Public defence in Acoustics and Audio Signal Processing, M.Sc. Alec Wright

The title of the thesis: Neural Modelling of Audio Effects

Doctoral student: Alec Wright
Opponents: Prof. Joshua D. Reiss, Centre for Digital Music, Queen Mary University of London, UK
Custos: Prof. Vesa Välimäki, Aalto University School of Electrical Engineering, Department of Information and Communications Engineering 

This thesis develops a number of pioneering new techniques integral to the ongoing development of next generation music software. The work focuses on creating high-quality digital emulations of analog musical hardware, dramatically simplifying the process of developing audio software for use in music production and performance. The results can be applied, for example, to emulating a wide range of guitar sounds, from the clean or crunchy tones used in rock and pop to the highly distorted guitar often found in heavy metal music. 

Existing research in the field often requires complex and time-consuming analysis and development for each new system being modelled. In contrast, the methods proposed in this thesis use neural networks, a subfield of machine learning, allowing for new models to be created using just a few minutes of audio data and a personal computer. 

Important results in the thesis include demonstrating that these methods achieve excellent perceptual quality - making them hard to distinguish from the real devices they are emulating. Additionally, software was developed demonstrating that the models are real-time capable and have a low computational cost, making them suitable for use in live music performance and production. Finally, new methods were developed to emulate audio effects processing from existing musical recordings, opening new possibilities for emulating sounds found on famous records. 

These results make a significant contribution toward the exciting and fast-developing field of music technology - which is rapidly enhancing and improving the tools available to musicians and expanding what is possible when producing and performing music.

Keywords: Deep learning, neural networks, guitar amplifiers, virtual analog

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

Contact:

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