A computational framework can extract spatio-spectro-temporal neural signatures corresponding to hierarchical prediction and prediction errors in a local-global auditory task.