Szczegóły publikacji
Opis bibliograficzny
Investigation of Whisper ASR hallucinations induced by non-speech audio / Mateusz BARAŃSKI, Jan JASIŃSKI, Julitta BARTOLEWSKA, Stanisław KACPRZAK, Marcin WITKOWSKI, Konrad KOWALCZYK // W: ICASSP 2025 [Dokument elektroniczny] : 2025 IEEE International Conference on Acoustics, Speech and Signal Processing : [6-11 April, 2025], Hyderabad, India : conference proceedings. — Wersja do Windows. — Dane tekstowe. — Piscataway : IEEE, cop. 2025. — ( Proceedings of the ... IEEE International Conference on Acoustics, Speech, and Signal Processing ; ISSN 1520-6149 ). — Dod. ISBN: 979-8-3503-6875-8. — e-ISBN: 979-8-3503-6874-1. — S. [1-5]. — Wymagania systemowe: Adobe Reader. — Bibliogr. s. [5], Abstr. — Publikacja dostępna online od: 2025-03-07
Autorzy (6)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 158903 |
|---|---|
| Data dodania do BaDAP | 2025-04-15 |
| Tekst źródłowy | URL |
| DOI | 10.1109/ICASSP49660.2025.10890105 |
| Rok publikacji | 2025 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Creative Commons | |
| Wydawca | Institute of Electrical and Electronics Engineers (IEEE) |
| Konferencja | IEEE International Conference on Acoustics, Speech and Signal Processing 2025 |
| Czasopismo/seria | Proceedings of the ... IEEE International Conference on Acoustics, Speech, and Signal Processing |
Abstract
Hallucinations of deep neural models are amongst key challenges in automatic speech recognition (ASR). In this paper, we investigate hallucinations of the Whisper ASR model induced by non-speech audio segments present during inference. By inducting hallucinations with various types of sounds, we show that there exists a set of hallucinations that appear frequently. We then study hallucinations caused by the augmentation of speech with such sounds. Finally, we describe the creation of a bag of hallucinations (BoH) that allows to remove the effect of hallucinations through the post-processing of text transcriptions. The results of our experiments show that such post-processing is capable of reducing word error rate (WER) and acts as a good safeguard against problematic hallucinations.
