@inproceedings{lee-etal-2025-hakka,
    title = "{H}akka Speech Recognition with Whisper and {P}inyin Post-processing for {FSR}-2025",
    author = "Lee, Chia-Hsin  and
      Chang, Yung-Jun  and
      Wu, Jin-Yan  and
      Chen, Kuan-Yu",
    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://preview.aclanthology.org/dashboard/2025.rocling-main.53/",
    pages = "454--458",
    ISBN = "979-8-89176-379-1",
    abstract = "本研究為參加 FSR-2025 客語語音辨識挑戰賽(Hakka ASR II)的技術報告,旨在推進客語自動語音辨識技術的發展。由於客語屬於低資源語言,且存在多種腔調,語音辨識面臨高度挑戰。我們以 Whisperlarge-v2 為骨幹模型,設計兩階段訓練流程:首先利用「Hakka Across Taiwan(HAT)」語料庫進行模型調適,以捕捉客語的一般聲學特徵;其次在賽事方提供的60 小時腔調語料上進行微調,以增強對目標資料的適應性。實驗發現,直接輸出客語漢字可達到良好的字錯率(CER),但由 於腔調差異與拼音規則變化多,拼音任務表現顯著下降。為解決此問題,我們以漢字模型的編碼器初始化拼音模型,並提出結合 RoBERTa 漢字轉拼音、腔調判斷與字典修正的後處理模組,期望可以在比賽中提升辨識的成效。"
}