摘要
对带噪语音信号进行增强,是语音信号处理中一个重要的研究课题。由于噪声影响语音质量,这抑制背景噪声,利用小波包良好的时频分析能力,能较好模拟人耳基底膜频率分析特性的特点,提出基于正交小波包的语音去噪增强算法,算法首先把含噪语音信号分解于不同的频率范围内,根据"3σ规则",确定不同频率下的阈值,并采用动态阈值法对各层进行阈值处理,最后对处理后的语音信号反变换得到增强后的信号。在MATLAB平台上,对带噪语音信号去噪增强,实验结果表明,方法在抑制白噪声的同时减小了语音信息的损失,输出语音在保证可懂度的同时达到了较好的输出语音效果。
The enhanced speech signal with noise is an important topic in the speech signal processing.On the base of characteristics of excellent time-frequency analysis ability of wavelet packlet,the frequency analysis characteristics of human ears can be well simulated.In this paper,a new algorithm is proposed for speech de-noise and enhancement using orthogonal wavelet packet decomposition.In this algorithm,the speech is first decomposed into multi-level wavlet,then,to estimate noise variance using different level noise mean,different level threshold was obtained according to 3σ rule.Finally,the de-noise and enhanced speeches were obtained via the inverse wavelet packet transform.Based on MATLAB,for the speech signal with noise was denoised and enhanced.The experments show that the performance is better in subjective test and output SNR test.This method restrains the white noise and reduces the loss of speech with high output SNR and speech intelligibility.
出处
《计算机仿真》
CSCD
北大核心
2011年第5期388-390,共3页
Computer Simulation
关键词
语音去噪增强
小波包
动态阈值法
Speech de-noise and enhancement
Wavelet packet
Dynamic threshold method
