Automatic Music Analysis References

Most research in music analysis focuses on the structural and stylistic elements at the symbolic level. That's interesting, but it can only be applied to abstract representations of music. I'm interested in the other kind of music analysis--the kind applicable to actual music.

My MSc dissertation is now hopelessly out of date, but this is good news, as it signifies progress in the field.

This is far from a complete list of people and research in the field of automatic music analysis, but it should help people just getting started (and it's a useful reference for me).

Conferences

ISMIR

Music Information Retrieval (feature extraction, indexing, searching, browsing, etc.) This includes things dealing with actual music, or a symbolic representation of it.

DAFX

Digital Audio Effects. (1998, 1999, 2000, 2001, 2002, 2003, 2004)

WASPAA

Workshop on the Applications of Signal Processing to Audio and Acoustics (1997, 1999, 2001, 2003)

Ongoing Public Discussion

• The MUSIC-IR mailing list - low traffic, high signal:noise

Research Groups

• Digital Music Research Network - a consortium of research groups, listing far more than I do here
• Music Technology Group at Pompeu Fabra University of Barcelona
• DSP & Multimedia Research Group at Queen Mary University of London
• MMM Group at University of Amsterdam
• Machine Listening @ MIT Media Lab
• Microsoft's Communication, Collaboration, and Signal Processing research group has done a number of interesting music-related things, including AutoDJ, which constructs music playlists from a seed of one or more songs using functions learned over metadata. They used high quality (manually assigned) metadata that common folks wouldn't normally have access to, but the technique should work with cheaper inputs.

Software

I'm only interested in Free software that runs on Linux.

• WEKA - a machine learning library in Java
• Octave - a Matlab clone, useful for working with numerical data
• FFTW - the Fastest Fourier Transform in the West
• MARSYAS - Manipulation Analysis, and Retrieval SYstems for Audio and Signals
• Maaate - a set of libraries that enable audio signal analysis and feature calculation in the compressed/frequency domain
• CLAM - C++ Library for Audio and Music
• Agnula - Linux distributions (based on Debian and Red Hat) specifically tuned for audio applications

People and projects

• HPdj
• SmartPlay attempts to learn which songs go with which other songs, but it doesn't generalize at all. Lessons learned about one pair of songs are not extrapolated to other pairs.
• Eric Scheirer has some great papers and references on non-symbolic methods of computational psychoacoustics, including a beat tracking algorithm (which is described better in his dissertation).
• Simon Dixon has also done some work on beat tracking in audio signals and musical expression inference.
• Prediction-driven computational auditory scene analysis
• Automatic Musical Instrument Identification
• Roger B. Dannenberg
• Dan Ellis
• George Tzanetakis
• Doug Eck
• Keith Martin

Publications

I can't keep up. Seek and ye shall find. I recommend checking conference proceedings.

tmp+15843@sethoscope.net