BASIS-Bench Data Resources

About

We release three data resources with BASIS-Bench: Spoken StereoSet+ is an improved stereotypical bias dataset for age and gender stereotypical bias assessment. BASIS-Health/Product are gender personalization datasets. ELIP-Pair is an age personalization dataset. All of these datasets contain synthetic voices only.

Downloading the data

• Spoken StereoSet+: Download link. CC-BY-NC-SA-4.0
• BASIS-Health/Product: Download link. NAVER LABS Europe's non-commercial license
• ELIP-Pair: Download link. CC-BY-NC-SA-4.0

Citing us

When using our datasets, please cite the following paper:

@misc{basisbench2026,
  title         = {Personalization Without Stereotyping: Do Speech {LLMs} Know When to Turn a Deaf Ear?},
  author        = {Ambilduke, Kshitij and Brun, Caroline and Thonet, Thibaut and Nikoulina, Vassilina and Boito, Marcely Zanon},
  year          = {2026},
  eprint        = {2610.13550},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2610.13550}
}