System Development Of Speaker Gender Identification And Age Estimation

The gender and age of the speaker is important to represent the identity of the person. It has been applied increasingly widely in the real world, such as criminal investigation, bank financial security and military fields, etc.. In daily life, people can easily judge a person’s gender and age according to the pitch, speed and content of the speaker’s voice, so the human-computer interaction system can also estimate the speaker’s gender and age according to his or her voice.This paper complete the gender identification and age estimation of the speaker based on their different features. In the process of speaker gender identification and age estimation system(hereinafter referred to as system), the main contents are as follows:1. Speech samples at different ages and different gender are recorded to establish three types of database. The first one is the gender recognition database, both female and male model are trained; the second one is the age estimation database, which divided all the samples into five intervals according to the different acoustic features; the third one is a new type I proposed referring to the former researches, which can realize gender classification and age estimation simultaneously, the method is based on gender classification and divided samples by age. The experiments can identify the gender, age, gender and age at the same time.2. A feature parameters, which mixes the short-time average magnitude, MFCC, and first-order differential MFCC is proposed. Moreover, it is compared with other methods based on the GMM model. According to three experiments we complete, the proposed feature parameters perform well in whichever experiment.3. A gender identification and age estimation system is designed. In this system, people can carry out the above three types of model training and recognition by changing the database. MFC is utilized in the system as the interface development language to design function modules.4. The factors affecting the accuracy of system identification are analyzed. The functions of both online and offline identification system are realized. Among them, for the gender recognition, the online average recognition rate is 91.95%, while the offline average recognition rate is 92.65%; for the age estimation, while the online average recognition rate is 75.34%, offline average recognition rate is 77.14%; for the simultaneous gender classification and age estimates, the online average recognition rate is 83.67%, while the offline average recognition rate is 83.99%.according the result, we prove that the method we propose can effectively improve the recognition accuracy of the age estimation.