Open-source · v3 pipeline · Runs locally

Speak Hindi.
The meeting hears English.

Sync Speak is a desktop app that listens to your Hindi voice, translates it in real time, and plays natural English into Google Meet, Zoom, or Teams — under a second of latency, with context awareness across your last five sentences.

Download for your platform See how it works →
  • < 1.2s perceived latency
  • 5-utterance rolling context
  • 100% local voice capture
Inside the app

Four screens. One fluid experience.

Live translation, local history, guided setup, and your keys — every pane a frosted lens over your desktop.

Sync Speak — Translate
Sync Speak Translate screen
Sync Speak History screen
Sync Speak Guide screen
Sync Speak Settings screen

Live transcription and translation with voice sensitivity control. The window is a frosted lens — your desktop shows through.

Why Sync Speak

Built for how meetings actually happen.

Neural VAD, not a fixed mic gate

WebRTC Neural VAD with a 500 ms pre-buffer catches every “kya” and “toh” instead of clipping the first syllable.

Context-aware translation

Groq Llama 3.3 70B sees your last five utterances so pronouns, names, and jargon carry from sentence to sentence.

Sentence-pipelined TTS

Sarvam Bulbul v3 starts speaking the first clause while the next is still being synthesized — you never wait for a full paragraph to finish.

Routes to any meeting app

VB-Cable virtual mic feeds translated audio into Google Meet, Zoom, Teams, Discord — anything that accepts a microphone input.

True desktop transparency

Tauri v2 shell with Liquid Glass chrome. A floating lens on your desktop — never a heavy Electron window.

Open-source, your keys

MIT-licensed. Sarvam and Groq keys live in your local config file, never uploaded, never shared.

The pipeline

Six stages. One fluid voice.

  1. 01

    Microphone

    Your real mic, captured at 16 kHz with WASAPI reinit for stability.

  2. 02

    Neural VAD

    WebRTC VAD detects speech vs. silence. 500 ms pre-buffer prevents clipping.

  3. 03

    Sarvam Saarika v2.5

    Hindi / Hinglish speech → text, with a correction table for common mis-mappings.

  4. 04

    Groq Llama 3.3 70B

    Translation with 5-utterance rolling context for coherent meeting speech.

  5. 05

    Sarvam Bulbul v3

    Text → speech, streamed sentence by sentence so playback starts instantly.

  6. 06

    VB-Cable out

    Virtual mic feeds Google Meet, Zoom, Teams, Discord — anything.

Read the full pipeline →

Try it in your next meeting.

Free. Open-source. Your voice stays on your machine.

Download for your platform Star on GitHub