Aarhus Universitets segl

Computational Music Analysis and Generation

Lecturer David Meredith (Aalborg University, Musik Informatics and Cognition Research Group)

Oplysninger om arrangementet

Tidspunkt

Torsdag 11. oktober 2018,  kl. 16:15 - 18:00

Sted

Kasernen, Langelandsgade 145, bygn. 1584, lokale 126, 8000 Aarhus C

The recent work of the Music Informatics and Cognition (MusIC) Research Group at Aalborg University will be presented. The group has developed computational models and methods that aim to simulate (and explain) human performance on a variety of musical tasks, including composition, motivic/thematic analysis, tonal analysis, composer and style discrimination and the perception of rhythmic structure (metre, syncopation and grouping). After giving an overview of the group’s work in these areas, three topics will be presented in more depth: (1) computationally modelling Milton Babbitt’s compositional processes, (2) convolutional methods for music analysis, and (3) geometric, compression-based pattern discovery in music. The group’s work on modelling Milton Babbitt’s compositional processes resulted in the first ever successful automatic generation of a completely novel all-partition array. This array was then used to automatically generate a novel piece in the style of Babbitt. The piece was premiered in London in 2016 and is featured on the front cover of the Spring 2016 issue of Computer Music Journal. The group has also developed convolutional methods (e.g., deep learning methods such as convolutional neural networks) for music segmentation and classification, that have been successfully applied in automatically distinguishing between composers (e.g., Mozart and Haydn) with an accuracy that matches human experts. Finally, a number of specialised compression algorithms have been developed that are capable of discovering repeated patterns in music that often correspond to the motives and themes identified by human analysts and listeners. The compact encodings generated by these algorithms have also been successfully used, in combination with a compression-based similarity metric, for various music classification tasks such as categorising folk songs into tune families.