metadata

license: apache-2.0
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: transcription
      dtype: string
  splits:
    - name: train
      num_bytes: 131422878531.915
      num_examples: 1026047
  download_size: 130501007846
  dataset_size: 131422878531.915
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
language:
  - vi
task_categories:
  - automatic-speech-recognition

VSV-1100: Vietnamese social voice dataset

Introdution

This dataset includes over 1,100 hours of speech data. The voice samples were collected from a variety of social resources, ensuring a diverse representation of accents (north, central, south), dialects, and speaking styles. This diversity makes the dataset particularly valuable for training and evaluating ASR models, as it enhances their ability to accurately recognize and transcribe speech across different linguistic and cultural contexts.

This dataset is developed by Multimedia Communications Laboratory - University of Information Technology (MMLab-UIT)

See our model: https://github.com/NhutP/ViWhisper

Statistics

Number of samples: 1,026,047
Sampling rate: 16000
Size: 131 GB

Example

{'audio': {'path': None,
           'array': array([....]),
           'sampling_rate': 16000},
 'transcription': 'bạn có thể tìm ra cách khắc phục giai đoạn hai và ba tốt hơn'}

Usage

from datasets import load_dataset
# load the dataset
ds = load_dataset("NhutP/VSV-1100", split= 'train')
# load in streaming mode
ds = load_dataset("NhutP/VSV-1100", split= 'train', streaming=True)
audio_array = ds[0]['audio']['array']
transcription = ds[0]['transcription']

Contributors

Citation

@misc{VSV-1100,
    author = {Pham Quang Nhut and Duong Pham Hoang Anh and Nguyen Vinh Tiep},
    title = {VSV-1100: Vietnamese social voice dataset},
    url = {https://github.com/NhutP/VSV-1100},
    year = {2024}
}

Also, please give us a star on github: https://github.com/NhutP/VSV-1100 if you find our project useful

Contact me at: [email protected] (Pham Quang Nhut)