Analysis and Synthesis of Directional Reverberation

Abstract

The reproduction of acoustics is one of the key challenges in spatial sound reproduction. In order to reach high levels of realism and immersion, reproduced signals must be processed through a set of filters accounting for real-life phenomena. From the scattering and absorption of sound on walls to its diffraction around objects and our head, the propagation of sound waves in a room creates a complex sound field around a listener. While recreating all of the underlying parts of sound propagation is not yet within our reach for real-time applications, spatial sound techniques aim to minimize the computational cost by focusing the efforts on the most perceptually critical components. In this dissertation, the reproduction of reverberant sound fields is investigated through the development of analysis methods, reverberation algorithms, and decorrelation techniques. A particular emphasis is given to the perception and reproduction of directional characteristics present in late reverberation. Most reverberation algorithms consider that the sound energy is evenly distributed across space in all directions, after an initial period of time, since the diffusion of energy leads to more homogeneous and isotropic sound fields. However, previous work has demonstrated that insufficient diffusion in a room leads to anisotropic, and directional, late reverberation.

In this dissertation, a complete framework is proposed for the objective and subjective analysis of directional characteristics as well as a novel delay-network reverberation method capable of producing direction-dependent decay properties. The reverberator is also expanded to offer efficient frequency- and direction-dependent processing. The proposed algorithm contains all the required elements for the auralization of reverberant sound fields, which may be modulated in real-time to support six-degrees-of-freedom sound reproduction. The decorrelation of audio signals, which occurs naturally during the diffusion of energy in a sound field, is another important aspect of sound reproduction. This dissertation considers the use of velvet-noise sequences, a special type of sparse noise signals, as decorrelation filters and offers a method to optimize their characteristics. Velvet noise is also proposed for the reproduction of an existing impulse response using a small set of time- and frequency-dependent information, along with a reverberator using velvet noise to improve the echo density of a delay network reverberator. Overall, the results contained in this dissertation offer new insights into the perceptual ramifications of reverberant sound fields containing directional characteristics and their reproduction. The methods presented bring applications in the context of immersive sound reproduction, such as in virtual and augmented reality.