Research On Voiceprint Recognition System Based On GMM

The voiceprint recognition,born under the background of information crisis,is a kind of biological authentication technology. It developed fast and drewworldwide attention with its unique convenience, accuracy and economy.However, in recently years, the research on voiceprint recognition has entered abottleneck period. Because of the limitation of the feature parameter and therecognition model, the recognition rate is hard to be improved. Therefore, Aimingat improving the recognition rate, the paper launched a deeply study on voiceprintrecognition.At the beginning, the paper elaborated the structure and the basic principle ofthe voiceprint recognition system as a whole, then it launched a detailed study oneach module of the system，such as feature parameter extraction, the pretreatmenttechnology,Gaussian mixture model and so on. In the feature parameterextraction module, first, introduced the basic theory of linear prediction analysis,From which the LPCC was derived. Then starting from fundamentals ofHomomorphism filtering, we got the MFCC. Lastly, this two kinds of featureparameters were improved and optimized.In terms of pattern matching, theextracted feature parameters were used,and the model initiation was madethrough K-means algorithm. Then training the model using the EM algorithm.Finally, got the Gaussian mixture model of the speaker recognition. In the lastpart of the paper, we integrated all of the modules and accomplished the overalldesign of voiceprint recognition system based on GMM, filled the relativefunction test experiment.The main innovation of this paper is that an improved endpoint detection wasproposed. The new algorithm had the background noise rid more thoroughly.Thus, we can get the more accurate and effective voice section, which after feature extraction,can reflect the speaker’s personal traits. Experiments show thatthe improved algorithm make the recognition rate increased by2%.