Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition

Yu-Sheng Huang; Wei-Cheng Hong; Xin-Yu Chen; Szu-Yin Lin

Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition

Yu-Sheng Huang, Wei-Cheng Hong, Xin-Yu Chen, Szu-Yin Lin

Correct Metadata for

Important: The Anthology treat PDFs as authoritative. Please use this form only to correct data that is out of line with the PDF. See our corrections guidelines if you need to change the PDF.

Title Adjust the title. Retain tags such as <fixed-case>.

Authors Adjust author names and order to match the PDF.

Abstract Correct abstract if needed. Retain XML formatting tags such as <tex-math>.

Verification against PDF Ensure that the new title/authors match the snapshot below. (If there is no snapshot or it is too small, consult the PDF.)

Authors concatenated from the text boxes above:

ALL author names match the snapshot above—including middle initials, hyphens, and accents.

Abstract

This study addresses the FSR 2025 Hakka speech recognition task by comparing two strategies: fine-tuning large pre-trained models and training from scratch. For character (Hanzi) recognition, we fine-tuned five different scales of the Whisper model, with large-v3-turbo achieving a 7.55% CER on the test set. For Pinyin recognition, a Branchformer model was compared against a LoRA fine-tuned Whisper-small, yielding WERs of 4.7% and 6.5% on the test set, respectively. Speed perturbation was the primary method used for data augmentation in our pre-processing pipeline.

Anthology ID:: 2025.rocling-main.50
Volume:: Proceedings of the 37th Conference on Computational Linguistics and Speech Processing (ROCLING 2025)
Month:: November
Year:: 2025
Address:: National Taiwan University, Taipei City, Taiwan
Editors:: Kai-Wei Chang, Ke-Han Lu, Chih-Kai Yang, Zhi-Rui Tam, Wen-Yu Chang, Chung-Che Wang
Venue:: ROCLING
SIG:
Publisher:: Association for Computational Linguistics
Note:
Pages:: 441–445
Language:
URL:: https://aclanthology.org/2025.rocling-main.50/
DOI:
Bibkey:
Cite (ACL):: Yu-Sheng Huang, Wei-Cheng Hong, Xin-Yu Chen, and Szu-Yin Lin. 2025. Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition. In Proceedings of the 37th Conference on Computational Linguistics and Speech Processing (ROCLING 2025), pages 441–445, National Taiwan University, Taipei City, Taiwan. Association for Computational Linguistics.
Cite (Informal):: Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition (Huang et al., ROCLING 2025)
Copy Citation:
PDF:: https://aclanthology.org/2025.rocling-main.50.pdf

PDF Cite Search Fix data

Export citation

BibTeX
MODS XML
Endnote
Preformatted

@inproceedings{huang-etal-2025-applying,
    title = "Applying Whisper Fine-tuning and Branchformer to {H}akka Speech Recognition",
    author = "Huang, Yu-Sheng  and
      Hong, Wei-Cheng  and
      Chen, Xin-Yu  and
      Lin, Szu-Yin",
    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.50/",
    pages = "441--445",
    ISBN = "979-8-89176-379-1",
    abstract = "This study addresses the FSR 2025 Hakka speech recognition task by comparing two strategies: fine-tuning large pre-trained models and training from scratch. For character (Hanzi) recognition, we fine-tuned five different scales of the Whisper model, with large-v3-turbo achieving a 7.55{\%} CER on the test set. For Pinyin recognition, a Branchformer model was compared against a LoRA fine-tuned Whisper-small, yielding WERs of 4.7{\%} and 6.5{\%} on the test set, respectively. Speed perturbation was the primary method used for data augmentation in our pre-processing pipeline."
}

Download as File

<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2025-applying">
    <titleInfo>
        <title>Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Yu-Sheng</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Wei-Cheng</namePart>
        <namePart type="family">Hong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xin-Yu</namePart>
        <namePart type="family">Chen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Szu-Yin</namePart>
        <namePart type="family">Lin</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>This study addresses the FSR 2025 Hakka speech recognition task by comparing two strategies: fine-tuning large pre-trained models and training from scratch. For character (Hanzi) recognition, we fine-tuned five different scales of the Whisper model, with large-v3-turbo achieving a 7.55% CER on the test set. For Pinyin recognition, a Branchformer model was compared against a LoRA fine-tuned Whisper-small, yielding WERs of 4.7% and 6.5% on the test set, respectively. Speed perturbation was the primary method used for data augmentation in our pre-processing pipeline.</abstract>
    <identifier type="citekey">huang-etal-2025-applying</identifier>
    <location>
        <url>https://aclanthology.org/2025.rocling-main.50/</url>
    </location>
    <part>
        <date>2025-11</date>
        <extent unit="page">
            <start>441</start>
            <end>445</end>
        </extent>
    </part>
</mods>
</modsCollection>

Download as File

%0 Conference Proceedings
%T Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition
%A Huang, Yu-Sheng
%A Hong, Wei-Cheng
%A Chen, Xin-Yu
%A Lin, Szu-Yin
%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 huang-etal-2025-applying
%X This study addresses the FSR 2025 Hakka speech recognition task by comparing two strategies: fine-tuning large pre-trained models and training from scratch. For character (Hanzi) recognition, we fine-tuned five different scales of the Whisper model, with large-v3-turbo achieving a 7.55% CER on the test set. For Pinyin recognition, a Branchformer model was compared against a LoRA fine-tuned Whisper-small, yielding WERs of 4.7% and 6.5% on the test set, respectively. Speed perturbation was the primary method used for data augmentation in our pre-processing pipeline.
%U https://aclanthology.org/2025.rocling-main.50/
%P 441-445

Download as File

Markdown (Informal)

[Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition](https://aclanthology.org/2025.rocling-main.50/) (Huang et al., ROCLING 2025)

Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition (Huang et al., ROCLING 2025)

ACL

Yu-Sheng Huang, Wei-Cheng Hong, Xin-Yu Chen, and Szu-Yin Lin. 2025. Applying Whisper Fine-tuning and Branchformer to Hakka Speech Recognition. In Proceedings of the 37th Conference on Computational Linguistics and Speech Processing (ROCLING 2025), pages 441–445, National Taiwan University, Taipei City, Taiwan. Association for Computational Linguistics.