Sparse stimuli for cochlear implants

The aim of the present research is to explore the application of sparse coding principles to the processing within a cochlear implant. These principles would determine what information in noisy speech should be extracted and used to excite the electrode array within the cochlea. The hypothesis is that reducing redundancy in the signal, making it more sparse, would improve speech recognition scores in noise. The proposed sparse coding strategy was based on a combination of ICA (independent component analysis) and PCA (principal component analysis), both operating on the spectrotemporal envelope of the speech signal. The algorithm is tested for speech in quiet and modulated babble noise conditions (signal-to-noise ratios, SNR=15 dB, 10 dB, 5 dB). Results show that the algorithm is beneficial, particularly when baseline performance of listeners is poor. This approach is applicable both to acoustical hearing aids and cochlear implants.