A machine learning approach for gender identification using statistical features of pitch in speeches. (1st January 2022)

• Record Type: Journal Article
• Authors: Shagi, G.U.; Aji, S.
• Is Part Of: Applied acoustics. Volume 185(2022)
• Publication Date: 2022-01-01
• Subjects: Gender identification -- Pitch -- Statistical measures -- Machine learning; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2021.108392
• Languages: English
• ISSNs: 0003-682X

Abstract

Abstract: Speech processing and recognition are some of the imperative research areas, remarkably in user authentication and entertainment. There are assorted well-known approaches in the domain, and the properties of pitch are broadly used in most of the works. The objective of this work is to explore the capabilities of statistical features of the pitch in the gender identification problem. We proposed an effective combination of features, PFG-Pitch Feature for Gender, for gender identification with machine learning algorithms. The feature sets from three datasets – TIMIT, CHAINS and SLR-63- were used in the experiments. We could attain the maximum output in some ideal situations with the classical learning methods- CNN, MLP, SVM and LR. It is renowned that the outstanding performance from most of the machine learning classifiers embassies the magnitude of statistical features in speech processing.