Saved in:
Bibliographic Details
Main Authors: Huo, Mingyue, Zhang, Yuheng, Tang, Yan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.07195
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912577811906560
author Huo, Mingyue
Zhang, Yuheng
Tang, Yan
author_facet Huo, Mingyue
Zhang, Yuheng
Tang, Yan
contents Modern end-to-end automatic speech recognition (ASR) models like Whisper not only suffer from reduced recognition accuracy in noise, but also exhibit overconfidence - assigning high confidence to wrong predictions. We conduct a systematic analysis of Whisper's behavior in additive noise conditions and find that overconfident errors increase dramatically at low signal-to-noise ratios, with 10-20% of tokens incorrectly predicted with confidence above 0.7. To mitigate this, we propose a lightweight, post-hoc calibration framework that detects potential overconfidence and applies temperature scaling selectively to those tokens, without altering the underlying ASR model. Evaluations on the R-SPIN dataset demonstrate that, in the low signal-to-noise ratio range (-18 to -5 dB), our method reduces the expected calibration error (ECE) by 58% and triples the normalized cross entropy (NCE), yielding more reliable confidence estimates under severe noise conditions.
format Preprint
id arxiv_https___arxiv_org_abs_2509_07195
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Identifying and Calibrating Overconfidence in Noisy Speech Recognition
Huo, Mingyue
Zhang, Yuheng
Tang, Yan
Audio and Speech Processing
Modern end-to-end automatic speech recognition (ASR) models like Whisper not only suffer from reduced recognition accuracy in noise, but also exhibit overconfidence - assigning high confidence to wrong predictions. We conduct a systematic analysis of Whisper's behavior in additive noise conditions and find that overconfident errors increase dramatically at low signal-to-noise ratios, with 10-20% of tokens incorrectly predicted with confidence above 0.7. To mitigate this, we propose a lightweight, post-hoc calibration framework that detects potential overconfidence and applies temperature scaling selectively to those tokens, without altering the underlying ASR model. Evaluations on the R-SPIN dataset demonstrate that, in the low signal-to-noise ratio range (-18 to -5 dB), our method reduces the expected calibration error (ECE) by 58% and triples the normalized cross entropy (NCE), yielding more reliable confidence estimates under severe noise conditions.
title Identifying and Calibrating Overconfidence in Noisy Speech Recognition
topic Audio and Speech Processing
url https://arxiv.org/abs/2509.07195