Classification of Right Hemisphere Damage Using MFCC Paralinguistic Voice Features

Right Hemisphere Damage (RHD) can profoundly impact the melodic and rhythmic qualities of speech—collectively known as prosody—and lead to social communication difficulties. This study proposes a machine learning workflow to identify RHD-related paralinguistic deficits from single-source audio. Audio signals from the NIH-funded TalkBank RHD dataset were denoised, voice activity filtered, and normalized, and Mel Frequency Cepstral Coefficients were extracted. Multiple supervised learning algorithms (Support Vector Machine, Random Forest, Logistic Regression, and k-Nearest Neighbors) were then trained separately and evaluated. Among these models, the Support Vector Machine with a radial basis function kernel achieved the highest accuracy of 0.79, indicating that short, standardized speech samples contain diagnostic cues for RHD. The contribution of this study is that this is the first application of such methods on the dataset for the objective. This approach highlights the feasibility of automated, non-invasive detection of RHD and offers a promising direction for adjunctive clinical assessment tools.