What is CLARUS? CLARUS is a final-year B.Tech research project studying fairness in AI-based lip-reading (visual speech recognition) systems — that is, whether these systems work equally well for everyone, regardless of skin tone, facial hair, lighting, or head angle.
Why we're collecting this data: Most existing lip-reading datasets come from narrow sources (mainly BBC/TED footage) that don't represent how diverse real speakers actually look. To properly test and improve fairness, we need our own diverse set of recordings — that's where you come in.
Why your contribution matters: Every recording directly helps us measure and reduce bias in this technology. Your data will only be used for this academic research project, as described in the consent notice below.
You'll record 5 short sentences, one at a time. Each recording is capped at 12 seconds and includes a short countdown first.
Make sure your face is clearly visible, well-lit, and that only you are in frame. The checklist below must all turn green before you can continue.
All 5 recordings have been uploaded successfully. You can close this tab now. Thank you for contributing to CLARUS!