A Speech Command Control-Based Recognition System for Dysarthric Patients Based on Deep Learning Technology

Voice control is an important way of controlling mobile devices; however, using it remains a challenge for dysarthric patients. Currently, there are many approaches, such as automatic speech recognition (ASR) systems, being used to help patients devices. However, the large computation power requirement ASR system increases implementation costs. To alleviate this problem, study proposed convolution neural network (CNN) with phonetic posteriorgram (PPG) feature recognize commands, called CNN–PPG; meanwhile, CNN model Mel-frequency cepstral coefficient (CNN–MFCC model) and ASR-based systems were comparison. The experiment results show that CNN–PPG provided 93.49% accuracy, better than CNN–MFCC (65.67%) (89.59%). Additionally, smaller size comprising only 54% parameter numbers compared system; hence, could reduce costs users. These findings suggest augment communication device via commands in future.