Built the SensAffect OmniCNN with five independently gated modalities:

* Visual expression, posture, and scene
* Vocal prosody
* Language
* Physiological signals
* Human-to-machine behavioural signals

Behaviour includes latency, hesitation, correction, repetition, navigation, interruptions, abandonment, help-seeking, overrides, cadence, pointer movement, turn-taking, and task progress.

Outputs include:

* 12-class emotion probabilities
* Valence–arousal–dominance
* Intensity and uncertainty
* Per-modality contribution weights
* A 128-dimensional SensAffect Imprint embedding for MMRY
* Missing-modality support
* ONNX export path for edge deployment
* Privacy-aware behavioural schema

[Download ](sandbox:/workspace/scratch/31eb0a4d311c/SensAffect-OmniCNN-v1.zip)

The model code passed Python compilation and the ZIP was integrity-checked. Full numerical testing requires installing the included PyTorch dependencies.
