Technology Of Model Training In Speaker Recognition Based On Adaptation And MCE

Due to its unique advantages of flexibility, economy, accuracy, extensibility, and so on, speaker recognition technique has a broad application future in biometrics security field. Although speaker recognition system runs well in the laboratory, its performance descents rapidly because of the influence of various factors in the real world. In order to make the system be more practical, many problems should be solved. One of the most significant problems is how to improve the system performance for lack of training data.This thesis adopts GMM-UBM when model speaker recognition system considering of lacking data. In the aspect of adapting in speaker recognition system modeling and parameter estimating, attentions are put on researching in how to improve recognition rate. In the side of adapting in speaker recognition system modeling, we will ameliorate conventional MAP (Maximum A Posterior Probability) means to get speaker recognition model, apply MLLR (Maximum Likelihood Linear Regression) and EigenVoice adaptation ways which used in speech recognition into adapting in speaker recognition system modeling, and compare the results with MAP mean. Considering the advantage and dis advantage of MAP and MLLR, integration incremental adaptation method could be introduced. In the side of estimating parameter, there is some disadvantages in the commonly used algorithm that base on ML(Maximum Likelihood) , because it could not depicts the difference between speakers clearly. Thus MCE (Minimum Classification Error) that used in speech recognition these years could be introduced, this method aims at minimum classification error in discriminating training. By means of analyzing, there is great importance when we apply MCE in improving speaker recognition system.The experiments show that if we use the appropriate means to adapt speaker recognition system model, and then improve the system with parameter estimation based on MCE, the recognition rate of the system will advance remarkable, we will obtain the best correct rate 91.2%.