Live Analysis
Run the full pipeline on any clip: speech-to-text, vocal affect, text sentiment, then fusion. Calibrated against the 2021 Abu Dhabi Grand Prix corpus.
Running on the ZeroGPU backend
This page is served by a static Space, which has no process to run a model in. The clip is sent to a second Space running the pipeline on ZeroGPU — the same pipeline/ package behind the Race Replay, copied in verbatim at build time rather than reimplemented.
The first call of the day wakes the Space and loads three models, so it can take a minute; later ones are quick. ZeroGPU time is charged to you, not to the Space — signing in to Hugging Face raises the allowance.
Input
Inference runs on a shared ZeroGPU slice, so a warm call takes a few seconds and a cold one takes a minute. The Race Replay is precomputed for exactly this reason.
Result
Nothing analysed yet. Upload or record a clip to see the transcript, the vocal affect, and the fused driver state.