Question: How to create “interesting” audio music sequences from large collections of sound samples? The team invented the notion of musacing, by analogy with mosaicing, and explore new forms of interaction with music catalogues, such as song sampling.

The idea of musaicing is a transposition of the notion of image mosaicing to the world of audio. Musaicing makes intensive use of large databases of audio samples, and allows user to create music without having to perform the tedious and difficult task of listening and selecting individual samples. Musaicing consists in creating automatically large databases of samples by segmenting existing songs. Then metadata is computed for each sample to describe it in terms of perceptive parameters (such as timbre, percussivity, energy, pitchness, etc.). Finally the user can express high-level constraints to specify the structure and nature of a target sequence of samples. Constraints can be of various types, such as continuity (produce a sequence of samples which are continuous, timbre-wise), distribution (select a percussive sample every beat with tempo = 120) or cardinality (include at least 40% of samples which come from a Beatles song), or any combinations of these.

Media

• Interface screenshot
  

Participants: François Pachet, Jean-Julien Aucouturier, Aymeric Zils

Tags: Music

References

Aucouturier, J.-J., Pachet, F. and Hanappe, P. From Sound Sampling To Song Sampling. Proceedings of the International Conference on Music Information Retrieval (ISMIR), pages 1-8, Barcelona, Spain, 2004. University Pompeu Fabra

Zils, A. and Pachet, F Musical Mosaicing. Proceedings of the COST G-6 Conference on Digital Audio Effects (DaFx-01), pages 39-44, Limerick, U.K., December 2001. University of Limerick

This entry was posted on Saturday, December 15th, 2001 at 9:09 am and is filed under Past Projects.