Differential beamformers with small-size microphone arrays are very attractive for audio and speech signal acquisition thanks to their high directivity and frequency-invariant spatial responses. However, such beamformers often suffer from significant white noise amplification at low frequencies, which makes their implementation in real-world systems challenging. One widely used way to circumvent this issue is to increase the number of microphones in the design of a given order differential beamformer, leading to the so-called robust differential beamformer in which the redundancy provided by the additional sensors are used to improve the white noise gain (WNG). But even with this robust solution, white noise amplification at low frequencies still exists. In this article, instead of trying to improve WNG, we adopt a method to render the white noise in such a way that it affects less the perception of the speech signal of interest. Specifically, we propose a binaural differential beamforming method in which a differential beamformer is designed with two sub-beamforming filters that simultaneously generates two outputs, one for the left ear and the other for the right ear. Motivated by psychoacoustic experiments, we design these two filters in such a way that they are orthogonal so that the coherence between the white noise at the binaural outputs is minimized while the coherence between diffuse noise is maximized. With the proposed binaural differential beamformers, the desired signal components and (amplified) white noise at the binaural differential beamformer's outputs are rendered into different directions or zones. Consequently, the human perception system can better distinguish the desired signal from white noise for improved speech intelligibility. The superiority of the proposed binaural beamforming technique is justified by simulations, experiments, as well as listening tests.

中文翻译：

---

用于提高语音清晰度的异相双耳差分波束成形

由于具有高方向性和频率不变的空间响应，具有小尺寸麦克风阵列的差分波束形成器对于音频和语音信号采集非常有吸引力。然而，这种波束形成器通常会在低频下受到显着的白噪声放大，这使得它们在现实世界系统中的实现具有挑战性。规避此问题的一种广泛使用的方法是在给定阶差分波束成形器的设计中增加麦克风的数量，从而产生所谓的稳健差分波束成形器，其中附加传感器提供的冗余用于改善白噪声增益 (WNG)。但即使有了这个强大的解决方案，低频的白噪声放大仍然存在。在本文中，与其尝试改进 WNG，我们采用一种方法来渲染白噪声，使其对感兴趣的语音信号的感知影响较小。具体来说，我们提出了一种双耳差分波束成形方法，其中差分波束成形器设计有两个子波束成形滤波器，可同时生成两个输出，一个用于左耳，另一个用于右耳。受心理声学实验的启发，我们以正交的方式设计这两个滤波器，以便最小化双耳输出白噪声之间的相干性，同时最大化漫射噪声之间的相干性。使用所提出的双耳差分波束形成器，双耳差分波束形成器输出处的所需信号分量和（放大的）白噪声被渲染到不同的方向或区域。最后，人类感知系统可以更好地将所需信号与白噪声区分开来，从而提高语音清晰度。模拟、实验以及听力测试证明了所提出的双耳波束成形技术的优越性。