一种低信噪比下的说话人识别算法研究

摘要

In order to improve the performance of speaker recognition system under low SNR, we present a speaker recognition algorithm which combines Gammatone filter bank and the speech enhancement with the spectral subtraction improved.It uses improved spectral subtractionas the preprocessor to further improve SNR of speech signals, and then deals with the enhanced speech signals of speaker through Gammatone filter bank, extracts feature parameters GFCC of speaker’ s speech signals, and applies feature parameters GFCC in speaker recognitionalgorithm.The simulation experiment is conducted in Gaussian mixture model recognition system.Experimental results show that by applying the algorithm in speaker recognition system, the system recognition rate and robustness are all obviously improved.%为了提高低信噪比下说话人识别系统的性能，提出一种Gammatone滤波器组与改进谱减法的语音增强相结合的说话人识别算法。将改进的谱减法作为预处理器，进一步提高语音信号的信噪比，再通过Gammatone滤波器组，对增强后的说话人语音信号进行处理，提取说话人语音信号的特征参数GFCC，进而将特征参数GFCC用于说话人识别算法中。仿真实验在高斯混合模型识别系统中进行。实验结果表明，采用这种算法应用于说话人识别系统，系统的识别率及鲁棒性都有明显的提高。