Musical Mosaicing

By Aymeric Zils and François Pachet.

In the Proceedings of the COST G-6 Conference on Digital Audio Effects (DAFx 2001), pp. 39–44, Limerick, Ireland, December 2001.

Read the paper (PDF)

About the paper

At DAFx 2001, Aymeric Zils and François Pachet pioneered a constraint-based approach to composing with recorded samples and coined the term musaicing, short for musical mosaicing. Inspired by photographic mosaics, they let a composer specify the properties of the whole sequence and had the system choose and order the individual samples.

Musical mosaicing builds new audio sequences from recorded samples. The composer specifies properties for the result, such as a melody, a rhythmic pattern, or continuity of timbre, and the system selects and orders samples to match.

Abstract summary

Zils and Pachet express the desired properties as constraints over descriptors of audio samples and use local search to assemble a sequence. The paper illustrates the method with rhythmic and melodic examples and reports using sample collections of more than 100,000 items.

Follow-up research

Audio Analogies (Simon, Basu, Salesin, and Agrawala, ICMC 2005) cites the 2001 paper and uses recorded performances to synthesize a new performance. Later work includes Let It Bee (Driedger, Prätzlich, and Müller, ISMIR 2015), which uses nonnegative matrix factorization, and The Concatenator (Tralie and Cantil, ISMIR 2024), which takes the approach into real-time performance.

Related work

The Musaicing project page collects the audio examples. Jamming with Plunderphonics describes a related use of concatenative synthesis in interactive performance.

Back to Publications