Hugging Face's logo

Datasets:
adjaysagar
/
700hrs-hindi-english-hinglish-high-quality-tts-data

License:
Dataset card Files
xet
 Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

700 hrs Hindi, English, Hinglish High-Quality TTS Data

Processed speech data: Hindi, English, and Hinglish (code-mixed). Includes train/val manifests and the preprocessing script used to generate them.

Download

  • voice_data.zip — Password-protected archive containing:
    • Data directory — audio files, train.jsonl, val.jsonl
    • Preprocessing script — Python script used to build this dataset

The zip file is password-protected. Contact the dataset maintainers for access.

Duration and statistics (from preprocessing logs)

Split Samples Duration
Train 313,703 737.5 h
Val 16,510 38.8 h
Total 330,213 776.2 h

Language breakdown (train):

  • Hindi: 11825 samples
  • English: 50127 samples
  • Hinglish: 69844 samples

Language breakdown (val):

  • Hindi: 7084 | English: 5934 | Hinglish: 3492

Approx. audio directory size: 229.7 GB (uncompressed).

Contents

  • audio/ — WAV files (44.1 kHz or 16 kHz as per preprocessing).
  • train.jsonl, val.jsonl — manifest lines (audio path, text, language, duration, etc.).
  • Preprocessing script — to reproduce or extend the dataset (see config in script for sources).

License / attribution

See individual dataset sources in the preprocessing script. Uploaded for research use.
Downloads last month
11
Total file size:
197 GB