Electroencephalogram-based ling six-sound perception classification using deep learning

Abstract

Ling Six-Sound is a typical sound material widely used in hearing assessment for speech perception, which consists of /m/, /u/, /a/, /i/, /sh/ and /s/. At present, the Ling Six-Sound clinical hearing assessment mainly through subjective audiometry or question-and-answer, there are inaccurate and incomplete drawbacks. With the development of brain-computerinterface, objective perception based on EEG has made great progress. In order to construct an objective evaluation method for Ling Six-Sound perception, this study used oddball paradigm to collect passive EEG induced by Ling Six-Sound, analyzed the time and spatial domain features of the EEG signals, and used four kinds of deep learning models to classify the collected EEG signals. The experimental results show that the performance of the temporal convolutional network with the fusion layer (TCNet-Fusion) is better than other models, and the accuracy for six-classification is 79.78%, which is much higher than the random level. Based on the above results, it can be concluded that the objective Ling Six-Sound perception assessment is feasible, and provide reference for introducing this objective assessment method into clinical hearing assessment in the future.