Analysis and Synthesis of Small Room Acoustics

混响是一种日常的声学体验，当声波在封闭的空间中来回反弹时就会发生。房间声学特征，如房间几何形状、源和听众位置以及边界材料，决定了所产生的混响的物理和感知特性。在许多应用中，需要人工再现混响效果。人工混响可以基于对房间声学的纯物理模拟。然而，所需的巨大计算复杂性限制了精度，具体取决于应用要求。因此，存在对物理上不一定准确但在感知上令人信服的高效算法的需求。Assai项目旨在研究从小房间测量中分析声学特征并合成感知准确的房间脉冲响应的技术，特别关注一种计算高效的方法，即反馈延迟网络(FDN)。FDN代表了一大类稀疏递归滤波器，因此本项目中的基础研究与广泛的数字信号处理应用相关，包括去相关、数字声音合成和物理建模。我们介绍了FDN的两个创新性扩展：控制房间模式和扩散。为了便于系统地研究感知平移，我们提出了从房间声学测量到参数估计、物理平移、模型简化和评估的端到端设计。我们专注于小房间声学，因为它在增强现实、仪器建模和电影后期制作等应用中具有特殊的相关性。小房间声学的测量和分析是基于空间房间脉冲响应和房间几何深度图，这反过来又驱动计算声学模拟。然后，利用感知激励的误差测量从物理仿真中导出FDN。该评估是根据真实世界的参考进行的，并调查增强现实应用的可行性。

Reverberation is an everyday acoustical experience which occurs when sound waves are bouncing back and forth in an enclosed space. Room acoustic features such as room geometry, source and listener positions and boundary materials determine the physical and perceptual properties of the resulting reverberation. In many applications, it is necessary to recreate the reverberation effect artificially. Artificial reverberation can be based on purely physical simulation of room acoustics. However, the required immense computational complexity limits the accuracy depending on the application requirements. Thus, there is a demand for efficient algorithms which are not necessarily physically accurate but which are perceptually convincing. The ASSAI-project aims to investigate techniques for analyzing acoustic features from small room measurements and synthesizing perceptually accurate room impulse responses with a particular focus on a computationally efficient method, i.e., feedback delay network (FDN). FDNs represent a broad class of sparse recursive filters such that the fundamental investigations in this project are relevant to a wide range of digital signal processing applications including decorrelation, numerical sound synthesis, and physical modeling. We introduce two innovative extensions to FDNs: control of room modes and diffusion. To facilitate systematic investigation of the perceptual translation, we propose an end-to-end design from room acoustic measurement to parameter estimation, physical translation, model reduction and evaluation. We focus on small room acoustics because of its particular relevance in applications such as augmented reality, instrument modeling, and movie post-production. The measurements and analysis of the small room acoustics are based on spatial room impulse responses and room geometry depth maps which in turn drives computational acoustic simulations. The FDN is then derived from the physical simulation with a perceptually motivated error measure. The evaluation is performed against real-world references and investigates viability for augmented reality applications.