一种基于Chirp原子分解的语音增强方法  

A Method of Speech Enhancement Based on Chirp Atoms Decomposition

作　　者：武明勤[1] 于凤芹[1] 韩鹍[1] 

机构地区：[1]江南大学通信与控制学院,江苏无锡214122

出　　处：《微电子学与计算机》2005年第12期74-77,共4页Microelectronics & Computer

基　　金：江苏省自然科学基金项目(BK2002068)

摘　　要：由于语音信号在时频面上具有局部连续结构,文章提出了一种基于Chirp时频原子分解的语音增强方法,该方法将含噪的语音信号使用匹配追踪算法分解成Chirp原子的组合,根据语音和噪声所对应的Chirp原子在参数上的不同,从中分离出属于语音的Chirp原子来重构语音信号,从而达到去除增强语音的目的。仿真实验结果表明,经该法处理后的语音信号的信噪比有较大的提高,主观试听效果也较好。According to the coherent structure of speech on the time-frequency plane, we proposed a speech enhancement algorithm . This algorithm first decompose the noisy speech signal into Chirp atoms using Matching pursuit, and then select the Chirp atoms according to the characters of speech on its time-frequency plane. Finally, noise is removed by reconstructing the selected Chirp atoms. Simulation to the noisy speech have shown that this algorithm works efficiently even when the signal to noise ratio （SNR） is very slow, and the enhanced speech has good audible quality.

关 键 词：语音增强 匹配追踪算法 Chirp原子分解 时频分布 

分 类 号：TN912.3[电子电信—通信与信息系统]