³ÉÈËÂÛ̳

For about a year now, ³ÉÈËÂÛ̳ R&D has been running a project called Multimedia Classification. The aim of this project is to look at new ways of classifying the archives by creating what is known as semantic metadata.

In many archives, including the ³ÉÈËÂÛ̳’s, each programme has an entry in a database which contains information about the programme. This database information is known as metadata – information about data. Currently, much of this metadata in the archives is factual, details on who was in the show, when it was transmitted and a genre classification, using a system called LONCLASS which is derived from the Universal Decimal Classification system (a more advanced and powerful version of the well known Dewey Decimal system often used in libraries). This metadata is useful for what is was designed for – to allow editors and producers to find specific clips and specific programmes in the archives.

The Multimedia Classification project is creating new metadata to allow members of the public to navigate the digitised archive. One of the ways to create metadata is to automatically analyse a TV programme – for example; finding out what people are talking about; how they are speaking (are they angry, happy, upset?); are there any exciting or dramatic scenes. Effectively we are looking at what is happening in the pictures and sound of a programme and trying to work out what this means. This is known as a semantic analysis, and the semantic information about a programme we create is the semantic metadata.

One of the aspects of the audio that we are looking at is the music within a programme – in particular the theme tune which is where Musical Moods comes in. We are using automatic techniques to analyse a theme tune to find different aspects of it – for example the key the music is in, what the tempo is and what the instrumentation is like. However, in order to be able to make sense of these we need people to tell us what they think of the theme tunes; are they exciting? Is the music heavy? Is the music happy?

When we’ve got public opinion from the Musical Moods experiment, we can train computers to identify which aspects of a theme tune relate to mood. Along with other analysis techniques of the programme video and other audio (speech, sound effects) we can improve our semantic metadata. Eventually, this could help people navigate through the ³ÉÈËÂÛ̳ Archive as its content is gradually made available online. We hope that Musical Moods is a step toward helping our audiences find ³ÉÈËÂÛ̳ programmes they never knew existed and they never knew they liked.

and another if you want to watch the video we have made about the project.