Adaptive Feature based Explainable Ensemble Model for Dysarthria Speech Detection using OpenSMILE Acoustic Features and Multi-Level Feature Selection

Speech-based effective communication is essential to human contact and has a significant impact on an individual’s quality of life. Dysarthria is a motor speech disorder that impedes natural expression and intelligibility by altering voice acoustics. However the conventional system often face the challenges due to noisy and redundant features. To address this challenge, the study proposed the Adaptive feature based Explainable ensemble mod-el for the automated detection of dysarthria. Subsequently many acoustic features of TORGO and UA-Speech data are extracted with the core feature extraction of module of OpenSMILE utilized in this study. The extracted features are selected with the Adaptive feature selection process to chosen the most informative features. The selected features are classified with the ensemble learning stacking model of Random forest, XG Boost and SVM. Finally the core model interprets with the global and local levels with hybrid developed model of SHAP and LIME.