Yves Raimond, Chris Lowis, Roderick Hodgson, BBC R&D
Jonathan Tweed, MetaBroadcast
Demo Track, WWW, Lyon
18 April 2012

The BBC archive

Increasing the value of archives by interlinking

Automatically extracting topics from speech audio and identify them with Linked Data URIs

Workflow

• Automated speech recognition
• Term identification
• Term disambiguation
• Ranking

More details about each step in our LDOW paper

The World Service archive

• Around 70,000 programmes
• Covering 6 decades
• ~ 3 years of continuous audio
• Sparse metadata
• ~ 500 TB of content

Processing the World Service archive

• Workflow steps isolated in individual workers
• Computation-intensive workers on Amazon Web Services
• All managed by Message Queues and an API centralising the data
• Only bottleneck: bandwidth to Amazon's servers

Some examples

Thank you!