AURA

INSTRUMENT · EVALUATE

Does the reading agree with the labels?

Score Aura's expressed-mood reading against video that people have already labelled for pleasantness (valence) and energy (arousal). Everything runs here, in this browser: no video, label or result leaves this machine.

1. Video and labels

One or more clips in any format this browser plays. Each clip is read at one moment per step, exactly as the live app would read it.

A CSV with a header row — file, time, valence, arousal — with time in seconds (leave it out for one label per clip), valence from −1 to 1 and arousal from 0 to 1. A 1–9 SAM scale is recognised. Files are matched by name.

2. Report

Nothing run yet. The self-check draws a face that smiles for ten seconds and rests for ten, with matching labels, and scores the reading against them — it proves the instrument, not the model.

The features CSV holds, for every moment, the labels and the strength of each of Aura's cues: the table a learned weighting is trained from.

3. Where labelled video comes from

Public, in-the-wild, per-moment labels: AFEW-VA, SEWA and Aff-Wild2 — each free for research after a licence request to its authors. Or a consented study of your own: volunteers, their own reports of their mood, and two independent raters, which also calibrates the model for the faces and light it will meet.