Speech recognition

Speech-to-text that does not need a data centre.

On-device STT means the weights sit on the SSD and inference uses Neural Engine / GPU on the Mac. VV’s English default is Parakeet. You can try other local models from Experimental settings; you do not have to.

This is slower to download the first time and faster every time after, because there is no upload. Offline aeroplane dictation works.

Deep dive: on-device feature. Competitor context: Wispr Flow.