A beamformer post-filter for cochlear implant noise reduction

Abstract

Cochlear implant users have limited ability to understand speech in noisy conditions. Signal processing methods to address this issue that use multiple microphones typically use beamforming to perform noise reduction. However, the effectiveness of the beamformer is diminished as the number of interfering noises increases and the acoustic environment becomes more diffuse. A multi-microphone noise reduction algorithm that aims to address this issue is presented in this study. The algorithm uses spatial filtering to estimate the signal-to-noise ratio (SNR) and attenuates time-frequency elements that have poor SNR. The algorithm was evaluated by measuring intelligibility of speech embedded in 4-talker babble where the interfering talkers were spatially separated and changed location during the test. Twelve cochlear implant users took part in the evaluation, which demonstrated a significant mean improvement of 4.6 dB (standard error 0.4, P < 0.001) in speech reception threshold compared to an adaptive beamformer. The results suggest that a substantial improvement in performance can be gained for cochlear implant users in noisy environments where the noise is spatially separated from the target speech.

Two-microphone spatial filtering provides speech reception benefits for cochlear implant users in difficult acoustic environments

“…Hersbach et al (2013) demonstrated an initial step in this direction by using aspects of spatial-filtering to serve as a post-filter for null-steering beamforming. They demonstrated that such a combination yielded synergistic benefits; however, more sophisticated combinations of null-steering beamforming and spatial-filtering could be implemented.…”

“…With a fast-acting time constant of only 10 ms, Fennec takes such an aggressive approach. Hersbach et al (2013) evaluated another related approach that uses forward and backward pointing directional microphones to estimate SNR and generate time/frequency attenuation analogous to Eq. (3).…”

“…This distinction can yield performance differences, particularly in more complex environments where the target speech is degraded by reverberation and by multiple, simultaneous, noise sources. Hersbach et al (2013) suggested it was this dependence on coherence which caused the Yousefian and Loizou (2013) algorithm to deteriorate in reverberant environments. Specifically, speech reception benefits provided by the coherence-based algorithm decreased from 5 to 10 dB in an anechoic environment to 0 to 2 dB in a moderately reverberant room (T 60 ¼ 465 ms).…”

“…Specifically, speech reception benefits provided by the coherence-based algorithm decreased from 5 to 10 dB in an anechoic environment to 0 to 2 dB in a moderately reverberant room (T 60 ¼ 465 ms). Hersbach et al (2013) suggested an alternate approach, using null-steering beamforming as a front-end to a secondary post-filter using spectral-attenuation of low-SNR components. As they only evaluated this method in a sound-treated, low-reverberation environment, it is unknown the extent to which that approach is robust to the detrimental effects of higher levels of reverberation.…”

“…They found that this benefit was affected by reverberation with benefits decreasing to 4 to 7 dB SRT and to 1 to 2 dB SRT when tested in rooms with reverberation times (T 60 ) of 220 and 465 ms, respectively. Hersbach et al (2013) investigated the combination of null-steering beamforming with a spatial-filtering algorithm similar to the approach of Goldsworthy (2009) and of Loizou (2012, 2013). Specifically, they used a null-steering beamformer to approximate the instantaneous SNR and then selectively attenuate spectral components that statistically appeared to be dominated by noise.…”