基于分形理论的语音增强  

Speech Enhancement Based on Fractal Interpolation

作　　者：赵晓群[1] 王彦[2] 苏彦莽[3] 王霞[3] 

机构地区：[1]同济大学电信学院,上海200092 [2]天津职业大学电子信息工程学院,天津300402 [3]河北工业大学信息工程学院,天津300130

出　　处：《电声技术》2005年第12期33-37,共5页Audio Engineering

基　　金：上海市自然科学基金(04ZR14138)

摘　　要：在分析了语音信号与随机噪声在不同尺度上进行小波变换时其传递特性不同的特征表现及其特点之 后,给出了一种新的基于分形维数的语音信号增强方法。该方法根据带噪语音信号及子波重构之后的分形维数不 同对语音信号的小波变换系数作不同的阚值处理,既抑制了噪声,又减少了语音段的信息的损失,提高了信噪比。 仿真结果表明,这是一种有效的语音增强、去噪的方法。After analyzing the transfer characteristic of the speech signal and the random noise on wavelet transform in different measures, in this paper,a novel denoising algorithm based on fractal dimension and wavelet field is presented. The characters and fractal dimension of noise and signal on wavelet transform are discussed. Compared with others,the process of the algorithm has some advantages, such as constructing efficiently,wiping off the noise effectively and programming easily. Simulation proves that the algorithm has better denosing performance.

关 键 词：语音增强 分形维数 小渡 去噪 非线性方法 

分 类 号：TN912[电子电信—通信与信息系统]