Public defence in Acoustics and Speech Technology, M.Sc.(Tech.) Jon Fagerström

The title of the thesis: Velvet Noise in Audio Processing

Thesis defender: Jon Fagerström
Opponent: Dr. Stephan Ewert, Carl von Ossietzky University of Oldenburg, Germany
Custos: Prof. Vesa Välimäki, Aalto University School of Electrical Engineering, Department of Information and Communications Engineering

This dissertation investigates the application of velvet noise, a specific type of sparse noise, to various audio processing tasks, including artificial reverberation, audio decorrelation, and variation synthesis. Velvet noise has previously been used for reverberation algorithms, audio decorrelation, and sound synthesis, but its potential applications have not been fully explored. This research aimed to enhance the understanding and effectiveness of velvet noise in these domains by developing new variants and novel audio signal processing algorithms based on them. 

With its smooth temporal envelope and minimal density, long velvet noise sequences are well-suited for late-reverberation modeling, whereas short velvet noise sequences provide efficient decorrelation and variation filters. This work builds upon previous knowledge and introduces novel methods to address gaps in modeling non-exponential late reverberation, decorrelation properties of feedback delay networks, and humanizing sampling synthesis. 

The primary outcomes of the research include the development of the binaural dark-velvet-noise reverberator, capable of accurately synthesizing binaural late-reverberation based on binaural target impulse responses, and the introduction of short velvet-noise filters for effective variation filtering of percussive sampled sounds. These results have significant practical applications in improving audio quality, particularly for perceptually accurate virtual audio scenes and sound synthesis. The dissertation also contributes to understanding the perceptual effects of late-reverberation models, decorrelation, and variation synthesis, highlighting the need for further research into the perception of these audio phenomena.

In conclusion, this dissertation advances the use of velvet noise in audio processing by introducing novel variants and demonstrating their potential in artificial reverberation, decorrelation, and variation filtering. These findings have significant implications for creating more realistic and immersive audio experiences in various applications.

Keywords: artificial reverberation, audio signal processing, decorrelation

Thesis available for public display 10 days prior to the defence at: https://aaltodoc.aalto.fi/doc_public/eonly/riiputus/

Contact: https://www.linkedin.com/in/jonfagerstrom/

Doctoral theses in the School of Electrical Engineering: https://aaltodoc.aalto.fi/handle/123456789/53