metadata

license: cc-by-4.0
task_categories:
  - automatic-speech-recognition
  - text-to-speech
language:
  - vi
pretty_name: VLSP 2020 - VinAI - ASR challenge dataset
size_categories:
  - 10K<n<100K
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: transcription
      dtype: string
  splits:
    - name: train
      num_bytes: 17159347574.893
      num_examples: 56427
  download_size: 11649243045
  dataset_size: 17159347574.893
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

unofficial mirror of VLSP 2020 - VinAI - ASR challenge dataset

official announcement:

tiếng việt: https://institute.vinbigdata.org/events/vinbigdata-chia-se-100-gio-du-lieu-tieng-noi-cho-cong-dong/
in eglish: https://institute.vinbigdata.org/en/events/vinbigdata-shares-100-hour-data-for-the-community/
VLSP 2020 workshop: https://vlsp.org.vn/vlsp2020

official download: https://drive.google.com/file/d/1vUSxdORDxk-ePUt-bUVDahpoXiqKchMx/view?usp=sharing

contact: info@vinbigdata.org

100h, 56.4k samples, accuracy 96%

pre-process: merge all transcript text files into 1, remove token <unk>

need to do: check misspelling, restore foreign words phonetised to vietnamese

usage with HuggingFace:

# pip install -q "datasets[audio]"
from datasets import load_dataset
from torch.utils.data import DataLoader

dataset = load_dataset("doof-ferb/vlsp2020_vinai_100h", split="train", streaming=True)
dataset.set_format(type="torch", columns=["audio", "transcription"])
dataloader = DataLoader(dataset, batch_size=4)