Why not just run a simple Whisper model on every audio batch? Whisper is designed for complete utterances, not real-time chunks. Processing small segments loses context, cuts off words mid-syllable, ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results