Design and Development of a Speech Recognition System for Autistic People

Abstract

Autism spectrum disorder (ASD) is one of the fast growing neurodevelopmental disorders across the world. By suitable interventions, their life can be improved a lot. The severity of disease and level of intervention required is different for different persons. Speech given by the ASD people are distorted ones. In this project a distorted Malayalam speech recognition system is developed using Mel-frequency cepstral coefficients(MFCCs) and its derivative features of speech. A Deep neural network is used for recognizing distorted speech. Based on the relevance of words in the life of an autistic patient, level of similarities and difficulties in pronunciation, a distorted Malayalam speech dataset is created. The distorted Malayalam speech recognition system got a Word error rate of 2.8%.