【信号处理】Matlab实现CDR-based噪声和混响抑制

1 内容介绍

We investigate the estimation of the time- and frequency-dependent coherent-to-diffuse ratio (CDR) from the measured spatial coherence between two omnidirectional microphones. We illustrate the relationship between several known CDR estimators using a geometric interpretation in the complex plane, discuss the problem of estimator bias, and propose unbiased versions of the estimators. Furthermore, we show that knowledge of either the direction of arrival (DOA) of the target source or the coherence of the noise field is sufficient for an unbiased CDR estimation. Finally, we apply the CDR estimators to the problem of dereverberation, using automatic speech recognition word error rate as objective performance measure.

2 部分代码

%ESTIMATE_CDR_ROBUST_UNBIASED

% Unbiased estimation of the Coherent-to-Diffuse Ratio (CDR) from the complex

% coherence of a mixed (noisy) signal, using knowledge of both signal and noise

% coherence. This is a variation of estimate_cdr_unbiased which shows better

% performance in practice. Equivalent to CDRprop2 in [1].

%

% CDR = estimate_cdr_nodiffuse(X, N, S)

%       X: complex coherence of mixed (noisy) signal

%       N: coherence of noise component (real-valued)

%       S: coherence of signal component (magnitude one)

%

% Reference:

% Andreas Schwarz, Walter Kellermann, "Coherent-to-Diffuse Power Ratio

% Estimation for Dereverberation", IEEE/ACM Trans. on Audio, Speech and

% Lang. Proc., 2015 (under review); preprint available: arXiv:1502.03784

% PDF: http://arxiv.org/pdf/1502.03784

%

% Andreas Schwarz (schwarz@lnt.de)

% Multimedia Communications and Signal Processing

% Friedrich-Alexander-Universitaet Erlangen-Nuernberg (FAU)

% Cauerstr. 7, 91058 Erlangen, Germany

function CDR = estimate_cdr_robust_unbiased(Cxx,Cnn,Css)

Css = bsxfun(@times, ones(size(Cxx)), Css);

Cnn = bsxfun(@times, ones(size(Cxx)), Cnn);

% limit the magnitude of Cxx to prevent numerical problems

magnitude_threshold = 1-1e-10;

critical = abs(Cxx)>magnitude_threshold;

Cxx(critical) = magnitude_threshold .* Cxx(critical) ./ abs(Cxx(critical));

CDR = 1./(-abs(Cnn-exp(1j*angle(Css)))./(Cnn.*cos(angle(Css))-1)).*abs((exp(-1j*angle(Css)).*Cnn - (exp(-1i*angle(Css)).*Cxx))./(real(exp(-1i*angle(Css)).*Cxx) - 1));

% Ensure we don't get any negative or complex results due to numerical effects

CDR = max(real(CDR),0);

end

3 运行结果

4 参考文献

[1] Schwarz, A. , and  W. Kellermann . "Unbiased coherent-to-diffuse ratio estimation for dereverberation." International Workshop on Acoustic Signal Enhancement IEEE, 2014.

博主简介：擅长智能优化算法、神经网络预测、信号处理、元胞自动机、图像处理、路径规划、无人机、雷达通信、无线传感器等多种领域的Matlab仿真，相关matlab代码问题可私信交流。

部分理论引用网络文献，若有侵权联系博主删除。