一种声纹美尔频率倒谱系数干扰消除算法研究

摘要

在伴随着外部噪声的情况下,待识别的声纹美尔频率倒谱系数特征各项属性很容易受到外部噪声的干扰发生改变,造成声纹特征的识别的精度不高.为提高精度,提出了一种用支持向量机的美尔频率倒谱系数特征干扰去除算法.确定分类决策函数时充分考虑美尔频率倒谱系数与声纹中心以及噪声之间的关系,并且将声纹特征引入核函数,将原空间样本数据通过非线性变换映射到高维特征空间,在高维空间中求最优或广义最优分类面,实现对语音特征的干扰消除.实验表明,利用改进算法实现了声纹特征中过零率,倒谱特征、矩形窗和汉明窗长的短时能量函数特征的优化.%With external noise, identification of voiceprint meyer frequency cepstrum coefficient features various attributes is very vulnerable to external noise interference change, causing low voiceprint characteristics of the recognition accuracy. For this, based on support vector machine (SVM), a Mel frequency cepstrum coefficient characteristics interference removal algorithm was proposed. The algorithm in determining the classification decision function fully considers Mel frequency cepstrum coefficient and the relationship between the voice print center and the relationship between the noise, and voiceprint features are introduced into the kernel function, and the space sample data through the nonlinear mapping transformation into a high dimensional feature space. In the high dimension space and optimal or generalized optimal classification surface, the speech characteristics of the interference elimination are realized. The experimental results show that the use of this kind of algorithm realyzes the optimization of zero - crossing rate of voiceprint, the spectrum characteristics, the rectangular window and hamming window long short-term energy function characteristics.