Skip to content

Dataset · cvssp

Audio Text Descriptions for Multimodal Models

cvssp/WavCaps

English audio captions generated with ChatGPT for audio-text multimodal research.

What's inside

A weakly-labelled dataset for English audio captioning tasks, created using ChatGPT. The audio clips are originally sourced from open sources FreeSound, BBC Sound Effects, SoundBible, and the labeled subset of AudioSet. The card contains text metadata and descriptions of sound events.

What you can do with it

  • Training audio captioning models to generate text descriptions from sound
  • Multimodal research linking text and audio events
Languages
English
Size
up to 1K rows
Downloads this month
21.5K
Likes
56

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("cvssp/WavCaps")