MDCCST Dataset

This repository contains labeled Mandarin translation to the MDCC Cantonese ASR dataset. [https://github.com/HLTCHKUST/cantonese-asr]

For the original Cantonese transcription and audios, please find in the above MDCC repository.

@inproceedings{zhuang2025towards,
  title={Towards Efficiently Whisper Fine-tuning with Monotonic Alignments},
  author={Zhuang, Ziyang and Wei, Tao and Fang, Ming and Cheng, Ning and Wang, Shaojun and Xiao, Jing},
  booktitle={Proc. Interspeech 2025},
  year={2025}
}

Please cite the above paper if you use the MDCCST Mandarin translation in your research.

Data format

All translated texts are splited to "train (65120 lines)/dev (5663 lines)/test (12492 lines)" as the below original repo did. [https://github.com/HLTCHKUST/cantonese-asr/tree/main/dataset]

Example:

Audio	MDCCST translation
447_2101051829_46868_983.96_985.56	小孩子不要撒谎