alt Hacker NewsLast time I tried whisper, it hallucinated an elaborate conversation from sounds of slapping and moaning and it took minutes to spit every single line of it.
Replies
dotancohen • today at 8:43 PM
If I remember correctly, the whisper documentation actually recommends to trim non-speech portions as the models halucinate heavily during those portions.
Parakeet has been trained to detect non-voice sounds and exclude that from identification, so you might have better luck with that family.