Revealing the Role of Audio Channels in ASR Performance Degradation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915456467599360 |
|---|---|
| author | Huang, Kuan-Tang Chen, Li-Wei Lee, Hung-Shin Chen, Berlin Wang, Hsin-Min |
| author_facet | Huang, Kuan-Tang Chen, Li-Wei Lee, Hung-Shin Chen, Berlin Wang, Hsin-Min |
| contents | Pre-trained automatic speech recognition (ASR) models have demonstrated strong performance on a variety of tasks. However, their performance can degrade substantially when the input audio comes from different recording channels. While previous studies have demonstrated this phenomenon, it is often attributed to the mismatch between training and testing corpora. This study argues that variations in speech characteristics caused by different recording channels can fundamentally harm ASR performance. To address this limitation, we propose a normalization technique designed to mitigate the impact of channel variation by aligning internal feature representations in the ASR model with those derived from a clean reference channel. This approach significantly improves ASR performance on previously unseen channels and languages, highlighting its ability to generalize across channel and language differences. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_08967 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Revealing the Role of Audio Channels in ASR Performance Degradation Huang, Kuan-Tang Chen, Li-Wei Lee, Hung-Shin Chen, Berlin Wang, Hsin-Min Sound Artificial Intelligence Computation and Language Pre-trained automatic speech recognition (ASR) models have demonstrated strong performance on a variety of tasks. However, their performance can degrade substantially when the input audio comes from different recording channels. While previous studies have demonstrated this phenomenon, it is often attributed to the mismatch between training and testing corpora. This study argues that variations in speech characteristics caused by different recording channels can fundamentally harm ASR performance. To address this limitation, we propose a normalization technique designed to mitigate the impact of channel variation by aligning internal feature representations in the ASR model with those derived from a clean reference channel. This approach significantly improves ASR performance on previously unseen channels and languages, highlighting its ability to generalize across channel and language differences. |
| title | Revealing the Role of Audio Channels in ASR Performance Degradation |
| topic | Sound Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2508.08967 |