DETECTION OF STAMMERING IN SPEECH USING MACHINE LEARNING WITH HUMAN EVALUATION

Abstract: Stammering, also known as stuttering, is a speech fluency disorder characterized by involuntary repetitions, prolongations, and blocks in speech production. This paper presents a comprehensive framework for automatic detection of stammering using machine learning, augmented with structured human evaluation. Feature extraction uses MFCC, pitch contour, zero-crossing rate, and nergy-based features. Multiple classifiers including SVM, Random Forest, LSTM, and CNN are trained and evaluated. A human evaluation protocol validated model predictions against speech-language pathologists (SLPs). The proposed hybrid LSTM+RF approach achieves 94.7% accuracy with an F1-score of 0.943, outperforming existing standalone methods. Human-Model Agreement of 91.5% with Cohen's Kappa k=0.83 confirms clinical reliability.

Keywords: Stammering Detection, Speech Processing, Machine Learning, SVM, CNN, Human Evaluation, Speech- Language Pathology.