OMRAS 2

Research Project, Tim Crawford, Michela Magas, Ben Fields, Polina Proutskova, Christophe Rhodes, Mark d'Inverno

The OMRAS2 project was designed from the outset to bring together individual tools into a system which can enable a broad range of tasks for a variety of users. The precise way in which tools are combined and their functions adjusted depends firstly on the task itself, and also crucially on the content and scale of the music collections involved, but just as much on the requirements of the end user.

With non-specialist users in mind, two OMRAS2 PhD students are working on very different approaches to the general problem of how to select ‘interesting’ music from large collections of unknown material (as, say, in an online music download service or as gathered from the web):

• Michela Magas has developed the highly intuitive MHashup browsing interface which uses OMRAS2’s audioDB to discover tracks containing music similar to a query example;
• Ben Fields is working on automated generation of playlists using on a combination of musical features with the behaviour patterns of social networks.

Tim Crawford’s own work in OMRAS2 is broadly motivated by the demands of what might be called ‘musicology’, but it also has other uses within the project, especially in evaluation (where we test the ability of the system to recognise and retrieve music against expert human judgements). One typical use is in managing small to medium specialised libraries of recorded music, such as the King’s Sound Archive (KSA) collection of several thousand digitised historical 78 rpm record sides. As well as dealing well with tasks such as searching for duplicate recordings (for example, to find a clean substitute for a badly scratched or distorted section on a record), using OMRAS2’s audioDB we can also find the various different performances of the same work (as would be required for a study of changing performance styles). The OMRAS2 web-browser interface we have devloped for such searches also allows us to search within a single track, to show an analysis of the musical structure of the work being performed, as here:

A further use of this OMRAS2 technology is of great potential use to music librarians, who face an enormous amount of tedious manual (and aural) labour in cataloguing newly-acquired recordings of classical music (which are often documented poorly, and sometimes not at all). By matching music tracks automatically against a standard ‘authority’ index, we shall be able instantly to provide library-quality metadata concerning the work performed and its composer.

In a similar vein, OMRAS2 PhD student Polina Proutskova is working on tools to aid ethnomusicologists in the task of annotating field recordings; that is, labelling sections as to whether they are vocal or instrumental, performed by single musicians or groups, and so on. Such work, essential to make the recordings into a useful online resource, is normally carried out by experts in laborious listening sessions, a problem which has long held back the use of computers in ethnomusicology.