语音信号准确性识别仿真研究

摘要

This article proposes a recognition method for speech signal accuracy based on modified complex wavelet.Firstly,the method decomposes speech signals with noise according to wavelet transform fundamental.It acquires wavelet entropy of signal sub-band section in different decomposition orders and calculates threshold of high frequency coefficient.Then,it builds threshold function of compromise index to eliminate noise contained by the high frequency coefficients in speech signal of grammar integrated with then calculation results.It also workes out pitch period and formant information of signal of language pronunciation integrated with the complex wavelet theory.Finally,it uses optimal complex Gauss wavelet to extract magnitude spectra and phase spectra of signal and completes recognition on accuracy of speech signal.Simulation results show that the method can reduce signal redundancy effectively.It can recognize the signal accuracy effectively.%对语音信号的准确性进行有效识别,可提升语音发音准确性.但是语音信号中包含大量冗余度很高的随机信号,需要对随机信号进行过滤,降低降低信号的冗余度.但是传统方法利用小波变换对语音信号进行适当的尺度分解,得到各尺度的频谱并进行DCT变换不同语音信号的特征参数完成识别,但是无法排除随机信号的干扰,存在语音信号准确性识别误差大的问题.提出一种改进复子波的语音信号准确性识别方法.依据小波变换基本原理对含噪声的语音信号进行分解,获取信号不同分解层次的子带区间小波熵,并对信号不同层次中高频系数阈值进行计算,构建折中指数阈值函数,与计算结果相结合去除语法语音信号中高频系数所含的噪声,融合于复子波分析原理,计算出语言发音信号的基音周期及共振峰信息,利用最优的复高斯子波提取语音信号的幅度谱和相位谱,并完成对语音信号准确性识别.仿真结果表明,所提方法可以有效地降低信号冗余度,可以对语音信号的准确性有效识别.