一种新的自适应语音增强系统

针对自适应噪声对消(ANC)语音增强系统的性能高度依赖于参考信号的质量，任何原始语音信号泄漏到参考信号中，都会导致原始语音信号失真和噪声抵消性能恶化这一问题，本文提出一种对泄漏不敏感的附加随机噪声(ARN)自适应噪声对消语音增强系统。它通过在参考信号中加入一个低功率的宽带随机训练信号，然后用该训练信号作参考信号对噪声传输函数(NTF)进行自适应建模，并在使用自适应预测滤波器(APF)消除NTF自适应建模的语音信号干扰的同时，用补偿滤波器(CPF)来修正由APF引起的参考信号失真。计算机仿真表明，这种ARNANC语音增强系统在泄漏情况下能将原始语音信号从带噪语音信号中有效分离出来。

A Novel Adaptive Speech Enhancement System

The performance of speech enhancement systems based on adaptive noise cancellation (ANC) is highly dependent on the quality of the reference noise. In the LMS algorithm, the leaking of primary speech signal into the noise reference signal leads to the primary speech signal distortion and poor noise cancellation. Accordingly, a novel adaptive speech enhancement system with adding random noise (ARNANC) is proposed that is much less sensitive to the leakage mentioned above. The ARNANC speech enhancement system is accomplished by adding a low-level, broadband random training signal to the noise reference signal, and adaptively modeling the transfer function of the noise (NTF) by taking the training signal as the reference signal. The ARNANC speech enhancement system uses an adaptive prediction filter (APF) to remove the speech signal embedded in the noisy speech signal that affects the convergence of the modeling filter, and uses a compensation filter (CPF) to modify the distortion of the training signal component due to the APF. Computer simulations demonstrate that the ARNANC speech enhancement system can effectively separate the primary speech signal from the noisy speech no matter under the leakage or leakage-free conditions.