Modelling the detectability of changes in auditory signals

A model for determining the detectability of changes to auditory signals in monaural two interval listening experiments is proposed. The sensation associated with each interval is modeled as a log excitation vector corrupted by noise. The decision process in the model is based on an optimum unweighted sum of the difference in the log excitations for the two intervals. The performance of the model is assessed using results from psychoacoustical experiments involving single component and multiple component target detection in broadband noise with and without uncertainty as to target frequency. The model is also used to predict results from experiments involving intensity discrimination, variable bandwidth maskers and masker addition. The application of the model to wideband audio coding is discussed.<>