A comparison of data augmentation methods in voice pathology detection

Farhad Javanmardi, Sudarsana Reddy Kadiri, Paavo Alku

Abstract: Highlights•Six data augmentation (DA) methods are compared in detection of pathological voice.•Three time domain, one time-frequency domain and two vocoder-based methods are used.•The DA methods are studied using four popular spectral features and four classifiers.•The SpecAugment DA method shows highest improvements using the CNN-based classifier.