NEON: Dutch subtitling.

There is a strong need for subtitling Dutch spoken TV programmes. User groups are pushing broadcasters and governmants to increase the amount of subtitled material. Currently VRT subtitles 65% of its Dutch spoken programmes, with the promise to grow to 95% by 2010. Likewise, NPO will have to subtitle 95% of its broadcasts in 2010 the latest. This huge increase in amount of programmes that need to be subtitled requires large investments in new technologies and additional manpower. One of the technologies that come to help is speech recognition. Experiments with speech recognition for subtitling have been carried out since the 1990's. Since then technology has evolved quickly and currently it is being used routinely by several broadcasters, in the form of "respeaking" (where the subtitle is spoken by a trained person). In this project a more advanced and less labour intensive application of speech recognition for subtitling will be implemented, namely alignment of existing texts or scripts with spoken audio. The alignment produces the exact timing of the spoken audio such that subtitles can be generated at the right time instances. After alignment, the aligned text is used as scuh, or is further condensed into shorter sentences that are better suited to serve as subtitles. The incoming audio stream is also segmented and clusterd into different speakers, such that a visible indication onf the screen of speaker identity is possible (e;g. by using coloured subtitles). This will lead to a demonstrator for (semi) automatic subtitling in Dutch. A speech recognizer is used at this end, which means that in the background ans as a fallback position, the speech recognition result as a direct transcription of the audio stream is always available. The envisaged demonstrator will allow the content providers who already use respeaking to evaluate the more advanced use of speech recognition.

Keywords:Speech and text alignment, Automated subtitling, Speaker segmentation, Speech recognition

Disciplines:Linguistics, Theory and methodology of linguistics, Other languages and literary studies, Modelling, Biological system engineering, Signal processing