Gender Recognition by Using Acoustic Features of Sound with Deep Learning and Data Mining Methods
Eyup Ensar Kalayci, Buket Doğan · 2020 Innovations in Intelligent Systems and Applications Conference (ASYU) · 2020
Human voice is a physical phenomenon that contains some information on it. This information has been examined in studies on sound for years. Studies such as recognition of speech, age and gender determination, and emotional analysis were carried out and successful results were obtained. In this study, gender was determined according to the acoustic characteristics of the speeches compiled from seven different languages by using the acoustic features of the voice. In addition to data mining techniques, deep learning was also used in the classification process. Our dataset consisting of 4739 different speech samples belonging to approximately 800 people. 95.59% and %92.15 best success rate was achieved respectively with gradient boosting and random forest in classification process. Deep learning method with Multilayer Perceptron (MLP) was achieved 96.22% correct classification rate in our analysis.