Whisper ASR hallucination detection studied via model internals

Arxiv CS.SD (Sound/Audio ML)

Two arXiv papers investigate detecting and mitigating Whisper hallucinations. One uses hidden representation steering and sparse autoencoders; the other studies Whisper large v3's internal metrics for detection.

Voice AISpeech-to-text

See 1 more source

Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncoders15 days agoGeorgii Aparin, Vadim Popov, Tasnima Sadekova, Assel Yermekova

Arxiv CS.SD (Sound/Audio ML)·Jan Jasi\'nskiMateusz Bara\'nskiJulitta BartolewskaMarcin WitkowskiKonrad Kowalczyk

23 hours ago