Identification of Autism Spectrum Disorder Markers in Spoken Conversations

Abstract

Autism spectrum disorder (ASD) is a disability that impacts the social behavior of a person. Diagnosis of ASD is a challenging task, as there is a spectrum of symptoms that can vary from person to person. One of the areas, which affect a person with autism is spoken conversation. This report focuses on recognizing autism markers in spoken conversation. Firstly the key symptoms with respect to spoken conversation will be discussed. To find a pattern in spoken conversation the audio needs to be digitized in form of text and with speaker identity. This report discusses various state-of-the-art machine learning models for speech-to-text translation or automatic speech recognition (ASR) and then the speaker diarization process. Autism detection videos can be very long as it’s a long process so time complexity will also be determined for ASR and speaker diarization process. For pattern recognition, various metrics from the speech will be calculated. Lastly, a conclusion will be made and the future scope of this project will be discussed.