The impact of noise power estimation on speech intelligibility in cochlear-implant speech coding strategies

Thomas Bentsen,Stefan J Mauger,Tobias May,Abigail A Kressner,Torsten Dau

doi:10.1121/1.5089887

Abstract

The advanced combination encoder (ACE™) is an established speech-coding strategy in cochlear-implant processing that selects a number of frequency channels based on amplitudes. However, speech intelligibility outcomes with this strategy are limited in noisy conditions. To improve speech intelligibility, either noise-dominant channels can be attenuated prior to ACE™ with noise reduction or, alternatively, channels can be selected based on estimated signal-to-noise ratios. A noise power estimation stage is, therefore, required. This study investigated the impact of noise power estimation in noise-reduction and channel-selection strategies. Results imply that estimation with improved noise-tracking capabilities does not necessarily translate into increased speech intelligibility.

Highlights

In cochlear implant (CI) processing, a signal is decomposed into frequency channels and the signal level is used to determine the electrode stimulation intensity
The present study investigated the impact of the SPPbased estimator in a range of noise-reduction and channelselection strategies on the speech intelligibility outcome in CI recipients
In speech-weighted noise the existing noise-reduction strategy (NR-MS & ACETM) improved the speech reception threshold (SRT) compared to ACETM alone by about 1.6 dB (p < 0.01) [Fig. 1(a)], which is consistent with previously reported findings (Dawson et al, 2011; Mauger et al, 2012a; Mauger et al, 2012b)

Summary

Introduction

In cochlear implant (CI) processing, a signal is decomposed into frequency channels and the signal level is used to determine the electrode stimulation intensity. Speech intelligibility outcomes with ACETM in noisy conditions with low signal-to-noise ratios (SNRs) are limited primarily because: (i) the channels with the largest amplitudes can be noise-dominated instead of speech-dominated and (ii) ACETM always selects a fixed number of channels when the signal amplitude is above a predefined threshold, irrespective of whether speech is present or absent (Hu and Loizou, 2008). In an attempt to improve the speech intelligibility in these noisy conditions, a range of different speech-coding strategies have been developed. Another group of strategies selects which channels to stimulate directly based on an SNR criterion (Hu and Loizou, 2008). A channel is selected only if the SNR is above a local criterion (LC)

Methods

Results

Conclusion