Speech Event Recognition Model for People with Dysarthria Based on Deep Learning

Abstract

Dysarthria is a problem faced by many patients with special diseases, which causes speakers to have unclear pronunciation. In order to better understand the speech events expressed by patients with dysarthria, this article proposes a new speech event recognition model based on deep learning. The model takes speech clips as input, uses Gram angle field to retain the original features of the time series, uses Conformer to extract local features and global features of the sequence, and finally uses ResNet as a classification model. Experimental results on the EasyCall corpus data set show that the model proposed in the article has good recognition results.