Dysarthric Speech Enhancement using Hybrid model involving Dynamic Time Warping and Feed Forward Neural Network

Abstract

Any neurological impairment to the speech production system results in a range of motor speech disorders known as dysarthria. Speech quality is degraded in dysarthric speech, which is characterized by poor speech articulation. Therefore, it is crucial to enhance or correct dysarthric speech in order to help those who have the condition communicate more effectively. The purpose of this work is to enhance the continuous speech of many dysarthria sufferers. To enhance the intelligibility of speech, Dynamic Time Warping (DTW) with Artificial Neural Network (DTWANN) Speech Enhancement (SE) system has been used. Here, reconstructing the dysarthric signal from the speech signal has been focused on improving intelligibility of the speech. In order to process the dysarthric speech signal in the testing phase, the proposed SE system first trains an Artificial Neural Network (ANN) model. Pairs of normal speech and dysarthric speech utterances is used for training. The outcomes demonstrated that the suggested strategy improves PESQ score to 2.550 which shows significant improvement over PESQ of 1.002 before enhancement.