Modelling Music Memory and the Perception of Melodic Similarity (M4S) was a 3-year project running from 2006-2009 at the ISMS group in the Computing Department of Goldsmiths, University of London. The project was funded by the British EPSRC (grant reference: EP/D0388551). The project was an interdisciplinary venture connecting perspectives from psychology, computer science, and musicology and focussing on the human cognition of melodies. Among its outcomes are an open source software toolbox for melody analysis, 22 presentations at international conferences and events including many papers published in conference proceedings, and 11 journal papers and book chapters, some of which them already published and some of which are still awaiting publication. These web pages document and report the project`s goals and achievements and point as well to the relevant publications where the actual research results can be found in full detail.
For a quick overview read the below summary of the project as it was conceived in 2006 and the section summarising the results written in 2009. For a first impression of the type of research that can be done using FANTASTIC, the software toolbox developed during the project, see this presentation given at the 2009 meeting of the German Society for Music Psychology (DGM) in September 2009.
The main goal of the project Modelling Music Memory and the Perception of Melodic Similarity (M4S) is to characterise how people represent melodies in their minds. This includes the perception of melodies as well as their representation in memory. Typical questions are:
- What can ordinary people remember from a tune they have heard just once?
- Which parts and which aspects of melodies are retained better than others (e.g. rhythm vs. pitch)?
- What are typical memory errors for melodies?
- What are the cognitively relevant features of melodies and how are the categorised?
The project focuses on people with no formal music training and little explicit knowledge about music structure. To demonstrate that even people with little explicit music knowledge may exhibit a good non-verbal and implicit understanding of the regularities (structure) in a musical style they are familiar with, we chose western pop music as the musical domain, listened to by a large part of the western urban population.
The tasks of the project are to
- define algorithmic features that can represent knowledge about pop music, and especially pop music melodies;
- test these features for their cognitive validity in psychological experiments;
- describe statistical distributions of these features in a representative corpus of western pop music;
- model human memory performance using these features in psychological experiments.
The main assets of the project are:
- A database of 14067 full score, high quality transcriptions of pop songs from the 1950s up to the present with matching metadata. The transcriptions were acquired as MIDI files from a commercial supplier (Geerdes MIDI music) that generally provides musicians with MIDI files and Karaoke tracks.
- The ISMS’ AMUSE database system and interfaces at Goldsmiths Department of Computing, which is used to represent the pop songs as database objects.
- The specialised software tools SIMILE and MELFEATURE, programmed by Klaus Frieler, for melodic similarity computation and melody feature extraction.
Possible domains of application of the project’s results include:
- The design of interfaces and structure of melody and music databases.
- The detection of cover songs and melody plagiarism.
- In general, exploring the inter-relatedness of songs and musical genres.
- The description of certain music repertoires in terms of features (e.g. the common features of hit songs).
- Tools for music analysis.
- Insights into implicit and explicit memory for music, a non-verbal but language related domain.