Re:An hour of audio transcribed in about 13s on MacBook Air
by mikemzz1 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
[ Parent
| on HN ]
Re:An hour of audio transcribed in about 13s on MacBook Air
by Shershah1024 on Tuesday October 13, 2026 @09:38AM
(#50071579)
That's close to what this is: the reader is a CTC head on the audio encoder, but its output words are EmbeddingGemma's own text embeddings for ~229k words, instead of a separately learned vocabulary. I also tried a plain letter-level CTC head on the encoder. It did a bit better on clean audiobooks (7.0 vs 8.2 WER at the time) but much worse on harder audio like a lecture (16.5 vs 10.8), mostly from misspellings.