Dataset · laion
Massive short audio corpus with text captions
laion/laion-audio-xl
Over 366 million audio clips with captions used to train LAION CLAP2 models.
What's inside
A large-scale dataset of 366,007,518 audio clips up to 30 seconds long with text descriptions, collected from 462 source datasets. Over 4.3 million recordings are stored directly in the repository as audio files (divided by license types), while the remaining 361.6 million are provided as links with the same descriptions and analysis fields. Audio, text, and tabular modalities are represented.
What you can do with it
- Training multimodal audio-text alignment models (such as CLAP)
- Training text-to-audio generation models
- Audio feature extraction and sound classification
- Tasks
- Audio classification, Text → audio, Embeddings
- Size
- 100M–1B rows
- Downloads this month
- 24.9K
- Likes
- 88
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("laion/laion-audio-xl")