Skip to content
conftrace_
2025 Interspeech INTERSPEECH 2025

Calm-Whisper: Reduce Whisper Hallucination On Non-Speech By Calming Crazy Heads Down

Abstract

OpenAI's Whisper has achieved significant success in Automatic Speech Recognition. However, it has consistently been found to exhibit hallucination issues, particularly in non-speech segments, which limits its broader application in complex industrial settings.

🌉 Interdisciplinary Bridge - Artificial Intelligence and Speech & Audio

Authors

Yingzhi Wang, Anas Alhmoud, Saad Alsahly, Muhammad Alqurishi, Mirco Ravanelli

Topics

Speech & Audio > Recognition > Automatic Speech Recognition Speech & Audio > Recognition > Speech Recognition Deep Learning > Learning Types > Robustness

Keywords

automatic speech recognition hallucination reduction speech foundation model non-speech segment
Download PDF

More from Interspeech 2025

All Interspeech 2025 papers →
StarGAN-Aug: A Cross-domain Fault Audio Generation Method for High-performance Fault Diagnosis of Power Transformers
SuPseudo: A Pseudo-supervised Learning Method for Neural Speech Enhancement in Far-field Speech Recognition
Lightweight Front-end Enhancement for Robust ASR via Frame Resampling and Sub-Band Pruning
HuBERT-VIC: Improving Noise-Robust Automatic Speech Recognition of Speech Foundation Model via Variance-Invariance-Covariance Regularization
MOVER: Combining Multiple Meeting Recognition Systems
EmbedAug: An Augmentation Scheme for End-to-End Automatic Speech Recognition