STUDS, Speech Therapy Utility for Detection and Analysis of Stuttering

Abstract: Stuttering, a complex speech disorder, presents significant challenges in both diagnosis and treatment. In this study, we propose a novel approach for predicting stuttering severity in Kannada speech, aimed at enhancing therapeutic interventions for individuals affected by stuttering. Leveraging a dataset comprising video recordings of therapy sessions, our methodology involves the extraction of acoustic features from 3-second audio segments, including mean pitch, intensity, speech rate, and MFCCs. These features, along with annotations for disfluency types such as prolongation, repetition, and block, form the basis of a comprehensive dataset. Through the application of a CNN-LSTM hybrid neural network, we demonstrate promising results in predicting stuttering severity, with implications for personalized therapy strategies. Our research underscores the potential of machine learning techniques in improving the diagnosis and treatment of stuttering, paving the way for more effective interventions and improved outcomes for individuals with this speech disorder.

Keywords: MFCCs, CNN-LSTM, Kannada speech, stuttering.