Wednesday 29 January 2020, 4.00PM to 5.30pm
Speaker(s): Dr Katerina Kosta
This presentation covers the role of creativity in performing a musical score and composing music. Initially we focus on how computational performance analysis can give insights into a performer’s idiosyncrasy by relating audio features, such as timing and dynamics, combined with features from the score. We will discuss findings from various studies which examine variations in performances of the same pieces by several pianists. In terms of automatic composition, music generative models are able to learn style features from a dataset and create a unique musical piece. The quality of the outcome, depends primarily on the system’s input and the system’s evaluation. The above characteristics contain some of the big challenges of generative models. The learnt features can be further enhanced by enriching the encoding beyond the pitch and duration. A proposed encoding is described. Finally, we will present new approaches to analysing and building creative systems.
Dr Katerina Kosta is a Senior Machine Learning Researcher at ByteDance's AI lab in London. She gained her PhD from the Computer Science and Electronic Engineering department of Queen Mary University of London in 2017, conducting research on computational modelling and quantitative analysis of dynamics in performed music. Her other research interests included custom data structures, pattern recognition and machine learning for music synthesis and analysis of perceived emotion in music audio. She received degrees from National and Kapodistrian University of Athens (Mathematics) and Filippos Nakas Conservatory, Athens (Piano), and a Sound and Music Computing Masters from the Music Technology Group, UPF, Barcelona.
Location: Rymer Auditorium, Music Research Centre, Campus West