Here is the model in action: note how the activity follows the "active speaker", thereby predicting the ventriloquist illusion. This is the first perceptual model capable of predicting the illusion from raw audiovisual footage

Here's the model's response to McGurk stimuli with varying levels of AV sync. The regions of the face with the highest audiovisual correlation elicit strong model activity, and the overall population response is highly correlated to the probability of perceiving the illusion.