Skip to content

Dataset · ARTPARK-IISc

Vaani Indian multilingual speech

ARTPARK-IISc/Vaani

31k hours of image-prompted spontaneous speech across 105 Indian languages, with transcripts for ASR and multimodal tasks.

What's inside

VAANI by IISc and ARTPARK is a multimodal multilingual dataset for India. In phases 1-2: about 31,278 hours of image-prompted spontaneous speech from 156k speakers across 165 districts; 288k images and 105 languages. Transcripts are available for 2,122 hours, distributed almost evenly across districts; languages include ne, as, ml, gu, or, en.

What you can do with it

  • train ASR for Indian languages
  • fine-tune a multimodal speech-image model
  • evaluate TTS and ASR on low-resource languages
Tasks
Automatic speech recognition, Text → speech, Image → text, Text → image
Languages
Nepali, Assamese, Malayalam, Gujarati, Odia, English, Tamil, Urdu and 4 more
Size
10M–100M rows
Downloads this month
37K
Likes
198
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("ARTPARK-IISc/Vaani")