Classification of Dysarthric Speech According to the Severity of Impairment: an Analysis of Acoustic Features

Abstract

The automatic speech recognition (ASR) system is increasingly being applied as assistive technology in the speech impaired community, for individuals with physical disabilities such as dysarthric speakers. However, the effectiveness of the ASR system in recognizing dysarthric speech can be disadvantaged by data sparsity, either in the coverage of the language, or the size of the existing speech database, not counting the severity of the speech impairment. This study examines the acoustic features and feature selection methods that can be used to improve the classification of dysarthric speech, based on the severity of the impairment. For the purpose of this study, we incorporated four acoustic features including prosody, spectral, cepstral, and voice quality and seven feature selection methods which encompassed Interaction Capping (ICAP), Conditional Information Feature Extraction (CIFE), Conditional Mutual Information Maximization (CMIM), Double Input Symmetrical Relevance (DISR), Joint Mutual Information (JMI), Conditional redundancy (Condred) and Relief. Further to that, we engaged six classification algorithms like Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Artificial Neural Network (ANN), Classification and Regression Tree (CART), Naive Bayes (NB), and Random Forest (RF) in our experiment. The classification accuracy of our experiments ranges from 40.41% to 95.80%.