mikemzz1
Re:An hour of audio transcribed in about 13s on MacBook Air
on Tuesday October 13, 2026 @07:39AM (#50070073)
interesting approach. But training ctc head on top of gemma's audio encoder will most likely produce better results and faster inference