This dataset contains Vietnamese speech data from MISA company, featuring multiple speakers reading various texts. The dataset is designed for Text-to-Speech (TTS) and Automatic Speech Recognition (ASR) tasks in Vietnamese.

Dataset Structure

Each sample contains:

audio: Audio file (WAV format, 22050 Hz)
text: Vietnamese transcript of the audio
speaker_id: Identifier for the speaker
sample_rate: Audio sample rate

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("nmcuong/misa_voice_private")

# Access a sample
sample = dataset["train"][0]
print(f"Text: {sample['text']}")
print(f"Speaker: {sample['speaker_id']}")
print(f"Duration: {sample['duration']}s")

# Play audio (requires IPython)
import IPython.display as ipd
ipd.Audio(sample["audio"]["array"], rate=sample["audio"]["sampling_rate"])

Data Processing

The audio files have been processed with:

Normalization to 22050 Hz sample rate
Silence removal and padding
Quality filtering and validation

License

This dataset is released under the MIT License.

Citation

@dataset{misa_vietnamese_voice_2025,
  title={MISA Vietnamese Voice Dataset},
  author={MISA Corporation},
  year={2025},
  url={https://huggingface.co/datasets/nmcuong/misa_voice_private}
}

Contact

For questions about this dataset, please contact MISA Corporation.

Downloads last month: 4