面向个人近场环境的局部3D声场合成技术研究

With the growing popularity of immersive virtual reality equipment，traditional perceptual simulation 3D audio technology is difficult to obtain personalized information and not able to create 3D audio effect for different individuals，which can hardly meet the requirement of personal VR experience. Sound field synthesis can physically reconstruct 3D sound field without relying on individual characteristics, which caused wide public concern. High quality sound field synthesis requires huge space sampling, just synthesis the region of interest (ROI) sound field is a popular way to ease the strict requirements. But ROI ignored the fact that only the sound field outside the ear determines the individual 3D sensing, which restricts the improvement of synthetic quality. Therefore, this application focuses on the local 3D sound field synthesis technology for personal near-field environments. At first, we will discuss the restricted spherical sampling method and aliasing mechanism, aiming to propose a method of modifying the weight function under restricted sampling, making it is possible to synthesize the local interesting sound field; Secondly, we will study the closed boundary and the human body reflection pattern and propose the near-field reflection sound cancellation algorithm to reduce the influence of the reflection. At last, we plan to utilize the human hearing perceptual characteristics to guide high frequency sound field synthesis, which can reduce 3D sound quality loss caused by the high frequency distortion. It is expected that the objective quality of the local sound field will be increased by 10% in the case of a considerable number of speakers, and subjective quality increased by 15%. The synthesized sound quality of the non-point source will be increased by 0.4 MOS points, finally meeting the virtual reality 3D audio technology upgrading needs.

随着沉浸式虚拟现实设备的日益平民化，传统的感知模拟3D音频技术难以获取个性化信息从而为不同个体精确呈现三维声效，无法满足个人虚拟现实体验需求。声场模拟不依赖个体重建三维物理声场，引起了广泛关注。针对声场合成巨大的空间采样要求，现有研究将感兴趣区域思想引入声场合成，但忽略了决定3D听感的只是耳外局部声场的事实，制约了合成质量的提升。为此，本申请研究面向个人近场环境的局部3D声场合成技术。首先，从受限球面采样方法和混叠机理出发，提出采样受限下权重函数修正方法，使局部声场合成成为可能；其次，研究封闭边界和人体的近场反射模式并提出抵消算法，降低近场反射声影响；最后，利用人耳空间感知特性指导高频信号合成，减小高频失真造成的3D音质损失。预计在扬声器数目相当的情况下将局部声场的客观质量提高10%，主观音质提高15%，对非点声源的合成音质提高0.4个MOS分，满足虚拟现实3D音频技术升级换代需求。

本项目针对声场合成巨大的空间采样需求难以满足的现状，将感兴趣区域思想引入声场合成，研究了非均匀采样理论及其在声场合成技术中的应用，采用重叠窗非均匀采样提升了感兴趣区域声场合成的质量，探索了三维声场的感知机理。首先，从受限球面采样方法和混叠机理出发，提出采样受限下非均匀采样方法，推导了理论声场失真并进行了仿真验证；其次，研究了基于重叠窗函数的非均匀采样算法，实现了信噪比在感兴趣区域内外的重新分配；最后，研究了人耳的三维空间感知特性，并用于指导三维声场的音源分类。在扬声器数目相当的情况下对感兴趣区域的声场信噪比可提高3.8~5.1dB，但同时会引起非兴趣区域信噪比下降。所提非均匀排布方法与典型的均匀排布的最佳模态带宽相比，对感兴趣区域音源的无混叠半径提高了12.9%~22.6%。三维音源的分类结果准确性在公共数据集上有2.8~4.2%的提高，整体上完成了申请书预定研究内容和目标。