The effect of gain thresholds on speech intelligibility for statistical model based noise reduction for cochlear implants: A simulation based verification

摘要

Noise corruption can dramatically decrease the speech intelligibility for listeners with cochlear implants (CI). Noise reduction is a key point in CI speech processing strategy. This paper proposes a statistical model based noise reduction algorithm for-CIs. A realistic noise estimator, which requires no prior knowledge of the noise, is adopted for noise estimation. An improved method for determining the user-specific gain function is proposed, in which the apparent gain threshold is incorporated to compute the optimal parameters, with which the optimal gain function for noise suppression can be determined accordingly. Vocoder simulation perceptual experiments with normal hearing listeners shows that the proposed algorithm can significantly improve the speech intelligibility of the denoised speech.