sdialog/voices-asrfairness

ASR Fairness Dataset voice samples for SDialog

Fair-Speech dataset, a public corpus containing 26,500 utterances from 593 compensated U.S. participants with self-reported demographics including age, gender, ethnicity, location, and native language status.
The dataset spans seven voice assistant domains: music, capture, utilities, notification control, messaging, calling, and dictation.

Total number of unique speakers: 593
Total number of utterances: 26,471
Average per utterance length: 7.42 seconds
100% US (excludes speakers in IL, WA and TX)

• https://ai.meta.com/datasets/speech-fairness-dataset/

• FairASR: Fair Audio Contrastive Learning for Automatic Speech Recognition

  • https://arxiv.org/pdf/2506.10747v1
  • code: https://github.com/JongSuk1/FairASR