I should also add that if you don’t want to reconfigure Jambi to use Whisper, you can try WhisperNow, which is built in Python and uses Whisper. I saw similar transcription performance when I used Whisper with Jambi and decided to move forward with Vosk after testing Whisper in both programs.

I use the Futo Keyboard on Android which already has this feature. It’s also opens source, and I’ve had a really good experience with it so far.

I’m not familiar with whisper.c++ but I did try faster-whisper. Unfortunately, the transcriptions took upward of 40sec and it didn’t offer live transcription, which is a nice feature of vosk. There’s a comparison in the readme with other differences. That said, it should be relatively modular. It shouldn’t take much to swap it back to whisper if that’s what you prefer to use. Whisper is in the nix flake as optional, and the program allows you to change models but i haven’t bothered trying to switch back to Whisper since Vosk has been more performant.

That means a lot to me. Thanks so much for your kind words!