Research Seminars

Analysing Recorded Music: A Pilot Project

Nicholas Cook, CHARM and Royal Holloway, University of London
Andrew Earis, CHARM and Royal College of Music, London

Wed 25 May 2005, 4:00 pm, Room 105

Abstract

This seminar outlines a project now beginning at the AHRC Research Centre for the History and Analysis of Recorded Music (CHARM), which is based at Royal Holloway. The seminar will consist of a (1) general introduction to the project by Nicholas Cook, which is directing it, and (2) a detailed account of the proposed means of data capture by Andrew Earis, software consultant to the project.

1. General Introduction to the Project

Within the last decade, powerful computational tools for addressing musicological questions (such as David Huron's Humdrum Toolkit) have been developed, but their use among historical and analytical musicologists (as opposed to specialists in computational musicology) has been very limited. Nevertheless they hold the promise of a reinvigorated style analysis which seeks to understand the relationship between structural context and performance interpretation on the basis of large data sets. This project will be based on Chopin's complete Mazurkas, involving analysis of a significant proportion of existing recordings of them, and making use of recently developed techniques for the mechanical capture of timing and dynamic information. The intention is not only to throw light on the interaction of compositional and performance style, using this in turn as a basis for interpreting geographical and chronological trends in the recorded performances, but also to explore the possibility of linking such analysis to the cultural meanings the Mazurkas have supported over the past 150 years.

2. Data Capture Methods

Measurable features of expressive musical performance include timing, dynamics, articulation and pedalling. Whilst such parameters can simply be measured by analysing MIDI performances (made on, for example, a Yamaha Disklavier piano), even more valuable information can be obtained by analysing acoustic (audio) recordings. Such recordings are universally available and provide a vast supply of invaluable musical information spanning over a century. Currently, the main method of analysing musical expression in audio recordings is by manual means - for example, timing is extracted by studying the waveform of the musical signal, using visual and auditory clues to estimate the onset time of each ‘event’. Recently developed techniques to automate this work using signal processing methods and other forms of acoustical analysis are presented, these potentially allowing a much larger quantity of musical data to be measured. Results are presented and future directions discussed.