Automatic Speech Recognition and Assessment Systems Incorporated into Digital Therapeutics for Children with Autism Spectrum Disorder

Children with autism spectrum disorder (ASD) frequently encounter challenges in social communication and interaction, which necessitates continuous, comprehensive interventions to enhance their communication skills. Despite increasing interest in digital therapeutics (DTx), research on speech-utilizing interventions for children with ASD remains limited. This study introduced speech-based technologies integrated into DTx software designed to support the development of communicative skills in children with ASD. We compiled a large speech corpus from both children with ASD and typically developing children, which included clinical scores on social communication severity and speech production, rated by certified speech and language pathologists. Then three speech-based technologies were developed: automatic speech recognition for verbal interaction within the DTx, an automatic assessment model for social communication severity to monitor progress, and an automatic speech production assessment model to facilitate speech production skills. The results were promising, demonstrating a syllable error rate of 12.36% in automatic speech recognition for target keywords, a correlation coefficient of 0.71 for assessing social communication severity, and a correlation coefficient of 0.75 for speech production assessment. These technologies are expected to improve the accessibility of interventions for children with ASD, overcoming barriers related to location, time, and human resources.