metadata

license: cc-by-sa-4.0
language:
  - th
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: sentence
      dtype: string
    - name: thai_sentence
      dtype: string
    - name: dialect_type
      dtype: string
    - name: utterance
      dtype: string
  splits:
    - name: train
      num_bytes: 554017600.528
      num_examples: 40704
  download_size: 508770535
  dataset_size: 554017600.528
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Porjai-Thai-voice-dataset-korat

This corpus contains a officially split of 700 hours for Central Thai, and 40 hours for the three dialect each. The corpus is designed such that there are some parallel sentences between the dialects, making it suitable for Speech and Machine translation research.

Our demo ASR model can be found at https://www.cmkl.ac.th/research/porjai. The Thai Central data was collected using Wang Data Market.

Since parts of this corpus are in the ML-SUPERB challenge, the test sets are not released in this github and would be released subsequently in ML-SUPERB.

The baseline models of our corpus are at:
Thai-central
Khummuang
Korat
Pattani

The Thai-dialect Corpus is licensed under CC-BY-SA 4.0.

Acknowledgements

This dataset was created with support from the PMU-C grant (Thai Language Automatic Speech Recognition Interface for Community E-Commerce, C10F630122) and compute support from the Apex cluster team. Some evaluation data was donated by Wang.

Citation

@inproceedings{suwanbandit23_interspeech,
  author={Artit Suwanbandit and Burin Naowarat and Orathai Sangpetch and Ekapol Chuangsuwanich},
  title={{Thai Dialect Corpus and Transfer-based Curriculum Learning Investigation for Dialect Automatic Speech Recognition}},
  year=2023,
  booktitle={Proc. INTERSPEECH 2023},
  pages={4069--4073},
  doi={10.21437/Interspeech.2023-1828}
}