Spectral-modulation Preserving Dynamic Range Compression Improves Simulated Speech-in-noise Performance of Cochlear Implant Users
* Presenting author
Abstract:
Cochlear implant (CI) users encounter profound difficulties when communicating in noisy environments, potentially due to the inability of exploiting spectral and temporal gaps. The limited spectral resolution and its significant correlation with deteriorated speech perception was shown in numerous studies. Recently proposed approaches to model aided impaired hearing with modified automatic speech recognition systems aim to disentangle the underlying mechanisms which affect speech recognition performance, and to eventually represent speech information in a more supportive way. In prior studies, the Framework for Auditory Discrimination Experiments (FADE) was successfully extended to incorporate hearing impairment by means of absolute hearing thresholds, a supra-threshold level uncertainty, and spectral smearing.In this study, parameter ranges for modeling performances of CI users with FADE were inferred from simulations of the Spectrally-Temporally Modified Ripple Test and the Matrix Sentence Test. This model was then used to evaluate a new approach to dynamic range compression, which enhances spectral modulations suspected to improve speech understanding of CI users in noise. The preservation of specific spectral modulations improved the simulated speech recognition performance, especially in a fluctuating noise condition and for a higher modulation-expanding factor by 7.3dB, indicating a potential to help provide masking release for actual CI users.