Personal tools
 
You are here: McMaster Institute for Music and the Mind > Publications > Analysis of Spatiotemporal Pattern Correction Using a Computational Model of the Auditory Periphery

Timothy j Zeyl and Ian C Bruce (2014)

Analysis of Spatiotemporal Pattern Correction Using a Computational Model of the Auditory Periphery

Ear and Hearing, 35:246-255.

Objectives: The purpose of this study was to determine the cause of poor experimental performance of a spatiotemporal pattern correction (SPC) scheme that has been proposed as a hearing aid algorithm and to determine contexts in which it may provide benefit. The SPC scheme is intended to compensate for altered phase response and group delay differences in the auditory nerve spiking patterns in impaired ears. Based on theoretical models of loudness and the hypothesized importance of temporal fine structure for intelligibility, the compensations of the SPC scheme are expected to provide benefit; however, preliminary experiments revealed that listeners preferred unprocessed or minimally processed speech as opposed to complete SPC processed speech.

Design: An improved version of the SPC scheme was evaluated with a computational auditory model in response to a synthesized vowel at multiple SPLs. The impaired model auditory nerve response to SPC-aided stimuli was compared to the unaided stimuli for spectrotemporal response similarity to the healthy auditory model. This comparison included analysis of synchronized rate across auditory nerve characteristic frequencies and a measure of relative phase response of auditory nerve fibers to complex stimuli derived from cross-correlations.

Results: Analysis indicates that SPC can improve a metric of relative phase response at low SPLs, but may do so at the cost of decreased spectrotemporal response similarity to the healthy auditory model and degraded synchrony to vowel formants. In-depth analysis identifies several technical and conceptual problems associated with SPC that need to be addressed. These include the following: (1) a nonflat frequency response through the analysis–synthesis filterbank that results from time-varying changes in the relative temporal alignment of filterbank channels, (2) group delay corrections that are based on incorrect frequencies because of spread of synchrony in auditory nerve responses, and (3) frequency modulations in the processed signal created by the insertion of delays.

Conclusions: Despite these issues, SPC provided benefit to an error metric derived from auditory nerve response cross-correlations at low SPLs, which may mean phase adjustment is achieved at the expense of other metrics, but could be beneficial for low-level speech.

hearing-aids, auditory nerve, computational modelling