考虑到语音信号方向向量估计误差对传统波束形成语音增强性能的影响,该文提出一种盲波束形成语音增强方法。由于采用阵列四阶互累积量和线性约束最小方差波束形成器相结合,使得该方法对语音信号方向向量误差具有一定韧性。此外,采用多通道后置滤波去除盲波束形成器输出端的残留噪声。仿真结果表明,在语音信号波达方向等先验信息未知的情况下,该文提出的盲波束形成语音增强方法仍具有较好的噪声抑制性能。
Considering the bad effect of steering vector errors on speech enhancement performance, a new blind beamforming based speech enhancement method is introduced in this paper. The blind beamforming combines fourth-order cross cumulant and Linear Constrained Minimum Variance (LCMV) beamformer, which makes it robust to the steering vector. A multi-channel post-processor is then used to suppress the residual nonstationary noise. Simulations results demonstrate that the proposed method exhibits good noise suppression performance even with the unknown prior knowledge about direction of arrive with respect to speech signal.