Speaker verification using pitch and melspec information

Recently, the research in the biometrics field acquired a great interest due to the growing needs on increasing the security, restricting the access to some exclusive places. This can be achieved by developing systems that not only recognize the keyword that the speaker tells, but also recognizes who is the speaker. This paper proposes a speaker verification system designed to verify the speaker identity through this/her voice. The system consists of two stages: the feature extraction stage, using pitch and Melspec information, and the speaker verification stage, using backpropagation neural network. In the feature extraction stage pitch information is used to obtain information about the speaker identity. Subsequently the Melspec information is obtained by using FFT and Gabor filtering. Computer simulation results show a fairly good performance with about 95% of recognition rate.