| 23 |
Unveiling the auditory system with a neural networkApproaches to cochlear implant and binaural processing 
|
|---|
Although there has been much research on human auditory characteristics, it is difficult to directly address the question of what kinds of input and training lead to the acquisition of these characteristics. In this work, we tackled the clinical and academic aspects of the question by using artificial neural networks (ANNs), and obtained new findings in each case. (1) It is known that people with hearing loss who wear cochlear implants (CIs) have difficulty with pitch perception, but we confirmed that the cochlear implant signal contains a certain amount of pitch information, suggesting that the difficulty in pitch perception is mainly due to physiological factors. (2) By measuring the response of a single unit in an artificial neural network trained to recognize natural sounds, we found out the ANN units (neurons) with the binaural processing characteristics were equivalent to those found in the auditory system of animals. We believe that cochlear implant users may be able to achieve normal pitch perception under a clean environment after an appropriate rehabilitation. We also hope to further develop AI technology and CI devices that behave in a human-like manner by advancing auditory information processing technology that is consistent with the auditory nervous system.
[1] T. Ashihara, S. Furukawa, M. Kashino, “F0 estimation from simulated cochlear-implant signals by using a DNN model,” Spring Meeting of Acoustic Society of Japan, 2022.
[2] T. Koumura, H. Terashima, S. Furukawa, “Emergence of ITD selectivity in a deep neural network trained for binaural natural sound detection,” in Proc. 42nd Association for Research in Otolaryngology (ARO) MidWinter Meeting, 2019.
[3] TC. Yin, JC. Chan, DR. Irvine, “Effects of interaural time delays of noise stimuli on low-frequency cells in the cat's inferior colliculus. I. Responses to wideband noise,” Journal of Neurophysiology, pp. 280–300, 1986.
Takanori Ashihara / Human Informatics Laboratories
Takuya Koumura / Sensory Representation Research Group, Human and Information Science Laboratory
Email: cs-openhouse-ml@hco.ntt.co.jp