local mode trails cloud on accuracy & naturalness
Business use may need notice of external processing
Keys to wider adoption: improving local inference accuracy and expanding the range of supported conferencing tools — with a subtitle mode and UI/UX improvements already underway.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…