Ambisonic diffuse sound rendering for real-time virtual reality simulations

Abstract

Regarding the recent trend and the rapid development of virtual reality (VR) media platforms that aim to provide users with immersive experiences, there is a rising demand for fast, high-quality acoustic simulations. The auralisation (the process of generating an audible result that corresponds to the acoustics of a virtual scene) of a 3D virtual environment involves modelling its acoustics and rendering the data into the binaural room impulse response (BRIR) through the simulation of the acoustic propagation in the room. The sound field within a virtual scene is composed of direct sound and specular reflections with well-defined directions, as well as late reverberation characterised by the sounds impinging from every direction enveloping the listener. In particular, the simulation of the BRIR for late reverberation plays a crucial role in the perceptual quality of the auralisation result. Despite its importance, late reverberation is often modelled as a simple diffuse field, and its precise modelling requires a large amount of computational resources due to its complex variation in space and time. This study explores acoustic simulation techniques and rendering methods for reproducing a virtual diffuse field and proposes new efficient methods motivated by Ambisonics used for sound field reproduction in spatial audio. The proposed methods are evaluated in comparison with a reference Brute-force rendering method, with regards to computational efficiency as well as the preservation of acoustic parameters, anisotropic reverberation, and the general perceptual quality of the rendered diffuse sound field in a virtual scene.