BBC R&D

Abstract

This paper describes work which is exploring how speech recognition can be used in programme production. It examines the performance of large vocabulary, speaker independent, speech recognition in a broadcast context and illustrates this with results showing the performance likely to be achieved under these conditions. These show how the variety of speech and audio qualities to be expected in broadcast programmes makes this a challenging task but nevertheless, performance adequate for a range of applications can be achieved.

Practical applications for the technology are described. Some of these are already demonstrating significant benefits, whilst others suggest completely new ways of working will be possible as this technology improves further.

It is concluded that speech recognition is now able to deliver worthwhile results if applications are considered carefully and that, over time, its introduction will make a more widespread impact.

Given at the International Broadcasting Convention (IBC 2000), Amsterdam, 7-12 September.