< Back to previous page

Project

SPRAAK2TAAL: extraction of tension and idelological context from political speeches - research on the integration of speech and language processing.

The SPRAAK2TAAL project will investigate how speech recognition (e.g. automatic transcription of interviews) and a language module (e.g. automatic summarization) can optimally cooperate in the Clarin framework. The chosen case study will investigate the extraction of tension and ideological context from political speeches. This choice is based on feasability (well articulated speech, independency from yet to be developed Clarin modules, a language module capable of using all modalities of a speech recognizer), interest from the humanities and extensibility of the methods. The SPRAAK2TAAL project will: - investigate the usefulness for the language module of the different types of output produced by the speech recognizer (best sentence, sentence with alternatives or word graph); - adapt the SPRAAK speech recognizer to this end, paying attention to conformity of the input/output with Clarin standards; - automatically adapt the language model of the speech recognizer to the language type (political speeches) by using latent probabilistic models - a method that is used both in language and speech processing.
Date:1 Oct 2010 →  30 Sep 2012
Keywords:Automatic summarization, Speech recognition
Disciplines:Computer hardware, Computer theory, Scientific computing, Other computer engineering, information technology and mathematical engineering, Artificial intelligence, Cognitive science and intelligent systems, Modelling, Biological system engineering, Signal processing