Speaker
Description
Rendering diffuse sound fields of late reverberation with computational efficiency and accurate spatial impression is a challenging task. This paper introduces spatialized velvet-noise, a framework that efficiently generates a decorrelated sound field from a single monaural reverberation in real time. The method utilizes a single velvet-noise sequence that is distributed spatially. The approach is compared to other computationally efficient decorrelation strategies, including allpass filters, simple delays, and alternative velvet-noise configurations. This comparison involves a technical analysis of spectral deviations and interaural coherence, alongside a formal listening experiment. Results reveal that spatialized velvet-noise significantly outperforms most of the tested configurations in providing an enveloping sensation or maintaining a neutral timbre. Notably, none of the tested method was found to surpass spatialized velvet-noise in the listening experiment. While simple delays are the only computationally more efficient option, spatialized velvet-noise significantly outperforms them in the listening test.