The Design Of A Speaker Recognition Software Based On Matlab

Due to the uniqueness and induplication of biosignatures such as voice,finger print and iris,they have become the main methods in the identity identification.Compared with the other biosignatures,voice print is more easily available,with the characteristics of long-distance transmission and not easily affected by the environment.It presents its own uniqueness in the identity identification aspect and has a broad application prospect.This thesis designs a speaker recognition software based on Matlab.The software can record speakers' voices and store them in the designed voice database.A visual graphic interface is also designed,which is operation-friendly,simple and highly controllable.In the voice recognition aspect,the software can sample and analyze the speaker's voice and find the right person from the voice database.Three voice recognition algorithms are accomplished in the software,which are MFCC characteristic parameters,LPCC characteristic parameters and GMM model parameters.The MFCC characteristic parameter method is based on human auditory model to calculate the Mel-frequencies and cepstral coefficients,and the voice recognition is conducted through visual quantization.The LPCC characteristic parameter method uses human speech model to calculate AR coefficients,and the voice recognition is conducted through calculating the Euclidian distance.The GMM model parameter method summarizes human voice as an integration of several Gaussian distributions.The characteristics of the voice can be specified as Gaussian parameters.The voice recognition is conducted through Kullback-Leibler distance.The three algorithms are compared through several experimental tests.The recognition results indicate that the MFCC method presents the highest recognition rate and efficiency.