Accelerated Interactive Auralization of Highly Reverberant Spaces using Graphics Hardware

TL;DR

This work presents a GPU-accelerated, multichannel auralization framework for real-time synthesis of highly reverberant spaces, integrating uniform partitioned convolution with real-time acoustic feedback cancellation. It demonstrates substantial latency reductions (up to ~60x) over CPU implementations and supports large channel counts and long reverberation filters at 48 kHz, achieving end-to-end latency around a dozen milliseconds for practical block sizes. The method leverages parallel processing on graphics hardware to maintain low latency while preserving stability in the presence of feedback, and the authors provide a Python-based implementation with public code. Future work includes optimizing performance and conducting perceptual evaluations.

Abstract

Interactive acoustic auralization allows users to explore virtual acoustic environments in real-time, enabling the acoustic recreation of concert hall or Historical Worship Spaces (HWS) that are either no longer accessible, acoustically altered, or impractical to visit. Interactive acoustic synthesis requires real-time convolution of input signals with a set of synthesis filters that model the space-time acoustic response of the space. The acoustics in concert halls and HWS are both characterized by a long reverberation time, resulting in synthesis filters containing many filter taps. As a result, the convolution process can be computationally demanding, introducing significant latency that limits the real-time interactivity of the auralization system. In this paper, the implementation of a real-time multichannel loudspeaker-based auralization system is presented. This system is capable of synthesizing the acoustics of highly reverberant spaces in real-time using GPU-acceleration. A comparison between traditional CPU-based convolution and GPU-accelerated convolution is presented, showing that the latter can achieve real-time performance with significantly lower latency. Additionally, the system integrates acoustic synthesis with acoustic feedback cancellation on the GPU, creating a unified loudspeaker-based auralization framework that minimizes processing latency.

Figures (6)

• Figure 1: Block diagram of an interactive multichannel auralization system containing $L$ loudspeakers and $P$ microphones. The system consists of a pre-processing stage, where the microphone signals are processed by a pre-processor function $\mathbf{G}$, a synthesis stage with synthesis filters $\mathbf{H}_\textrm{aur}$, and an acoustic feedback cancellation stage which subtracts the estimated feedback signals from the microphone input. The feedback path for each loudspeaker-microphone pair is estimated a-priori and stored in the estimated feedback path matrix $\mathbf{\hat{F}}$.
• Figure 2: Block diagram of the uniform partitioned convolution algorithm used in wefers2010high. The algorithm is visualized for a single-channel input block $\mathbf{x}_n$ and output block $\mathbf{y}_n$. The -size for the input and filters was set to $n_f = 2 n_x$, where $n_x$ is the block size. The frequency-domain input blocks are denoted by $\mathbf{\bar{x}}_n$. The resulting output block is obtained using the method. The multiplication and summation of the input block spectrum with each of the sub-filter responses can be performed entirely in the frequency-domain on the .
• Figure 3: Performance of the and implementations of the partitioned convolution algorithm and the real-time auralization system as a function of the block size.
• Figure 4: Performance of the and implementations of the partitioned convolution algorithm and the real-time auralization system as a function of the filter length.
• Figure 5: Performance of the and implementations of the partitioned convolution algorithm and the real-time auralization system as a function of the number of output channels.