| In recent years, the study to the speaker recognition technology has become a hot topic at home and abroad, while the speaker recognition system of uyghur has been initial. With the popularity of telephone and mobile phone, telephone voice plays an important role in daily life, and the use of telephone voice for speaker recognition has been convenient, Therefore, it has great value and practical significance to bulid the speaker recognition system of uyghur. In this paper, we combines the characteristics and rules of uyghur pronunciation to achieve a practical application-oriented speaker recognition system , mainly to do the following work:(1)According to the design ruler of uyghur speech database, we established 278 phone free conversation speech database of uyghur.(2)For the case of using only channel cepstrum parameters of support vector machine (SVM) speaker model training and low efficiency in large samples, this paper proposes a combination of statistical parameters of the GMM-UBM speaker model, the Target speaker and the background speaker's Gaussian mixture model (GMM) were extracted, then build statistical models and train SVM speaker model. It is very effective to solve the problem of the large sample for the SVM training.(3)We researched and discussed the influness of the incentive characteristics and the dynamic parameters carried by the speech signal to how to recognition performance, introduced a main-auxiliary system policy which take a source of information as the auxiliary system, the main system used GMM-UBM mode and the auxiliary system used SVM mode to further improve the system recognition rate.(4) On the platform of the embedded Linux, we achieved the uyghur speaker recognition system written in language Perl and C language by taking the uyghur as object .(5) When an effective target voice was set in 20 seconds and test speech was set in 10 seconds in the training set of 50 targets, It would got the best recognition accuracy - 94.2%. |
PDF Full Text Request