论文部分内容阅读
本文提出了一种基于MFCC(MelFrequencyCepstralCoefficients)和LSP(LineSpectrumPair)的混合语音特征参数,有效地弥补了单纯使用MFCC的不足,实验中利用加权的欧几里德距离计算特征矢量的失真距离,结果表明新的特征矢量能够很好的表征语音信号的特征信息,能有效降低系统的误识率。
In this paper, a mixed speech feature parameter based on MFCC (Mel Frequency Cepstral Coefficients) and LSP (LineSpectrumPair) is proposed to effectively compensate for the lack of MFCC. In the experiment, the weighted Euclidean distance is used to calculate the distortion distance of eigenvectors. The new feature vector can characterize the characteristic information of the speech signal well, which can effectively reduce the error rate of the system.