This is a 5 minute presentation I gave on 18th December 2007 at the DMRN+2 workshop. Slides and some very rough speaking notes are provided, give a shout if there's anything you'd like more information about.

A very brief overview of how OMRAS2 is using RDF and semantic web technologies

0 - OMRAS2 + RDF

• Taking a resource-oriented rather than service-oriented view
• => Focus on suitable description formats rather than wrapping things as web services
• Using W3C's RDF (Resource Description Framework) technology
• which allows very versatile interlinking of different data sources
• There is an active community in the Linking Open Data project, seeking to bootstrap this web
• A lot of the music-related stuff has been done by Yves Raimond, here at QM

So I'll talk about a few of the sources of data we might be bringing into this web :

1 - Music Ontology

• Community effort, used mainly for traditional editorial music information
• eg. Wikipedia (via DBpedia), Musicbrainz (via Zitgist)
• Actually more expressive than these datasets show (eg. John Peel sessions, talking about different performances and recordings of a particular work, and the way they were played on the radio)
• Various translations from RDBs, XML, etc. to RDF

2 - RDF + Vamp

• We can also add audio feature data to this web
• So taking Vamp, an existing plugin format
• An RDF-aware Vamp host can generate RDF data from running a Vamp plugin
• This data can be directly interlinked with other heterogeneous data
• We're still working on how best to describe the output, but we've had some useful experimenting which I'll come to in a minute

3 - Chord Ontology

• As Matthias showed
• Human or automatic transcription generates RDF data
• This lives alongside the editorial text data, or the audio feature data, all interlinked to the same underlying Signal or Track resource

So now let's look at how we might start using this music web of data

4 - GNAT + GNARQL

• GNAT : Small tool which finds URIs for each track in your local music collection
• (Using fingerprinting or metadata-based search)
• GNARQL : Local music information store, loads in GNAT's results and aggregates additional info
• Can then do queries like :
• Find me all tracks produced by somebody who's worked with Pink Floyd
• Find me music by artists from the Eastern hemisphere (because linked with geonames)

And another little application :

5 - Timeline widget combining heterogeneous data

• Take :
  • human chord transcription in RDF
  • output from an RDF-aware vamp host running an onset detection plugin
  • Also Vamp-produced key estimation
• Since they're all attached to the same timeline resource, we can simply combine the RDF data and pass the file to a timeline visualiser widget
• This could be incorporated into a GNARQL interface, to display all known temporal information about a track