Are you fascinated by music and computing? Join the
Music Information Computing Group as a PhD candidate to work on computational modelling of musical style – a key challenge in Music Information Retrieval (MIR). Style understanding drives user modelling, cultural heritage preservation, music categorization, transcription, recommendation and historical analysis. The project aims at explainable models with broad musical coverage to deepen insights into musical style, perception and preferences.
Your jobThe core aim of the project is to design explainable computational models that drive high-impact applications in Music Information Retrieval (MIR), while advancing our computational understanding of musical style: what defines it, what are its elements, how is it structured and perceived, how does it vary?
The project will build on various theories. A promising starting point is Leonard B. Meyer’s theory of musical style, which defines style as a replication of patterning. A central challenge in this approach is to identify structural elements of music as instances of patterns. What these patterns are differs culturally and historically. The body of literature on topic theory (founded by Leonard Ratner) offers a point of departure to identify such patterns and to understand the way in which these are replicated and perceived. For example, a fragment of music could allude to a ‘fanfare’, or to a ‘horn call’, to just mention two examples out of many. These kinds of patterns have many occurrences throughout music history. Can we design computational models for such topics? How do topics function in game music and film music? How are different musical styles interconnected by occurrences of topics?
We also envision to connect with current understanding of music cognition, specifically building on insights on musical memory. There is a class of modular cognitive models of music processing that include a ‘musical lexicon’ as one of the cognitive modules. This ‘musical lexicon’ determines for a given listener what musical patterns can be recognized. Understanding of this personalized music perception plays a role in user modelling for interactive music systems.
An important challenge lies in designing models that go beyond merely achieving high accuracy in classifying musical styles or genres, or in detecting specific musical patterns. The process of modelling facilitates the understanding of the patterns through a computational lens. This calls for strong expertise in computational methods, machine learning, and data modelling combined with solid knowledge of music. We particularly aim to cover a broad range of musical traditions and cultures world-wide, both contemporary and historical.
In this project you will:
- prepare data sets representing a wide range of musical styles, including both audio and symbolic formats;
- design appropriate data structures for representation of music;
- design computational detectors for various musical patterns;
- design, implement and evaluate computational models of musical style;
- apply these models in several case studies.
Furthermore, you will communicate results in academic presentations and publications, and ultimately in a PhD thesis. During the project, you will expand your academic network. A moderate percentage of the time will be spent on teaching tasks within the department, providing you with the opportunity to gain experience in teaching.