PROJECT DESCRIPTION

PSYC450: Neural Network Models of Social + Cognitive Processes
Dr. Stephen Read | University of Southern California | Fall 2013

Given a musical stimulus, this neural network model, created with Emergent Neural Network Simulator, can predict the elicited affective response of a listener. 

HOW IT WORKS

I trained the network to measure the psychoacoustic factors of tempo, volume, sharpness, and novelty of a stimulus, and use the data to quantify a listener's experience of arousal and valence. 

Because the relationship between music and emotion is not linear, this model takes a spatiotemporal approach by utilizing attributes of long and short-term memory in the form of connection weights.

INPUT LAYERS -- 10x1 GEOMETRY

1. Tempo /beats per measure
2. Volume /piano to forte
3. Sharpness /staccato to legato
4. Novelty /degree of familiarity
Input layers composed of ten nodes each, creating a sliding scale that can measure continuous quantities. 

HIDDEN LAYER -- 25x1 GEOMETRY

Input layers connected to hidden layer; hidden layer bi-directionally connected to output layers. 

OUTPUT LAYERS -- SCALAR VALUES

1. Arousal /physiological response
2. Valence /positive or negative emotional value