基于多统计模型和人耳听觉特性的麦克风阵列后滤波语音增强算法

Microphone Array Post-filter Based on Multi-statistical Models and Perceptual Properties of Human Ears

针对麦克风阵列后滤波语音增强算法的不足,结合人耳的听觉掩蔽效应,提出了改进的后滤波语音增强算法.提出了最大化目标语音存在概率来确定信号子空间维度的方法,在噪声子空间上,利用条件概率估计出噪声功率谱.基于人耳的听觉掩蔽效应,提出了后滤波器的一种合理的设计方法.实验证明,所提的噪声估计方法比传统方法更加准确,所提的后滤波算法比传统的后滤波算法更好,在多项语音评价指标上,都取得了更好的实验效果.

To overcome the drawbacks of the conventional microphone array post-filter speech enhancement method, some improvements are proposed using the masking properties of human ears. A subspace selection method is proposed by maximizing the present probability of the target speech. In the noise subspace, the conditional probability is used to estimate the noise power spectrum. A novel post-filter is proposed based on the masking properties of human ears. Experiments prove that the proposed noise estimation method and post-filter are much better than the conventional ones. The proposed speech enhancement technique has shown to produce impressive results in terms of quality measures of the enhanced speech.