Auditory modelling of the perceptual similarity between piano sounds

In this study an auditory model which predicts psychoacoustic data was applied to the problem of perceptual similarity between piano sounds. The sounds correspond to loudness-balanced recordings of one note played on seven historical pianos that differ in timbre. The similarity between sound pairs was quantified using a 3-AFC discrimination task. To model perceptual similarity, two expected signals were included in the decision stage of an existing auditory model. The simulations were maximally correlated with experimental data when only the initial part of the sounds (0.2 s) was used as input to the model.