README.md

metadata

license: cc-by-4.0
task_categories:
  - text-to-speech
language:
  - gu
size_categories:
  - 1K<n<10K

Gujarati TTS Dataset

SPICOR Gujarati Female TTS dataset in WebDataset format.

Dataset Details

• Total Files: 9100
• Duration: ~33.6 hours
• Speaker: Spk0001 (Female, Age 33)
• Language: Gujarati
• Domains: 19 domains (Agriculture, Entertainment, Finance, Health, Science, Sports, etc.)
• Recording: 48kHz, 24-bit, Studio quality

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("webdataset", data_dir="Chakshu/gujarati-tts/resolve/main/data")

# Access audio and text
for sample in dataset["train"]:
    audio = sample["wav"]  # Audio bytes (WAV format)
    metadata = sample["json"]  # Metadata dict
    print(metadata["text"])  # Gujarati transcription
    print(metadata["speaker_id"])  # Speaker ID
    print(metadata["domain"])  # Domain

Columns

Each sample contains:

• audio (.wav file): Raw WAV audio bytes
• metadata (.json file):
  • text: Gujarati transcription
  • file_id: Unique identifier
  • category: Category code (e.g., SPOR, AGRI)
  • domain: Full domain name
  • speaker_id: Spk0001
  • speaker_gender: Female
  • speaker_age: 33
  • language: gu (Gujarati)

Splits

• train: 8242 files in 17 TAR shards
• test: 858 files in 2 TAR shards

Citation

@misc{SPICOR_TTS_2.0_Corpus,
    Title = {SPICOR TTS_2.0 Corpus - A 57+ hour domain-rich Gujarati TTS Corpus},
    Authors = {Abhayjeet Et al.},
    Year = {2025}
}

License

CC-BY-4.0