利用相位时频掩蔽的麦克风阵列噪声消除方法

本文提出了一种在干扰声源和背景噪声存在条件下麦克风阵列噪声消除的方法。麦克风阵列通过波束形成增强由导向矢量所指定方向的目标声源来抑制背景噪声。然而，现有的波束形成算法在干扰声源存在的情况下，无法进行准确的导向矢量估计。为此，本文提出一种基于音频信号互相关功率谱相位的麦克风阵列噪声消除方法。首先通过音频信号的相位时频掩码估计导向矢量，并对其进行波束形成，从而有效抑制干扰声源和背景噪声；然后利用语音存在概率，采用最大似然的方法估计波束形成后信号中残留的干扰噪声功率谱密度，对其进行后处理，进一步抑制残留干扰和噪声。实验结果表明在干扰声源和背景噪声存在的条件下，所提方法有效地实现了麦克风阵列噪声消除，且各种性能指标优于基线方法。 

Microphone Array Noise Cancellation Method Using Phase Time-frequency Masking

A noise cancellation method is studied for microphone array under the interfering sound source and background noise. The microphone array suppresses background noise by reinforcing the target sound source in the direction specified by the steering vector by beamforming. However, the present beamforming algorithms cannot accurately estimate the steering vector when interfering sound source presents. Therefore, this paper presents a noise cancellation method using the phase of audio signal cross-correlation power spectrum based on the microphone array. Firstly, the proposed method estimates the steering vector by the phase time-frequency mask of the audio signal and carries out the beamforming, so as to effectively suppress the interference sound source and background noise. Then the speech presence probability is calculated to estimate residual interference and noise power spectral density using maximum likelihood method and employed on the post-beamforming signal to further suppress residual interference and noise. The experimental results demonstrate that under the interfering sound source and background noise, the proposed microphone array method can effectively suppress noise, which is superior to the baseline methods.