Profile
Project Title:
Time-domain Music Source Separation: Developing Novel Tools for Music Production
Abstract:
This project approaches source separation from an audio production perspective, where we develop source separation tools to be able to solve various live/studio production tasks. This includes use cases such as noise suppression in studio recordings, binaural rendering, 3-D upmixing, remixing and intelligent mixing. Such tasks require high fidelity source separation models where the type of sources present in the mixture are not fixed. We intend our system to introduce minimal artifacts while being robust towards a variety of timbres, audio effects and recording conditions. Taking motivation from the success of recent time-domain speech enhancement methods, we focus our research on time-domain deep learning architectures. We introduce novel datasets enabling monotimbral separation research for chamber ensembles and explore the impact of various mixing practices and recording scenarios on source separation based applications like bleed reduction, room and microphone correction.
C4DM theme affiliation:
Music Informatics
Teaching
Music Informatics (Postgraduate/Undergraduate)
This module introduces students to state-of-the-art methods for the analysis of music data, with a focus on music audio. It presents in-depth studies of general approaches to the low-level analysis of audio signals, and follows these with specialised methods for the high-level analysis of music signals, including the extraction of information related to the rhythm, melody, harmony, form and instrumentation of recorded music. This is followed by an examination of the most important methods of extracting high-level musical content, sound source separation, and on analysing multimodal music data.
Research Methods and Responsible Innovation (Postgraduate)
This module will teach generic high-level research and transferable skills applicable to pure and applied research in computer science and engineering. The module fosters the development of practical understanding of established approaches, methods and techniques of research; conceptual understanding that enables critical and rigorous evaluation of research; ability to communicate ideas and conclusions logically and fluently in both written and oral contexts. It will also discuss responsible research and innovation practices, acknowledging that science can raise questions and dilemmas, is often ambiguous in terms of purposes and motivations and unpredictable in terms of impacts. Topics include research writing with an introduction to LaTeX, research ethics and responsible innovation, quantitative, qualitative and reproducible research methods, including experiment design and basic statistical analysis with an introduction to statistical programming and an introduction to scientific programming environments and version control systems.
Research
Research Interests:
Audio Source Separation, Intelligent Music Production, Intelligent Audio Enhancement, Perceptual Evalutation.
