Speech and music recognition with acoustic simulations of a harmonic single sideband encoding strategy for cochlear implants.

Abstract

Cochlear implant users experience difficulties in speech perception in noise and music perception. Previous studies suggested that these difficulties are partially due to the lack of temporal fine structure (TFS) coding in cochlear implants [Moore, JARO (2008)]. TFSs are generally high-frequency signals; therefore, delivering perceivable TFS cues in electrical hearing is a challenge because the temporal sensitivity in cochlear implants is typically restricted to 300 Hz and below. A recently-proposed strategy, harmonic single sideband encoder (HSSE), could potentially use the harmonic structure of sounds to coherently demodulate TFS information [Li et al., ICASSP (2010)]. To evaluate these potential benefits to speech and music perception, this study compared the performance of five normal hearing listeners using four- and eight-channel sinusoidal vocoder simulations of HSSE and the continuous interleaved stimulation (CIS) strategy. Scores on the following tasks were measured: sentence recognition in noise and single-talker maskers, melody and timbre recognition, and Mandarin tone discrimination. In all of the tasks, HSSE listeners obtained a significant performance improvement over the CIS vocoder. Furthermore, simulation results from an electrically-stimulated model of the auditory nerve fibers demonstrated that HSSE is capable of encoding temporal pitch cues better than CIS. [This study was supported by NIH R01-DC007525, P30-DC004661, and T32-DC005361; AFOSR Grant FA9550061019; CGF 657635; ITHS 620491; Advanced Bionics Graduate Fellowship; and NSF TG-IBN090004.]