摘要：In this paper, a novel algorithm is proposed that uses the concepts of Kepstrum approach and Mel filtering for processing of speech signals for cochlear implants. Here the speech signals are first divided into various frames of approximately 10 mS duration. FFT and periodogram of each frame is calculated after which the signal is passed through the Mel filter bank. Then the kepstrum coefficients are calculated, which are used for estimation of the coefficients of the whitening filter. This filter is used for the noise suppression of the input speech signal. The algorithm is tested on various speech signals and the SNR for the speech signal before and after processing are tabulated. The results show that there is a good amount of increase in the SNR which indicates the efficiency of the algorithm in noise suppression.
