基于EMD与RLS的说话人识别抗噪算法

A SPEAKER RECOGNITION ANTI-NOISE ALGORITHM BASED ON EMD AND RLS

语音信号是一种典型的非平稳信号。为了分析语音的非平稳特性,提出一种基于经验模态分解(EMD)与递推最小二乘算法(RLS)自适应滤波器相结合的语音信号去噪预处理器,进一步提高语音信号的信噪比和可懂度。再通过Gammatone滤波器组,对去噪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。仿真实验在高斯混合模型识别系统中进行。实验结果表明,采用这种方法应用于说话人识别抗噪系统,系统的识别率及鲁棒性都有明显提高。

Speech signal is a kind of typical non-stationary signal. In order to analyse the non-stationary characteristic of speech signal,in the paper we present a speech signal denoising pre-processor,which is based on the combination of empirical mode decomposition（ EMD）and recursive least-squares（ RLS） adaptive filter,and further improves SNR and speech intelligibility of signals. Then through Gammatone filter bank it deals with the denoised speech signals,and extracts the feature parameters GFCC of speaker speech signals. We conduct the simulation experiment in Gaussian mixture model recognition system. Experimental results show that applying the algorithm in speaker recognition anti-noise system,the recognition rate and robustness of the system are all obviously improved.