Gender identification based on time-varying multi-instance learning
Ming Liang Gu · Jisuanji gongcheng yu sheji · 2013
A method for gender identification based on Time-varying Multi-Instance Learning is presented to improve the recognition rate.The method regards the speech regions as multi-instance bag,and the acoustic features of the speech regions are clustered into instances of the bag using K-means clustering algorithm.After labeling male and female speech bags into different kinds,EM-DD algorithm is used to get male and female speech diverse density points,and Bags-K neighbor classification algorithm is put forward for identification.Experiments show the average recognition rate of the gender identification system could reach as high as 97%.