VoxCeleb1

Speaker identification/verification corpus of 153,516 utterances from 1251 celebrities extracted from YouTube interview videos.

• Task: speaker
• Languages: multi
• Hours: 352
• Domain: celebrity interviews (YouTube)
• License: CC BY-SA 4.0
• Homepage: https://www.robots.ox.ac.uk/~vgg/data/voxceleb/vox1.html
• Paper: https://arxiv.org/abs/1706.08612

Recommendation

Default in-the-wild speaker-verification benchmark — pick when you want a public, well-comparable baseline before scaling to VoxCeleb2. Speakers are multi-nationality (speech predominantly English); YouTube provenance means clip-level quality varies and some links rot.

Getting the data

Obtain from the dataset homepage.

Same VGG distribution model as VoxCeleb2; requires accepting the access terms on the VGG site even though the corpus itself is CC BY-SA 4.0.

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at voxkitchen/templates/pipelines/speaker-analysis.yaml — run it with vkit docker run.