说话人识别算法的研究

On the speaker recognition algorithm

针对说话人识别问题,基于概率神经网络PNN,实验比较MFCC,ΔMFCC+MFCC分别与PNN相结合时的识别率。仿真结果表明,在文本有关情况下,当说话人说话内容为0～9的发音时,ΔMFCC+MFCC优于MFCC,使用PNN算法的识别率能够满足说话人识别的实际要求。

Based on PNN,experiments are setup to compare the speaker recognition rates of MFCC and △ MFCC ＋ MFCC.Simulating results show that,for a speaker,in textdependent circumstances,△ MFCC ＋ MFCC is better than MFCC when the content of speech is 0 to 9;and upto the recognition rate,the algorithm based on PNN can meet the practical requirements for speaker recognition.