Modern cochlear implants convert sound into digital impulses, which are sent via implanted electrodes. Each electrode corresponds to a limited range of audio signal frequencies, so the audio is divided into appropriate bands. Currently, a filter bank is used for this purpose, but it is computationally complex and uses wide-bandwidth filters.
This paper presents an innovative method of converting audio signals into neuromorphic impulses using mathematical transformations. To select the best method, the speech signal was encoded using five types of transformations, two sampling methods, and two impulse-generation models. To ascertain the quality of the transmitted signal, the artificial intelligence models XLSR-531 and Whisper-large-v32 were used to transform speech to text.
The experiments carried out showed a significant advantage of logarithmic signal sampling over regular sampling. The transformation for which the speech-to-text models performed best was the discrete wavelet transform (DWT) with a Daubechies-5 wavelet and regular sampling. The Fast Fourier Transform (FFT) with logarithmic sampling yielded slightly worse results.
The constant-Q (CQT) and Walsh-Hadamard (FWHT) transformations performed the worst overall and are not recommended for this system. The applied pulse-generation models did not affect the test results, even after adding Gaussian noise to the generated impulse signal. This neuromorphic approach, particularly when employing the discrete wavelet transform with regular sampling, shows promise for cochlear implantation strategies for partial deafness, enabling precise high-frequency electrical stimulation with shorter electrode arrays while preserving low-frequency residual acoustic hearing, thereby potentially improving speech perception and reducing channel interaction.
This research received no external funding and was conducted as part of the University and the Institute academic and statutory activities. Faculty of Electronics, Telecommunications, and Informatics, Multimedia Systems Department, Gdańsk University of Technology, Gdańsk, 80-233, Poland Institute of Physiology and Pathology of Hearing, Kajetany, 05-830, Poland The authors declare no competing interests. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material.
Extract — continue reading at the source.
