Feature Request: Add Whisper/whisper.cpp support as alternative to Vosk

[nishantkumar1292]

The current Vosk models provide decent recognition but fall noticeably behind modern alternatives in accuracy, especially for accents and noisy environments.
Request: Add support for whisper.cpp as an alternative speech recognition backend.
Why Whisper?

Significantly better accuracy than Vosk (especially Whisper small/medium/large)
whisper.cpp is optimised for on-device inference and runs well on mobile
Already proven on Android via FUTO Voice Input
Multi-language support built-in

This could be implemented as:

A separate backend option alongside Vosk
Or a complete migration to whisper.cpp

Related: #97 (Moonshine request)
Would love to hear if this is feasible or if there are blockers. Happy to help test.

[m13v]

everyone recommends whisper as the obvious upgrade from vosk, and for accuracy it clearly is. but there's a real tradeoff people skip over: whisper wasn't designed for the "type as you speak" use case. it works on complete utterances, not streaming partial results. vosk gives you word-level partials in real time which is what makes a keyboard feel responsive. whisper.cpp now has a streaming mode but it buffers ~2-3 seconds before emitting, which feels laggy for typing. parakeet-tdt handles this better with its transducer architecture, worth looking at as a middle ground between vosk's speed and whisper's accuracy.