Speech perception difficulties with cochlear implants in noisy environments: investigating effects of channel interaction and enhancing speech perception by using machine learning Seminar

Cochlear implant (CI) listeners struggle to understand speech in background noise. Interactions between electrode channels due to current spread increase the masking of speech by noise and reduce the effective number of channels a CI provides. Therefore, strategies that can reduce channel interaction or the impact of background noise have the potential to improve speech-in-noise perception by CI listeners. I will describe the outcomes of two recent studies that tackled this challenge from two angles:

In the first study, we investigated the effects of channel interaction on speech-in-noise perception and its association with spectro-temporal acuity in a listening study with 12 CI users. We measured speech reception thresholds in noise as a function of the amount of channel interaction (“blurring”) applied to either all, or 5 out of 15, electrode channels. Performance for each listener remained roughly constant as the amount of blurring applied to all channels increased up to some knee point, above which it deteriorated. This knee point correlated with performance on a non-speech spectro-temporal task. Surprisingly, even extreme amounts of blurring applied to only 5 out of 15 channels did not affect performance overall. Implications of these findings will be discussed.

In the second study, we evaluated a machine learning algorithm, a recurrent neural network (RNN), for enhancing speech in non-stationary noise and its benefits were evaluated with 10 CI users. The RNN was trained using speech from many talkers mixed with multi-talker babble or traffic noise recordings. Its performance was evaluated using speech from an unseen talker mixed with different noise recordings of the same class, either babble or traffic noise. The experimental results showed significantly improved intelligibility of speech in babble noise but not in traffic noise. CI subjects rated the processed stimuli as significantly better in terms of perceptual quality than unprocessed stimuli for both babble and traffic noise. These results confirm and extend previous findings of improved speech perception by CI users to mostly unseen acoustic conditions with realistic and non-stationary noise.

Acknowledgements
This work was funded by Action on Hearing Loss (UK, grant number 82 and F92).