So it's mostly intended to be a better replacement for whisper? Mostly? With better support for more models and maybe acceleration backends?

More or less yes, for whisper.cpp, just trying to make local transcription more accessible to anyone building an app, etc