Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Implementation of a live dialectal media subtitling system

: Stadtschnitzer, Michael; Schmidt, Christoph Andreas

Fulltext (PDF; )

International Speech Communication Association -ISCA-:
Speech beyond speech towards a better understanding of the most important biosignal. Vol.2 : 16th Annual Conference of the International Speech Communication Association (INTERSPEECH 2015); Dresden, Germany, 6-10 September 2015
Red Hook, NY: Curran, 2016
ISBN: 978-1-5108-1790-6
International Speech Communication Association (Interspeech Annual Conference) <16, 2015, Dresden>
Conference Paper, Electronic Publication
Fraunhofer IAIS ()

Subtitling is a useful technique to fulfil the information needs of deaf and hearing impaired people. Live subtitling is needed especially for live events and is not restricted to television, but can also be provided to persons on site, e.g. to a deaf politician during a parliamentary debate. Live subtitling is demanding since the audio information has to be transformed into text within a few seconds. In this demonstration we present the Fraunhofer IAIS audio and video live subtitling system for standard German and Bavarian. The system was developed in the “Live-Caption” project and consists of an online speech recognition and speaker diarisation system. We employ our entire large annotated standard German broadcast corpus for training. In addition, we apply the system to Bavarian dialect by adapting the acoustic models and the pronunciation lexicon, exploiting a Bavarian media corpus. Due to the real-time restrictions and the spontaneous character of dialectal speech, the system performance is far from perfect, but encouraging.