Data Augmentation and Deep Learning Methods in Sound Classification: A Systematic Review

The aim of this systematic literature review (SLR) is to identify and critically evaluate current research advancements with respect small data the use augmentation methods increase amount available for deep learning classifiers sound (including voice, speech, related audio signals) classification. Methodology: This SLR was carried out based on standard guidelines PRISMA, three bibliographic databases were examined, namely, Web Science, SCOPUS, IEEE Xplore. Findings. initial search findings using variety keyword combinations in last five years (2017–2021) resulted a total 131 papers. To select relevant articles that are within scope study, we adopted some screening exclusion criteria snowballing (forward backward snowballing) which 56 selected articles. Originality: Shortcomings previous studies include lack sufficient data, weakly labelled unbalanced datasets, noisy poor representations features, effective approach affecting overall performance classifiers, discuss article. Following analysis identified articles, overview feature extraction methods, techniques, its applications different areas classification problem. Finally, conclude summary SLR, answers questions, recommendations task.