Automatic speech-based classification of gender, age and accent

Abstract

This paper presents an automatic speech-based classification scheme to classify speaker characteristics. In the training phase, speech data are grouped into speaker groups according to speakers' gender, age and accent. Voice features are then extracted to feature vectors which are used to train speaker characteristic models with different techniques which are Vector Quantization, Gaussian Mixture Model and Support Vector Machine. Fusion of classification results from those groups is then performed to obtain final classification results for each characteristic. The Australian National Database of Spoken Language (ANDOSL) corpus was used for evaluation of gender, age and accent classification. Experiments showed high performance for the proposed classification scheme. © 2010 Springer-Verlag Berlin Heidelberg.