Mini VoxCeleb1 Speaker Recognition Dataset

A 2022 subset of the VoxCeleb1 dataset curated by s3prl for speaker recognition tasks. It contains a reduced number of audio clips from celebrity interviews sourced from YouTube videos. The dataset is hosted on Hugging Face.

Use Cases

• Train speaker embedding models using audio clips labeled by speaker identity.
• Benchmark speaker verification systems by matching or distinguishing audio samples from known speakers.
• Perform speaker diarization experiments by identifying segments from different speakers within audio.
• Develop few-shot learning models for speaker recognition using the limited speaker set.

Strengths

• Subset derived from the well-known VoxCeleb1 benchmark dataset.
• Hosted on Hugging Face for easy integration with modern ML frameworks.

Limitations

• Small sample size (n1K category) limits statistical power and model generalization.
• Limited demographic and geographic diversity, primarily containing English-language speech from celebrities.

Provenance

Source

VoxCeleb1 dataset (YouTube interviews).

Collection Method

Audio extracted from online video interviews.

Time Range

Original VoxCeleb1 videos span 2010-2017.

Freshness

Last updated in June 2022.

Geography

Primarily English-speaking regions, with a focus on celebrities.