For several years the BBC has been using speaker-independent speech recognition to assist television subtitling and dramatically reduce the time taken for a skilled subtitler to complete a programme. This paper describes recent developments in this technology. Our new system for robust generation of subtitles for live television is also introduced. This system uses speaker-specific speech recognition and novel networking software to allow subtitlers to work from home or other remote locations whilst permitting straightforward handovers between subtitlers or television services.

This was an invited paper published by the Technological Advancement Organization of Japan in the Proceedings of the TAO Workshop on Universal Design for Information, Communication and Broadcasting Technologies, Tokyo, 7 June 2003.

Gratefully dedicated to David Kirby.