Neural-network clustering for evaluating immersive sound fields

Evaluating immersive sound fields requires a group of trained and experienced listeners. This study investigates whether unsupervised clustering using a self-organizing map can be used to group listeners based on their perceptual responses to immersive reproduction of two orchestral music recordings. The study trained a clustering model using 82 subjects' subjective responses, which consisted of five attribute ratings and one preference rating. Subsequently, the model was validated with a new data set from 19 subjects. The results showed that the model was 84% accurate compared to manual clustering. The clustered group's cognitive characteristics were similar to a previous study, supporting the efficacy of the proposed neural-network clustering. The authors will continue to collect more data to further validate and update the model, which will reliably and quickly evaluate a participant's degree of training and experience to determine their eligibility for relevant subjective evaluations of immersive auditory experiences.