A Comparison of Cepstral Features in the Detection of Pathological Voices by Varying the Input and Filterbank of the Cepstrum Computation

Automatic voice pathology detection enables objective assessment of pathologies that affect the production mechanism. Detection systems have been developed using traditional pipeline approach (consisting feature extraction part and part) modern deep learning -based end-to-end approach. Due to lack vast amounts training data in study area pathological voice, former is still a valid choice. In existing based on approach, mel-frequency cepstral coefficient (MFCC) features can be regarded as defacto standard set. this study, automatic investigated by comparing performance various MFCC variants derived considering two factors: input filterbank cepstrum computation. For first factor, three inputs (the signal, glottal source vocal tract) are compared. The tract estimated quasi-closed phase inverse filtering method. second linear-frequency filterbanks Experiments were conducted separately six databases consisting voices produced speakers suffering from one four disorders (dysphonia, Parkinson’s disease, laryngitis, or heart failure) healthy speakers. Support vector machine (SVM) was used classifier. results show combining mel- coefficients tract, better overall accuracy obtained compared signal. Furthermore, combination provided comparable than techniques clean high signal-to-noise ratio (SNR) conditions.