Perceptual Sound Field Reconstruction and Coherent Emulation

该项目涉及开发一种新的5- 10声道音频技术，该技术将在（a）真实性、（B）听觉视角的准确性和稳定性、（c）最佳点的大小和（d）听觉体验方面优于现有技术。由于新技术旨在创建360度听觉视角，因此再现将在位于正多边形顶点的扬声器上进行。每个扬声器将由两个部件组成，一个部件将向收听者辐射直接声场，另一个部件将通过引入额外的散射来再现漫射声场。下面列出的特定任务的目标是找到捕获声场线索的最佳方式，并使用所提出的回放系统以将提供原始或期望声场的最令人信服的错觉的方式来渲染它们。(i)最佳麦克风阵列的建议播放系统将进行调查。所考虑的阵列将由放置在正多边形的顶点处的水平面中的麦克风组成，麦克风的数量等于扬声器的数量。对于每个阵列，将考虑从接近重合到稍微超过最佳值的范围内的不同直径以及不同的麦克风方向性图案。这些研究将针对扬声器配置的几个直径进行重复，以调查最佳阵列直径是否取决于扬声器布局的大小，如果是这样，则描述该依赖关系。最佳麦克风的方向性模式和阵列直径之间可能的依赖关系也将进行调查和表征。将根据上述标准（a）-（d）在关键听力测试中对阵列进行评估。实验将通过模拟来指导，这将提供对听音区域内产生的ITD和ILD线索的初步客观评估。同时，将研究所提出的技术产生的声场的数学模型，这可以为最佳麦克风阵列设计提供一些额外的见解。(ii)将系统地研究具有串扰消除的回放的影响。现有的串扰消除算法将首先使用，如果必要的话，新的算法，这是数字上的效率和有效的范围内的收听环境将被开发。然后，将研究用于具有串扰消除的回放的最佳麦克风阵列，即，将重复在（i）下描述的工作以用于具有串扰消除的再现。最后，将比较具有和不具有串扰消除的最佳系统。(iii)将研究直接/扩散声场分离的算法。当乐器的数量不超过麦克风的数量时，可以使用多通道均衡技术来找到干源信号，然后可以将干源信号与房间脉冲响应的直达/混响部分进行卷积，以分别获得直达/扩散声场分量。然而，音频中的多通道均衡由于过长的脉冲响应而特别具有挑战性，并且我们将开发用于音频应用的多通道均衡的数值有效算法。然后，我们将研究直接/扩散声场分解的心理声学近似，对声源的数量没有限制。(iv)将根据标准（a）-（d）在关键听力测试中系统地研究用于获取直接声场线索的近一致定向麦克风阵列和用于获取扩散声场线索的基于全向或双向麦克风的宽间隔阵列的组合。这种方法将与（i）-（iii）中描述的方法进行比较，其中相同的阵列用于两个声场分量。

The project is concerned with the development of a new 5--10 channel audio technology which would improve over existing ones in terms of (a) realism, (b) accuracy and stability of the auditory perspective, (c) size of the sweet spot, and (d) the envelopment experience. Since the new technology aims to create a 360 degrees auditory perspective, the reproduction will take place over speakers positioned at vertices of a regular polygon. Each speaker will consist of two components, one which will radiate the direct sound field toward a listener, and another which will reproduce diffuse sound field by introducing additional scattering. The goal of the particular tasks, listed below, is to find optimal ways to capture sound field cues and render them using the proposed playback system in a manner which would provide the most convincing illusion of the original or desired sound field.(i) Optimal microphone arrays for the proposed play-back system will be investigated. Arrays considered will consist of microphones placed in the horizontal plane at the vertices of a regular polygon, with the number of microphones equal to the number of speakers. For each array, different diameters, in the range from near coincident up to somewhat beyond the optimal value, and different microphone directivity patterns will be considered. These studies will be repreated for a few diameters of the speaker configuration to investigate if the optimal array diameter depends on the size of the speaker lay-out, and if so to characterize that dependence. Possible dependencies between the optimal microphone directivity patterns and array diameters will be also investigated and characterized. Arrays will be evaluated in critical listening tests according to criteria (a)--(d) stated in the above. Experiments will be guided by simulations which would provide initial objective assessment of ITD and ILD cues generated within the listening area. In parallel, mathematical models of sound fields generated by the proposed technology will be investigated, which could provide some additional insight into the optimal microphone array design. (ii) The impact of play-back with cross-talk cancellation will be be systematically investigated. Existing cross-talk cancellation algorithms will be first used, and if necessary, new algorithms which are numerically efficient and effective in a range of listening environments will be developed. Then optimal microphone arrays for play back with cross-talk cancellation will be investigated, i.e. the work described under (i) will be repeated for reproduction with cross-talk cancellation. Finally, optimal systems with and without cross-talk cancellation will be compared.(iii) Algorithms for direct/diffuse sound field separation will be studied. When the number of instruments does not exceed the number of microphones, multichannel equalization techniques can be used to find dry source signals, which can then be convolved with direct/reverberant parts of room impulse responses to obtain direct/diffuse sound field components, respectively. Multichannel equalization in audio is, however, particularly challenging owing to excessively long impulse responses, and we will develop numerically efficient algorithms for multichannel equalization for audio applications. Then we will study psychoacoustic approximation to direct/diffuse sound field decomposition with no restriction on the number of sources. (iv) Combinations of near-coincident directional microphone arrays, for acquiring direct sound field cues, and widely spaced arrays based on omni-directional or bi-directional microphones, for acquiring diffuse sound field cues, will be systematically investigated in critical listening tests according to criteria (a)--(d). This approach will be evaluated in comparison with the approach described in (i)--(iii) where the same array is used for both sound field components.