Optimization-based reproduction of diffuse audio objects

The creation of a diffuse sound event from a single audio signal is an important signal processing task, for instance in spatial audio reproduction or audio coding. Current algorithms based on decorrelation filters or frequency-dependent panning typically cause artifacts due to transients or time-domain aliasing. In this paper, we propose an optimization-based approach to diffusion that creates a set of filters to approximate a desired distribution of frequency-dependent propagation directions to create the perception of a diffuse sound field with a multi-channel audio system. Thus, the diffusion can be optimally adapted to a specific reproduction scenario. In addition, the transient response can be purposefully improved by imposing constraints on the time-domain filter coefficients.