Modelling and measurement of timbre perception in the electrically stimulated auditory system

Abstract

Music perception in cochlear implant (CI) listeners has been found to be generally unsatisfactory. An improved understanding of music perception in CIs is thus required, where research into the perception of timbre, an important aspect of music, could assist in improving this knowledge base. The aim of this study was to determine what underlies measured timbre perception in cochlear implantees. This was investigated by means of an experimental component and the development of a model of timbre perception in the electrically stimulated auditory system. Timbre perception was first measured in five normal-hearing (NH) and five CI listeners by means of three important timbre features, namely the spectral centroid, the logarithm of the rise time and the spectral irregularity. Discriminations of synthesised tones where these features were independently varied revealed that CI listeners had substantially larger threshold values than NH listeners for each of the timbre features investigated. Confusions of musical instrument timbres were also determined in five CI and five NH listeners by similarity ratings of original and acoustic simulations of musical instrument timbres, respectively. An acoustic model based on a six-channel advanced combination encoder (ACE) processor was developed in order to process 10 musical instrument timbres. The results of the similarity ratings revealed differences in the information conveyed through the timbre features for NH and CI listeners, and indicated that the acoustic model did not accurately represent timbre in the electrically stimulated auditory system, but provided reasonable measurable results which could be compared to timbre perception model predictions. A model of timbre perception was developed by combining the results of the discrimination tasks with signal detection theory, in an attempt to predict the amount of information conveyed through each of the timbre features to both NH and CI listeners. The model was found to predict the experimental results obtained from the similarity ratings for both NH and CI listeners acceptably for each of the three timbre features. This outcome also confirmed the validity of the choice of the three timbre features as the primary features contributing to timbre perception, implying that timbre perception through a CI would be improved if CI processors could be optimised for the transmission of these three important timbre perception features. The model of timbre perception therefore has application in advancing CI research to facilitate music perception in the electrically stimulated auditory system.