Applying Perceptual Distance to the Discrimination of Sounds

In sound compression and synthesis, it is valuable to have objective distance measures which can predict the perceived dissimilarity of two sounds. Furthermore it is desirable also to be able to estimate the extent of the dissimilarity or the quality difference between sounds. Since it is the perceived difference that needs to be quantified it can be expected that measures that are derived from models of hearing will provide the best predictions. An overview of the topic of auditory modelling is presented. An objective measure based on a distance metric applied to auditory excitation patterns is described. This perceptual measure is compared with a traditional spectral distortion measure in the prediction of the discrimination of spectral envelope distortions of vowel-like sounds. The results of a subjective experiment indicate that the perceptual measure is significantly superior.