Benchmark RadarAI BENCHMARK PROFILE
RedVox
RedVox is a multilingual safety and fairness benchmark for speech models, built on real voices. It covers unsafe and unfair stereotypical requests across five languages and evaluates models under naturalistic conditions.
- Released
- 2026-06-25
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Speech models are deployed globally but safety evaluations are mostly English-only. RedVox provides a reusable benchmark to assess cross-lingual safety gaps and the impact of spoken input.
Motivation
Speech-capable models are increasingly deployed in real-world applications across languages.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.