基于零陷谱减的GSC二元麦克风小阵列语音增强算法

A GSC algorithm based on null spectral subtraction for dual small microphone array speech enhancement

为了提高广义旁瓣抵消器语音增强算法在二元麦克风小阵列中的噪声抑制能力,提出一种基于零陷谱减二元麦克风小阵列广义旁瓣抵消器的改进语音增强算法.在广义旁瓣抵消器固定波束支路上利用谱减法抑制目标语音零陷方向噪声能量以提高其信噪比,在自适应支路使用基于动态收敛步长的快速分块最小均方自适应滤波器进一步抑制剩余噪声,以降低算法复杂度并提升自适应滤波器的收敛性.实验结果表明,相对于其他二元麦克风小阵列波束形成语音增强算法,该算法可以在任意方向获得较高质量的目标语音.

To improve the denoising ability of generalized sidelobe canceller（GSC） algorithm for dual small microphone array,a new GSC based on null spectral subtraction is proposed.In fixed beamforming branch of GSC,noise energy of desired speech in null direction is suppressed by spectral subtraction to improve the signal to noise ratio（SNR）.At the same time,in adaptive beamforming branch of GSC,a fast block least mean square（FBLMS） based adaptive filter further suppresses the residual noise,in which a dynamic convergence step is used to reduce the computational complexity and improve the convergence of adaptive filter.Compared with other denoising algorithms used for dual small microphone array,experimental results show that the desired speech obtained by the proposed algorithm is always of higher quality in any direction.