Skip to content

Dataset · facebook

Multimodal human interaction corpus

facebook/seamless-interaction

4,000+ hours of audio and video of in-person interactions by 4,000+ participants for AI research.

What's inside

The dataset contains over 4,000 hours of face-to-face in-person audio and video interactions. More than 4,000 people participated in the recordings. This is a large-scale multimodal corpus for AI research on human communication.

What you can do with it

  • Train a multimodal audio-video interaction model
  • Investigate non-verbal cues in communication
  • Evaluate human interaction analysis systems
Downloads this month
195K
Likes
223

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/seamless-interaction")