Dissertation
Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Electro-acoustic technology and speech signal processing > Speech Signal Processing > Speech Recognition and equipment

Research of Speaker Recognition Based on the Improvement Feature Parameters

Author LiuMingJuan
Tutor ZhangZuo
School Northeastern University
Course Basic mathematics
Keywords speaker recognition feature extraction principal component analysis pitch cycle gaussian mixture model
CLC TN912.34
Type Master's thesis
Year 2010
Downloads 46
Quotes 0
Download Dissertation

In this paper a small text-independent speaker recognition are researched. The main research work and achievements are as following:In the pretreatment part, in order to guarantee the rate of detection, we adopt an endpoint detection method based on the short energy and short-time zero-crossing rate of the speech.In the feature extraction part, we elaborate the feature parameters which are often used for the speaker recognition, such as LPC, LPCC, MFCC and so on. We use the principal component analysis method to improve the feature parameters performance based on the mel frequency cepstral coefficients (MFCC), and then combineding with the pitch feature as the new feature parameters. In addition, the performance of the new parameters are analysed, the analysis results show that the capacity of characting the speaker is stronger than MFCC.In the recognition part, the gaussian mixture model(GMM) are mainly introduced. In this paper, a speech corpus used for speaker recognition is established and we use the new feature parameters in the GMM, the results of the simulation experiment indicates the recognition of the new characteristic parameters are higher than MFCC.The all algorithms of this paper could be achieved by programming. Simulation experiment indicates the effect of our algorithms.

Related Dissertations
More Dissertations