Speech Technology: Television broadcasts are automatically transformed to text, enabling content search

zurück

Pressemitteilung - ( 27.10.2009 )

Speech Technology: Television broadcasts are automatically transformed to text, enabling content search

Collaboration between “Bayerischer Rundfunk” (Bavarian Broadcasting Corporation) and European Media Laboratory - automatic transcription of excerpts from the TV series “BR-Alpha Forum” – “It´s not a matter of artificial intelligence but of intelligent speech technology” says Siegfried Kunzmann, head of R & D at EML

Heidelberg/Munich. Automatic transformation of spoken language into text (transcription) has made great progress over the past years, due to continuously improved recognition algorithms, increased computer performance and the possibility of processing huge amounts of data. Speech technology experts of the European Media Laboratory (EML) have developed a transcription system which is able to automatically transform TV broadcasts into text. Before using this system, it has to be adapted to the application area, the so-called language domain.

In a new joint project with the “Bayerischer Rundfunk” (Bavarian Broadcasting Corporation, BR) the system is being adapted to the programme “BR Alpha Forum”. “BR Alpha Forum” is a daily broadcast in which experts from politics, economy, science religion, and culture give 45-minute presentations. Using some 100 recorded presentations, the transcription system will be trained. By processing both the spoken words and the accompanying manuscripts it “learns” the vocabulary of such broadcasts and the proper pronunciation. Based on this, it will be able to “understand” future presentations in real time.

“It is not a matter of artificial intelligence, but of intelligent speech technology,” explains EML R&D manager Dr. Siegfried Kunzmann. “We train the speech recognition system with large volumes of text and voice recordings of the respective domain.” Recognition errors are fixed by grammatical analysis ,and non-spoken punctuation marks are added. The resulting transcribed text is well-structured and of high accuracy..

The aim of the pilot project between EML and BR is to automatically transcribe the “BR Alpha Forum” conversations and to transform them into manuscripts. “Furthermore, the retrieval and replay of broadcasts in the video archive via content search becomes possible,” says Herbert Tillmann, technical director at BR. Eventually, it will be possible to show a transcribed text in a web browser or as video broadcast with subtitles.
The project was commissioned by the non-profit Klaus Tschira Foundation and will take six months.

For further information please contact:

Dr. Peter Saueressig
Presse- und Öffentlichkeitsarbeit
European Media Laboratory GmbH
Tel.: +49-(0)6221-533245
Peter.Saueressig@eml.org

Rudolf Küffner
BR-Pressestelle
Bayerischer Rundfunk
Tel.: +49-(0) 89-5900-2176
rudolf.kueffner@brnet.de

Dr. Peter Saueressig
European Media Laboratory GmbH
Schloss-Wolfsbrunnenweg 33
69118 Heidelberg
Telefon: 06221-533-245
Fax: 06221-533-198
Email: peter.saueressig (at) eml.villa-bosch.de