Acoustic models of consonant recognition in cochlear implant users
Normal-hearing adults have no difficulty in recognising consonants accurately, even in moderately adverse listening conditions. By contrast, users of multichannel cochlear implants have difficulty with the accurate perception of consonants, even in good listening conditions. Cochlear implant users are known to show systematic deficits in recognition of consonant features, with perception of the place feature, which relies on spectral information, being worst. These deficits may be attributed both to signal distortions introduced by the processing of the implants and to other factors, in particular the spectrotemporal distortions which occur at the interface between electrode array and auditory nervous system, including cross-channel interaction. The objective of the work reported here was to attempt to partial out the relative contribution of these different factors to consonant recognition. This was achieved by comparing cochlear implant users’ perceptual errors, analysed in terms of information transmission, with errors made by normal-hearing subjects listening to acoustic models of implant processing, in various conditions. Two initial experiments were undertaken to develop and refine an acoustic model of the Nucleus 24 cochlear implant. Findings from these two experiments informed the design of the main acoustic model experiment, which was undertaken in parallel with a further experiment involving users of the Nucleus 24 device. In both experiments, subjects listened to nonsense syllables with and without the addition of stationary background noise, in three different configurations of implant processing parameters. Additionally, in the acoustic model experiment, a simulation of cross-channel spread of excitation, or “channel interaction”, was varied. Results showed that acoustic model experiments were predictive of the pattern of consonant feature transmission in cochlear implant users with better baseline consonant recognition scores. Deficits in consonant recognition in this subgroup could be explained by the loss of phonemically relevant acoustic information in speech due to the nature of cochlear implant processing, while channel interaction appeared to play a smaller role in accounting for problems in consonant recognition. The work also evaluated the effect of changes in channel number and stimulation rate and failed to find any changes in consonant recognition as these parameters were varied. The lack of a stimulation rate effect was consistent with acoustic measurements of the temporal modulation transfer function of the processor, which showed almost no change across stimulation rates.