Enhanced sparse speech processing strategy for cochlear implants

To improve the performance of cochlear implants (CI) in noisy environments, a novel strategy is proposed by combining two established algorithms. By combining the Kalman Expectation-Maximize (KEM) algorithm for noise reduction and the SPARSE algorithm for information selection, the new “enhanced SPARSE” strategy gains from the benefits of both. In the new algorithm, the noisy speech is first transferred to the time-frequency domain via a 22- channel filter bank and the envelope in each frequency channel is extracted; secondly, KEM filtering is applied to the envelope in each channel; finally, the SPARSE strategy developed previously in our group is used to generate more sparse stimuli. Here we present results of objective and subjective experiments where KEM was applied to the standard CI speech strategy (Advanced Combination Encoder, ACE) and the novel SPARSE strategies for comparison. Sparseness, measured by kurtosis, of the KEM enhanced simulations was higher than the corresponding output without noise reduction. In subjective listening experiments, six normal hearing listeners were tested with sentences in noise at three signal-to-noise ratios (SNR): 0, 5, 10 dB. Speech intelligibility was better using the enhanced ACE algorithm than the original ACE for all these three SNR, enhanced SPARSE yielded better performance in low SNR (0 dB) than SPARSE, but the advantage was not obvious for the higher SNR (5, 10 dB). We conclude that the suggested strategy shows promise for achieving better speech perception for CI users in the future.