基于最小二乘支持向量机的含噪语音识别算法  

Noisy Speech Recognition Algorithm Based on Least Squares Support Vector Machine

作　　者：熊卫华[1] 梁坤[1] 

机构地区：[1]浙江理工大学机械与自动控制学院,浙江杭州310018

出　　处：《工业控制计算机》2018年第10期86-88,共3页Industrial Control Computer

基　　金：国家自然科学基金项目(61503341)

摘　　要：针对目前含噪语音识别准确率的问题,提出了一种基于集总经验模态分解(EEMD)和最小二乘支持向量机(LSSVM)的语音识别算法。首先将语音信号经过EEMD分解为特征模态函数(IMF)分量,然后经过筛选IMF分量实现去噪,再从剩下的IMF分量中提取美尔倒谱系数(MFCC)作为特征参数,利用动态时间规整法规整MFCC的特征帧,最后将MFCC导入LS-SVM进行训练,从而识别各种语音信号。仿真实验结果表明,该方法能够快速有效地识别各种语音信号,与EEMD结合BP神经网络方法相比,识别准确率更高,抗干扰能力更强。In light of the accuracy of noisy speech recognition,a speech recognition algorithm based on ensemble empirical mode decomposition（EEMD） and least squares support vector machine（LS-SVM） are proposed in this paper.Firstly,this method decomposes the speech signal through EEMD to obtain the intrinsic mode function（IMF） components,and then filters them to achieve denoising,and the Mel-Frequency cepstrum coefficient（MFCC） is extracted from the remaining IMF components.Next,the feature frames of MFCC are regulated through the dynamic time warping.Finally,the MFCC is imported into the LS-SVM for training,so as to recognize various speech signals.

关 键 词：语音识别 集总经验模态分解 最小二乘支持向量机 美尔倒谱系数 频谱分析 

分 类 号：TN912.34[电子电信—通信与信息系统]