@inproceedings{hsu-etal-2025-improving,
title = "Improving Low-Resource Speech Recognition with Whisper-{M}o{E} and Synthetic Data Augmentation: A Case Study on {H}akka",
author = "Hsu, Yuan-Chi and
Fang, Liang-Chun and
Dai, Hong-Jie",
editor = "Chang, Kai-Wei and
Lu, Ke-Han and
Yang, Chih-Kai and
Tam, Zhi-Rui and
Chang, Wen-Yu and
Wang, Chung-Che",
booktitle = "Proceedings of the 37th Conference on Computational Linguistics and Speech Processing (ROCLING 2025)",
month = nov,
year = "2025",
address = "National Taiwan University, Taipei City, Taiwan",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.rocling-main.51/",
pages = "446--449",
ISBN = "979-8-89176-379-1",
abstract = "The objective of this study is to improve speech recognition performance for low-resource Hakka, a language spoken by a specific ethnic group. Our team conducted experiments by fine-tuning different base versions of Whisper (e.g., the original model and the Mandarin-focused Belle model). We found that fine-tuning on different bases yielded distinct advantages and varying results in Hakka character and phonetic recognition tasks. To further enhance model accuracy, we experimented with replacing the q, k, and v linear layers in the attention blocks of the Whisper encoder with a mixture-of-experts model combined with RoLA. In addition, we augmented the training data with synthesized speech generated with diverse voice styles and varying speaking rates. The results showed a 0.73{\%} reduction in character error rate for Task 1 and a 0.2{\%} reduction in word error rate for Task 2. These findings confirm that both architectural adjustments to the model and the strategic use of limited synthetic speech data in low-resource dialect corpora can effectively improve recognition performance."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hsu-etal-2025-improving">
<titleInfo>
<title>Improving Low-Resource Speech Recognition with Whisper-MoE and Synthetic Data Augmentation: A Case Study on Hakka</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yuan-Chi</namePart>
<namePart type="family">Hsu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liang-Chun</namePart>
<namePart type="family">Fang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hong-Jie</namePart>
<namePart type="family">Dai</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 37th Conference on Computational Linguistics and Speech Processing (ROCLING 2025)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kai-Wei</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ke-Han</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chih-Kai</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhi-Rui</namePart>
<namePart type="family">Tam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wen-Yu</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chung-Che</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">National Taiwan University, Taipei City, Taiwan</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-379-1</identifier>
</relatedItem>
<abstract>The objective of this study is to improve speech recognition performance for low-resource Hakka, a language spoken by a specific ethnic group. Our team conducted experiments by fine-tuning different base versions of Whisper (e.g., the original model and the Mandarin-focused Belle model). We found that fine-tuning on different bases yielded distinct advantages and varying results in Hakka character and phonetic recognition tasks. To further enhance model accuracy, we experimented with replacing the q, k, and v linear layers in the attention blocks of the Whisper encoder with a mixture-of-experts model combined with RoLA. In addition, we augmented the training data with synthesized speech generated with diverse voice styles and varying speaking rates. The results showed a 0.73% reduction in character error rate for Task 1 and a 0.2% reduction in word error rate for Task 2. These findings confirm that both architectural adjustments to the model and the strategic use of limited synthetic speech data in low-resource dialect corpora can effectively improve recognition performance.</abstract>
<identifier type="citekey">hsu-etal-2025-improving</identifier>
<location>
<url>https://aclanthology.org/2025.rocling-main.51/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>446</start>
<end>449</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Improving Low-Resource Speech Recognition with Whisper-MoE and Synthetic Data Augmentation: A Case Study on Hakka
%A Hsu, Yuan-Chi
%A Fang, Liang-Chun
%A Dai, Hong-Jie
%Y Chang, Kai-Wei
%Y Lu, Ke-Han
%Y Yang, Chih-Kai
%Y Tam, Zhi-Rui
%Y Chang, Wen-Yu
%Y Wang, Chung-Che
%S Proceedings of the 37th Conference on Computational Linguistics and Speech Processing (ROCLING 2025)
%D 2025
%8 November
%I Association for Computational Linguistics
%C National Taiwan University, Taipei City, Taiwan
%@ 979-8-89176-379-1
%F hsu-etal-2025-improving
%X The objective of this study is to improve speech recognition performance for low-resource Hakka, a language spoken by a specific ethnic group. Our team conducted experiments by fine-tuning different base versions of Whisper (e.g., the original model and the Mandarin-focused Belle model). We found that fine-tuning on different bases yielded distinct advantages and varying results in Hakka character and phonetic recognition tasks. To further enhance model accuracy, we experimented with replacing the q, k, and v linear layers in the attention blocks of the Whisper encoder with a mixture-of-experts model combined with RoLA. In addition, we augmented the training data with synthesized speech generated with diverse voice styles and varying speaking rates. The results showed a 0.73% reduction in character error rate for Task 1 and a 0.2% reduction in word error rate for Task 2. These findings confirm that both architectural adjustments to the model and the strategic use of limited synthetic speech data in low-resource dialect corpora can effectively improve recognition performance.
%U https://aclanthology.org/2025.rocling-main.51/
%P 446-449
Markdown (Informal)
[Improving Low-Resource Speech Recognition with Whisper-MoE and Synthetic Data Augmentation: A Case Study on Hakka](https://aclanthology.org/2025.rocling-main.51/) (Hsu et al., ROCLING 2025)
ACL