Hybrid MFCC–HHT modeling for dysarthria and severity-level classification

Life Cycle Reliability and Safety Engineering · Springer Nature · 2026-09-16 · DOI: 10.1007/s10772-026-10288-3

Аннотация

Dysarthria is a neuromotor speech disorder that affects speech intelligibility and requires objective assessment methods. Conventional features such as Mel-Frequency Cepstral Coefficients (MFCCs) capture spectral characteristics but are limited in representing nonlinear and non-stationary properties of pathological speech. This study proposes a hybrid MFCC–Hilbert–Huang Transform (HHT) framework that combines spectral and time-frequency features for dysarthria classification. Statistical descriptors extracted from intrinsic mode functions are fused with MFCC features to form a compact representation, which is evaluated using Support Vector Machine (SVM) and Random Forest (RF) classifiers on UA-Speech and TORGO datasets under a speaker-independent protocol. For multiclass classification, the hybrid features achieve accuracies of 95.0% (RF) and 68.1% (SVM) on UA-Speech, and 94.4% (RF) and 74.0% (SVM) on TORGO, while for binary classification the system achieves 98.0% (RF) and 95.1% (SVM) on UA-Speech, and 95.4% (RF) and 93.2% (SVM) on TORGO. The results demonstrate that combining MFCC and HHT features improves classification performance, with Random Forest consistently outperforming SVM, providing an effective and computationally efficient solution for dysarthria detection and multiclass speech disorder classification.

Читать у издателя →

Похожие статьи