Bidirectional sound transport

Abstract

We present a new sound propagation algorithm, Bidirectional Sound Transport (BST), based on bidirectional path tracing. Current state-of-the-art geometric acoustic method handles diffuse reflection by backward path tracing and uses diffuse rain to improve the validity of generated paths. We show that this can be viewed as a special case of bidirectional path tracing. By allowing the connections to be established between any nodes of the subpaths, we are able to improve the sampling quality when sound sources locate near scene objects. This ensures more stable rendering quality and eases ray budget selection. We propose a new metric based on the signal-to-noise (SNR) of the energy response to evaluate the performance of Monte-Carlo path tracing method for sound. Based on the metric, we develop an iterative algorithm to redistribute the samples among bounce numbers according to the statistic characteristics of the sampling of previous frames. We show that the sample redistribution algorithm converges and better balances between early and late reverberation. We evaluate our approach on different benchmarks and demonstrate significant speedup over prior geometric acoustic algorithms. We also discuss clustering algorithms used to improve the scalability for bidirectional sound transport.