Hugging Face's logo

Datasets:
Cossale
/
gujarati-speech-corpus

Tasks:
Automatic Speech Recognition
Modalities:
Audio
Text
Formats:
parquet
Languages:
Gujarati
Size:
100K - 1M
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio 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.

Gujarati Speech Corpus

A merged and standardized collection of 9 Gujarati speech datasets — 1,115 hours of audio across 365,000 samples (~75 GB).

Sources

Source Hours Samples Sample Rate
gujarati-asr-16kHz 400.1h 81,035 16 kHz
SPRING_INX_Gujarati_R1 394.2h 72,058 16 kHz
SPRING_INX_Gujarati_R2 138.8h 26,927 16 kHz
gujarati-interspeech 90.0h 25,882 16 kHz
gujarati-speech-dataset 68.5h 152,713 16 kHz
Gujarati_VDB_Female 8.6h 2,219 48 kHz
Gujarati_VDB_Male 7.2h 2,053 48 kHz
Gujarati_Vistaar_DB 6.2h 1,766 16 kHz
speaker1-gujarati-tts-dataset 1.6h 347 22.05 kHz

Columns

  • audio - Embedded WAV audio bytes (struct with bytes field)
  • text - Gujarati transcript (standardized from transcript/transcription/text)
  • source - Original dataset name
  • sample_rate - Original sample rate in Hz (16kHz, 22.05kHz, or 48kHz)
  • utterance_id - Utterance ID (nullable, from SPRING_INX datasets)
  • speaker_id - Speaker ID (nullable, from gujarati-interspeech)
  • id - Sample ID (nullable, from speaker1-gujarati-tts-dataset)
  • split - Original split (train/test/validation)

Usage 

from datasets import load_dataset
ds = load_dataset("Cossale/gujarati-speech-corpus")

Notes

  • Audio is preserved at original sample rates (no resampling)
  • Text columns were standardized to text across all sources
  • Each parquet shard is under 5 GB for HuggingFace compatibility
Downloads last month
13