VoxCeleb: 1 Million+ Audiovisual Clips for Speaker Recognition

VoxCeleb and VoxCeleb2 provide over 1 million audiovisual clips of human speech from celebrities, compiled by researchers at the University of Oxford. This repository aggregates both versions into a single source containing MP4 video and AAC/WAV audio files.

Use Cases

• Speaker identification using AAC audio tracks
• Lip-reading analysis using MP4 video frames
• Cross-modal verification between facial images and voice recordings

Strengths

• Over 1,000,000 records
• Multi-modal data including MP4 and AAC formats
• CC BY 4.0 license

Limitations

• Requires manual concatenation of multipart zip files before use
• Potential background noise from original broadcast environments

Provenance

Source

Nagrani et al. (University of Oxford)

Collection Method

Extracted from YouTube audiovisual content

Freshness

Last updated February 2026.

Geography

Global