@inproceedings{wang-etal-2025-dynamic,
title = "Dynamic Model-Bank Test-Time Adaptation for Automatic Speech Recognition",
author = "Wang, Yanshuo and
Zhou, Yanghao and
Lin, Yukang and
Chen, Haoxing and
Zhang, Jin and
Zhu, Wentao and
Hong, Jie and
Li, Xuesong",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.emnlp-main.1107/",
doi = "10.18653/v1/2025.emnlp-main.1107",
pages = "21831--21841",
ISBN = "979-8-89176-332-6",
abstract = "End-to-end automatic speech recognition (ASR) based on deep learning has achieved impressive progress in recent years. However, the performance of ASR foundation model often degrades significantly on out-of-domain data due to real-world domain shifts. Test-Time Adaptation (TTA) methods aim to mitigate this issue by adapting models during inference without access to source data. Despite recent progress, existing ASR TTA methods often struggle with instability under continual and long-term distribution shifts. To alleviate the risk of performance collapse due to error accumulation, we propose Dynamic Model-bank Single-Utterance Test-time Adaptation (DMSUTA), a sustainable continual TTA framework based on adaptive ASR model ensembling. DMSUTA maintains a dynamic model bank, from which a subset of checkpoints is selected for each test sample based on confidence and uncertainty criteria. To preserve both model plasticity and long-term stability, DMSUTA actively manages the bank by filtering out potentially collapsed models. This design allows DMSUTA to continually adapt to evolving domain shifts in ASR test-time scenarios. Experiments on diverse, continuously shifting ASR TTA benchmarks show that DMSUTA consistently outperforms existing continual TTA baselines, demonstrating superior robustness to domain shifts in ASR."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2025-dynamic">
<titleInfo>
<title>Dynamic Model-Bank Test-Time Adaptation for Automatic Speech Recognition</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yanshuo</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanghao</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yukang</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoxing</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jin</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wentao</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jie</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuesong</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-332-6</identifier>
</relatedItem>
<abstract>End-to-end automatic speech recognition (ASR) based on deep learning has achieved impressive progress in recent years. However, the performance of ASR foundation model often degrades significantly on out-of-domain data due to real-world domain shifts. Test-Time Adaptation (TTA) methods aim to mitigate this issue by adapting models during inference without access to source data. Despite recent progress, existing ASR TTA methods often struggle with instability under continual and long-term distribution shifts. To alleviate the risk of performance collapse due to error accumulation, we propose Dynamic Model-bank Single-Utterance Test-time Adaptation (DMSUTA), a sustainable continual TTA framework based on adaptive ASR model ensembling. DMSUTA maintains a dynamic model bank, from which a subset of checkpoints is selected for each test sample based on confidence and uncertainty criteria. To preserve both model plasticity and long-term stability, DMSUTA actively manages the bank by filtering out potentially collapsed models. This design allows DMSUTA to continually adapt to evolving domain shifts in ASR test-time scenarios. Experiments on diverse, continuously shifting ASR TTA benchmarks show that DMSUTA consistently outperforms existing continual TTA baselines, demonstrating superior robustness to domain shifts in ASR.</abstract>
<identifier type="citekey">wang-etal-2025-dynamic</identifier>
<identifier type="doi">10.18653/v1/2025.emnlp-main.1107</identifier>
<location>
<url>https://aclanthology.org/2025.emnlp-main.1107/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>21831</start>
<end>21841</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Dynamic Model-Bank Test-Time Adaptation for Automatic Speech Recognition
%A Wang, Yanshuo
%A Zhou, Yanghao
%A Lin, Yukang
%A Chen, Haoxing
%A Zhang, Jin
%A Zhu, Wentao
%A Hong, Jie
%A Li, Xuesong
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-332-6
%F wang-etal-2025-dynamic
%X End-to-end automatic speech recognition (ASR) based on deep learning has achieved impressive progress in recent years. However, the performance of ASR foundation model often degrades significantly on out-of-domain data due to real-world domain shifts. Test-Time Adaptation (TTA) methods aim to mitigate this issue by adapting models during inference without access to source data. Despite recent progress, existing ASR TTA methods often struggle with instability under continual and long-term distribution shifts. To alleviate the risk of performance collapse due to error accumulation, we propose Dynamic Model-bank Single-Utterance Test-time Adaptation (DMSUTA), a sustainable continual TTA framework based on adaptive ASR model ensembling. DMSUTA maintains a dynamic model bank, from which a subset of checkpoints is selected for each test sample based on confidence and uncertainty criteria. To preserve both model plasticity and long-term stability, DMSUTA actively manages the bank by filtering out potentially collapsed models. This design allows DMSUTA to continually adapt to evolving domain shifts in ASR test-time scenarios. Experiments on diverse, continuously shifting ASR TTA benchmarks show that DMSUTA consistently outperforms existing continual TTA baselines, demonstrating superior robustness to domain shifts in ASR.
%R 10.18653/v1/2025.emnlp-main.1107
%U https://aclanthology.org/2025.emnlp-main.1107/
%U https://doi.org/10.18653/v1/2025.emnlp-main.1107
%P 21831-21841
Markdown (Informal)
[Dynamic Model-Bank Test-Time Adaptation for Automatic Speech Recognition](https://aclanthology.org/2025.emnlp-main.1107/) (Wang et al., EMNLP 2025)
ACL
- Yanshuo Wang, Yanghao Zhou, Yukang Lin, Haoxing Chen, Jin Zhang, Wentao Zhu, Jie Hong, and Xuesong Li. 2025. Dynamic Model-Bank Test-Time Adaptation for Automatic Speech Recognition. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 21831–21841, Suzhou, China. Association for Computational Linguistics.