Matthias Mauch

Contact Details

Title: Research Student
Tel: Internal: [13] 5528
National: 020 7882 5528
International: +44 20 7882 5528
Fax: National: 020 7882 7997
International: +44 20 7882 7997
Email: matthias.mauch@elec.qmul.ac.uk
Website: http://matthiasmauch.net/
Room: Eng. 112

Research Group: Centre for Digital Music

Supervisor: Simon Dixon

Research Topic: Automatic Harmony Analysis

One of the important parameters of Western music is harmony, and it has been covered extensively in musicologist literature. Harmonies (or chords) are generally thought of as vertical tonal content, i.e. notes appearing simultaneously in a piece of music.

Automatic analysis of harmony is currently applicable only to singular, rather homogeneous data sets and a more general approach is still very much an open research question. Only recently have the first automatic key induction systems working on audio been developed, and the first working chord extraction algorithms still do not achieve the accuracy of a trained human expert.

The focus of our work is twofold. On the one hand research on the automatic extraction of a harmonic representations from audio, on the other hand statistical learning from existing (symbolic) representations. The latter has been necessary in order to study the behaviour of chord progressions in the ideal case of human transcription. To this end, we have built a framework that empowers us to flexibly search collections of chord transcriptions stored in formats convertible to the Chord Ontology, thereby making constant use of the Music Ontology, the emerging backbone of the OMRAS2 project. Considering a bottom-up harmony paradigm devoid of global key, the features we are currently able to extract include simple relative frequency measures, but also structural border detection ("cadence finder"), and functional chord similarity (through divergence of context distributions).

We are currently investigating how statistical Natural Language Processing methods can be applied to study idioms and surprise in chord progressions. We have reason to believe that the endeavours in the symbolic domain will provide us with the necessary understanding and the data to improve results in the audio domain: The wealth of data can provide us with hand-tailored prior knowledge aiding the processing of audio.

Publications

M. Mauch, S. Dixon, C. Harte, M. Casey and B. Fields. Discovering chord idioms through Beatles and Real Book songs. In Proceedings of ISMIR 2007 Vienna, Austria, pp 255-258, 2007.