Live Caption and Translation for Firefox

A Firefox extension paired with a Windows backend that generates real-time English captions for whatever video is playing in the browser, translating on the fly when the audio is in another language.

How it works

Turn it on per tab. The extension finds the video that is actually playing audio, streams it to the local backend, and overlays the captions on the video. The caption UI is customisable and re-fits itself to the video layout as the Firefox window is resized. Spoken language is detected in real time, so a video that switches languages keeps producing English.

The backend is a tray application. Its AI models only spin up while the extension is enabled for some tab and audio is actually streaming, so an idle browser costs nothing beyond the loaded models.

Requirements

A CUDA GPU is needed, CUDA 13 capable recommended. The extension cannot run GPU-accelerated models on its own, so the backend is mandatory, and that backend is Windows only. Install it from the release files (the setup executable plus its .bin parts in one folder), then add the extension from addons.mozilla.org and restart the browser.

Coverage is strong for English, Spanish, French, Italian, Portuguese, Dutch, German, Turkish, Russian, Arabic, Hindi, Japanese, Korean, Vietnamese and Ukrainian, with a second tier of European languages and Mandarin that works but less reliably.

Footprint

On the development machine (RTX 4090, Threadripper 7960X) the backend holds roughly 5 GB of VRAM with the models loaded and under 6 GB while transcribing and translating, at well under a third of one CPU core. On a similar machine it is unnoticeable in use; your mileage may vary on smaller GPUs.