Glottal features for classification of phonation type from speech and neck surface accelerometer signals

Glottal source characteristics vary between phonation types due to the tension of laryngeal muscles with respiratory effort. Previous studies in classification type have mainly used speech signals recorded by microphone. Recently, two were published using neck surface accelerometer (NSA) signals. However, there are no previous comparing use acoustic signal vs. NSA as input classifying type. Therefore, current study investigates simultaneously and three (breathy, modal, pressed). The general goal is understand which (speech NSA) more effective task. We hypothesize that same feature set for both signals, accuracy higher signal, closely related physical vibration vocal folds less affected tract compared acoustical signal. waveforms computed processing methods, quasi-closed phase (QCP) glottal inverse filtering zero frequency (ZFF), a group time-domain frequency-domain scalar features from obtained waveforms. In addition, investigated mel-frequency cepstral coefficients (MFCCs) derived QCP ZFF. Classification experiments support vector machine classifiers revealed showed better discrimination when was used. Furthermore, it observed complementary information conventional MFCC resulting best (86.9%) (80.6%).