I get the impression that recreating the Whisper training code is possibly doable, but the data is a bigger task.

This is a possible Whisper alternative with maybe similar issues: https://petewarden.com/2024/10/21/introducing-moonshine-the-new-state-of-the-art-for-speech-to-text/

Yes it's nice that the phone app is free but STT is the difficult and important part. With Moonshine it might be possible to run the transcriber completely on the phone instead of having the STT on a remote server.

It's interesting that they are able to do all that speaker distinguishing with just a single mic as found on both phones. There was a thread about phone features recently. Given this STT stuff, it could be useful to have a phone with 3 or 4 mics in the corners of the phone, like one of those tabletop conference mics, so it can figure out directionality of sound sources.