What is omras2?
Online Music Recognition And Searching
Ontology-driven Music Retrieval & Annotation Sharing service
omras2 is a framework for annotating and searching collections of both recorded music and digital score representations such as MIDI. The project is funded by the EPSRC.
omras2 defines two core standards for annotating music:
- a set of extensible ontologies allowing data to be shared over the web and linked to existing resources;
- a plugin interface for audio feature extraction and annotation modules.
Numerous annotation modules are already available as Vamp plugins. The Vamp interface can support a wide range of outputs, from low-level audio features (spectrograms, etc.) to high-level musical annotations (tempo, key, musical structure, time-alignment with other performances, etc.). A development kit is available making it easy to wrap any C or C++ code as a Vamp plugin, with Python compatibility (VamPy) and also further enhancements in the pipeline.
Sonic Visualiser is omras2’s powerful desktop application for running Vamp plugins on individual tracks, and for viewing the output or enhancing it with manual annotations. Sonic Annotator is a corresponding command line tool enabling you to run plugins easily on all files in a collection. Both applications are cross-platform and open source. Sonic Annotator allows Vamp output to be saved in a variety of formats, in particular as RDF using the Audio Features Ontology. A Web Application (SAWA) is available for researchers to familiarise with OMRAS2 technologies. SAWA provides online audio feature extraction services using Vamp plugins. omras2’s Harmonic Visualiser is a new Windows application that allows the visualisation and editing of individual notes within a polyphonic spectrum.
Data is shared in omras2 as RDF, which can be published very simply to the web and accessed directly over http, or stored in a triple store (an RDF database). RDF data can be browsed just like conventional HTML web pages, as RDF documents are hyper-linked in a similar manner. The omras2 application GNAT can be used to scan all audio files in a collection and generate an RDF document linking them to existing information on the web. Used in parallel with Sonic Annotator, this enables you automatically to create annotations for every track in your collection and to link them to high-quality editorial information such as the MusicBrainz recording catalogue. To make it straightforward to use RDF annotations in your own code, we provide tutorials on using standard RDF client libraries in C++, Java and Python.
omras2 music retrieval software currently includes two audio content-based search engines. audioDB is a powerful frame-level search engine which can find matching passages of music in large audio collections. SoundBite is a lightweight self-contained playlist generation engine.
Ongoing research in omras2 includes:
- the deployment of “Knowledge Machines” – machines serving RDF over the web which are able to do on-the-fly audio processing and annotation using Vamp plugins; and
- the development of logic-based inference engines to create new knowledge from an existing web of RDF data, for example by inferring rules of harmony from information about individual chords.
Some annotation modules are already available as dynamic RDF services, including a chord-labelling service that works with MIDI input.