Generate captions locally with sherpa-onnx. Choose the streaming tab and enable Share tab audio.
Only the selected model downloads, after you start and authorize audio. Downloads are cached on this device when storage is available. Audio and transcripts are not uploaded. Keep this window open while listening. This language choice is separate from the video’s native subtitle language.
Generated captions appear over the video on the original streaming page. A copy appears below in this window. If neither appears, check the status below for an audio or recognition error.
Connecting to Ctrl-h…
Firefox receives audio exposed by the playing video over a local browser-to-browser connection with no relay servers. Only connection setup messages use userscript-manager storage, not audio. Protected media is not captured; there is no microphone fallback.
When the browser isolates the windows, recent caption text is passed through local userscript-manager storage and cleared on normal stop. No audio is stored there.
Tab audio is not supported by every browser. Protected playback may provide no audio. There is no microphone fallback. This feature cannot unlock protected media.
Recognition uses your CPU in a background worker. It can increase battery use and captions may be delayed or inaccurate. If recognition falls behind, it stops rather than building an unlimited backlog.
This clears the speech runtime/model cache only, not your replacement list or other app data. Browser HTTP cache may retain downloaded assets.