面向语音通信与交互的麦克风阵列波束形成方法

临境语音通信与智能语音交互都面临复杂声学环境中的远距离高保真拾音难题，解决这一难题的有效途径是使用由多个麦克风传感器组成的麦克风阵列或多通道拾音系统，这种系统的核心是信号处理，通过对空间采样的声场信息进行时、空、频三域的联合处理来实现声源定向/定位、信号增强、噪声抑制、混响抑制、声源分离、声场参数估计等功能。麦克风阵列信号处理的方法有很多，其中研究的最多、使用的最广的方法是波束形成。本文对麦克风阵列波束形成的原理、进展以及当前常用的方法进行简要综述，内容涵盖延迟求和、超指向、差分、正交级数展开、Kronecker和自适应波束形成方法等。论文侧重于方法原理、机理和架构方面的探讨，具体的算法实现细节感兴趣的读者可以参考相应的文献。 

Microphone Array Beamforming: an Overview

Immersive communication and human-machine speech interface systems have to face the problem of distant sound acquisition in complex acoustic environments where noise, reverberation, echo, and competing sources may coexist. To deal with this problem, microphone arrays or more generally multichannel sound acquisition systems have to be used. The major difference between a microphone array and a multichannel system lies in how the microphone sensors are selected and organized. In the former, the microphone sensors are carefully selected to have the same or similar sensitivity, signal-to-noise ratio (SNR), and responses and they are well organized into a particular geometry. All the sensors' signals in such an array are pre-amplified with a same gain and converted to the digital domain with a same clock. In contrast, a multichannel system (either centralized or distributed) may use multiple clocks and have sensors with different characteristics. The central component of both systems are signal processing, which operates on the sensors' outputs to achieve a certain objective such as source localization, noise reduction, signal enhancement, dereverberation, source separation, to name but a few. Considering the maturity level as well as the application breadth so far, we focus in this paper on microphone arrays and present a systematic overview of the state of the art of the associated beamforming algorithms including the delay-and-sum, superdirective, differential, and adaptive beamformers. We discuss the basic principles underlying those beamformers and their pros and cons. Due to space limit, however, we will not present much detail on the derivation of every individual algorithm. The interested reader is therefore encouraged to follow the respective references listed in the paper.