Enhancement of Speech Auditory Brainstem Responses Using Adaptive Filters

Abstract

Several adaptive filters were investigated to enhance speech auditory brainstem responses (speech ABR). The objective was to shorten the long recording time currently needed by the standard coherent averaging method to obtain acceptable performance, which has limited the clinical adoption of speech ABR. Five algorithms were implemented: Wiener Filter (WF), Steepest Descent (SD), Adaptive Noise Cancellation (ANC) based on Least-Mean-Square error (LMS) and normalized LMS error (nLMS), and a multi-adaptive cascade combination of SD and LMS. The performance of the adaptive filters was assessed on speech ABR data gathered from several subjects and compared with coherent averaging using the overall Signal-to-Noise Ratio (SNR), the local SNR around the fundamental frequency and the first formant, and Mean-Square-Error (MSE) in the time and frequency domains. The adaptive filters could reduce the time needed, by at least one order of magnitude, for obtaining comparable signal quality as that obtained with coherent averaging.