AI Live Voice Translator

Free AI Live Voice Translator

Speak into your device and see what you said with its translation beside it, as the conversation happens — for example Chinese speech with a Thai or English translation. Each phrase is transcribed as you say it and translated the moment you pause. Switch to full-screen captions to put the text up on a screen or a TV for a room, a class, a meeting or a video call.

Translate YouTube, Teams and Zoom

Choose Tab or screen sound to translate what’s playing on your computer instead of your own voice: a YouTube video, a webinar, a podcast, or a Microsoft Teams or Zoom call. Share the browser tab with its audio — or, for the Teams and Zoom desktop apps, the entire screen with system audio — and the captions follow along. Meeting mode mixes that sound with your microphone, so both sides of the call are translated (wear headphones). Open Floating captions for a small window that stays on top of the video or meeting (Chrome and Edge on a computer).

Auto-detect the spoken language

Choose Auto-detect and list the languages people may speak — say English and Chinese. Each phrase is recognised by OpenAI’s Whisper model running in your browser, which also decides which of those languages it was, so two people can take turns in different languages without touching anything. Each phrase appears as spoken, translated into the language you chose — and a phrase already in that language is translated into the other speaker’s language, so a two-way conversation is translated both ways.

Two ways to translate

• Gemini AI — the most natural translations, sent to our server (text only) and metered daily.
• Chrome built-in — Chrome’s on-device translator: near-instant and unlimited; each language pair downloads once. Desktop Chrome 138 and later.

What runs where

With browser recognition, speech-to-text is your browser’s Web Speech API (in Chrome that is Google’s speech service); the audio is never uploaded to us. With local Whisper — always used for sound shared from a tab or the screen — the audio never leaves your device at all. Only with the Gemini engine is the recognised text sent to our server for translation, and it is not stored. With local Whisper and Chrome’s built-in translator, the whole tool works without sending anything anywhere.

Speed and devices

• Local Whisper uses your graphics card through WebGPU when the browser supports it, and the CPU otherwise — a few seconds per phrase on an ordinary CPU, faster with a graphics card.
• The Whisper model downloads once and is cached by the browser; the next visit starts in seconds.
• Browser recognition shows words while you talk and works in Chrome, Edge and Safari; Firefox has none, so it uses Whisper.

Please do not rely on it unchecked

Two lossy steps run back to back: speech to text, then machine translation. Each is usually good and occasionally confidently wrong, and a misheard word becomes a fluent, plausible mistranslation. That is fine for following a conversation or taking notes — not for medical, legal, financial or safety-critical situations, where you should use a human interpreter.