MIR book reviews

by graham

Computer Music Modeling and Retrieval 2003 – Wiil, Uffe K. (Ed.)

Weird potpourris of composition, music retrieval, system architecture, and musicological research papers are entertaining, if you like that sort of thing. As with other sequence analysis domains, the Similarity Matrix is key. My favorites:

Real Time Beat- Estimation Using Feature Abstraction – Jensen and Anderson

Beat and tempo extraction is a core MIR problem. They present a straightforward probability-based approach. Results are OK but on not much data, it is always hard to tell how things like this will actually perform. To evaluate it, one can only implement it and test it versus others. As the problem and field matures formal evaluation procedures (mirex 2005 results!) will decide how they do in comparison. Interesting because I plan on implementing a beat extractor soon.

The Study of Musical Styles in Central Africa: the Use of Interactive Experimental Methods – Marandola

Musicologist describes the use of interactive computer techniques to learn the basic scale and tone structure of two native music traditions—the Hocket Instrumental Polyphonies of Ouldeme flutes (Paris), and the Vocal Polyphonies of Bezdan Pygmies (Cameroon). In the latter case, polyphonic pieces were recorded on multiple tracks and re-synthesized by the computer based on models of harmony. The performers would reject the models until they were somewhat accurate.

Evolving Automatically High-Level Music Descriptors from Music Signals – the power team Zils and Pachet
(pdf of a similar work, access to CMMR paper closed)

High-Level music descriptors, like the presence of certain instruments, tempo, genre, are difficult things for which to design algorithms. In this work, signal networks are built up from low-level descriptors (FFTs, filters, correlations, etc) and grown using genetic algorithms. They attack the problems of Sine+Colored Noise, Presence of Voice, and Perceived Intensity. Since these are not the hardest problems around, this isn’t earth shattering, but it is pretty cool.

Other interesting papers on physical modeling synthesis, structural patterns in pop music and self-organizing maps.

Leave a Reply