Enhanced Acoustic Modelling for Auralisation using Hybrid Boundary Integral Methods

该项目将开发一种新的声学建模方法，非常适合模拟房间和城市广场，它将在准确性或计算效率方面优于现有方法。开发这样的算法是一个特别关注的从业人员使用听觉作为一个咨询工具，在声学设计的建筑空间。在这个过程中，来自模拟模型的数据通过扬声器系统呈现为声音，允许客户或利益相关者（不太可能是声学专家）对声学设计是否符合他们的需求做出判断。当然，这个过程只有在声学模型能够准确预测声音在空间中的表现时才有效，而目前的商业软件并不总是能成功完成这项任务，因为它所基于的高频几何传播假设在低频和衍射效应显著的空间中会失效。虽然存在替代的数值方法，但它们通常仅限于模拟低频率，因为随着频率或时间分辨率的增加，它们的计算成本变得不切实际。针对这些缺点，本项目将开发一种新的混合方法，该方法结合了几何方法和全数值边界元法（BEM）求解器的最佳特征，提供一种继承两种方法的理想特征的方案；即，完全误差可控的方案，在中低频段比几何方法更精确，但与标准BEM相比，在更高频率下的计算成本更低，所有这些都在一个统一的框架内实现。这样的模型可能包括所有的波项，几何和衍射，但低能量反射只包括在必要时达到给定的精度标准（例如声压级阈值或耳朵的可感知差异阈值的函数），因此计算效率将最大化。在听觉化过程中引入交互性元素，用户可以探索空间和/或对来源、建筑几何或材料进行动态更改，这从咨询生产力的角度来看是可取的，但对声学模型提出了极高的要求。不仅模型必须动态更新以反映用户所做的修改，而且对准确性的要求更加紧迫，因为客户选择引入的任何特征都必须准确渲染，即使它具有强烈的声学效果（例如凹聚焦表面，房间共振，不寻常的回声模式），专家很少或根本没有机会检查声音是否真实。我们提出的新算法将满足这些需求，因为除了提高精度外，它还具有理想的特性，即当建筑物几何形状或源位置发生变化时，只有一小部分容易识别的声学相互作用数据需要重新计算；因此，合并对时变建模和交互场景的支持将相对简单。为了实现这一目标，该项目还将开发一个新的面向听觉化的音频平台，该平台将通过数字滤波器网络代表声学相互作用，并将声音直接输出到音频硬件，模拟算法将面向以这种格式输出减少的声学模型。试点研究将调查如何支持交互性，如场景对象和相应的过滤网络元素的动态修改，以及如何将标准集总参数隔音和随机混响模型结合起来。该项目将以一个工作包结束，该工作包专门用于模拟一些现实世界的场景，这些场景会给当前的声学建模软件带来困难。

This project will develop a new acoustic modelling method, ideally suited for simulation of rooms and city squares, which will outperform existing methods either in its accuracy or computational efficiency. Developing such an algorithm is a particular concern for practitioners who use auralisation as a consultation tool in acoustic design of built spaces. In this process the data from the simulation model is rendered as sound by a loudspeaker system allowing a client or stakeholder, who is unlikely to be an expert in acoustics, to form a judgement on whether the acoustic design fits their needs. This process is of course only valid if the acoustic model delivers accurate prediction of how sound behaves in the space, and current commercial software does not always succeed in this task because the high-frequency geometric propagation assumption on which it is based breaks down at low frequencies and in spaces where diffraction effects are significant. Although alternate numerical methods exist they are typically limited to modelling only low frequencies since their computational cost becomes impractical as frequency or time-resolution is increased. In response to these shortcomings, this project will develop a new hybrid method which combines the best features of geometric methods and fully numerical boundary element method (BEM) solvers to provide a scheme that inherits desirable characteristics from both approaches; i.e. fully error controllable schemes, more accurate than geometric methods for low to mid range frequencies, but with reduced computational cost at higher frequencies compared to standard BEM, all achieved within a single unified framework. Such a model would potentially include all wave terms, geometric and diffracted, but lower energy reflections would only be included where necessary to achieve a given accuracy criterion (e.g. an SPL threshold or a function of the ear's perceptible difference limen) hence computational efficiency would be maximised.Introducing an element of interactivity to the auralisation process, where a user would be able to explore the space and/or make dynamic changes to the sources and building geometry or materials, would be desirable from a consultation-productivity perspective but place extremely high demands on the acoustic model. Not only must the model dynamically update to reflect the modifications made by the user, but the requirement for accuracy is even more pressing since any feature the client chooses to introduce must be accurately rendered, even if it has a strong acoustic effect (e.g. concave focussing surfaces, room resonances, unusual echo patterns), and there will be little or no opportunity for an expert to check that the sound is realistic. The new algorithm we propose will address these needs since, as well as having improved accuracy, it also has the desirable characteristic that only a small easily identified subset of the acoustic interaction data needs to be re-computed when a change in building geometry or source location occurs; incorporating support for modelling time variant and interactive scenarios would hence be relatively straightforward. Towards this goal the project will also develop a new auralisation orientated audio platform which will represent acoustic interactions by a network of digital filters and output sound direct to audio hardware, and the simulation algorithm will be geared towards outputting reduced acoustic models in this format. Pilot studies will investigate how interactivity might be supported, as dynamic modifications of scenario objects and corresponding filter network elements, and how standard lumped parameter sound insulation and stochastic reverberation models may be incorporated. The project will conclude with a work package dedicated to modelling some real-world scenarios which would cause difficulties for current acoustic modelling software.