Automated generation of semantic metadata from music in TV and radio programmes

Aims
• To create a new interdisciplinary collaboration across Salford University between Music Composition and Acoustic Engineering, strengthening the Media, Digital Technology and the Creative Economy Theme.
• To build relationships with BBC Archives Research and Development (a part of the BBC which isn’t part of the new BBC Audio Research Partnership), so we can bid to external funders on a related project concerning sound identification.
• To strengthen our research in metadata, which has been highlighted as a key area of research by funders such as the TSB.
• To develop new audio signal processing algorithms which can create impact through high quality academic publications, and exploitation by BBC Archives or through commercialisation.

There is a need to produce metadata from TV and radio programme content using automatic techniques. Currently, there is great interest in semantic metadata for indexing and searching, for instance the BBC is looking to automatically analyse audio and video content, extracting key semantic themes and ideas to allow programme classification and clustering. The DTA student would look at analysing music within programmes to ascertain what semantic or emotive information (e.g. happy, sad, exciting, calm) can be extracted via Music Information Retrieval (MIR) techniques. Music is well documented as setting or heightening semantic intent within broadcast programmes. By extracting these semantics much richer metadata would be created, allowing for better classification and searching for viewers. Trevor Cox has recently completed a short pilot project with the BBC on Theme Music, the proposed project will extend the work by looking at the use of music more widely within programmes.

For audio processing, initially music needs to be identified within the programme and separated from competing sounds. A good starting point is the previous work done at the BBC Audio R&D exploiting zero-crossings. Then the music’s key features will be extracted e.g. tempo, tonality. The audio features will give basic information about the content of the programme music. They will to be used to identify the music’s semantic concepts using an experiential based learning algorithm, such as a support vector machine. This will enable an automated semantic metadata extraction system for music.

Although computer scientists have studied automatic extraction of semantics from audio before, the use of knowledge from music is surprisingly lacking from much of the work, and this gives great potential for the proposed collaboration between an acoustic engineer and a music composer to break new ground. Furthermore, music within TV and radio has largely been ignored by academia and consequently new research and publications in musicology can be achieved.

How to apply:
The application form should be completed online at:
http://www.salford.ac.uk/study/postgraduate/how_apply/

The closing date for applications is 30th September 2011

Enquiries should be made to Catriona Barkley [Email Address Removed]

For detailed information about any of these projects please contact the named supervisor(s).