Research Activities

My main research interest is within speech processing and especially automatic speech recognition. A general summary on the activities on speech processing at our group is given here.

The last years I have mainly been involved in discriminatory methods for training speech recognizers. This includes both methods for training the 'classical' Gaussian HMM-structures, pure Neural Network based structures and hybrid variants of the two structures.
As to applications for Norwegian as a language, I am involved both in a project dealing with large vocabulary recognition (on line captioning of TV-programs) and in two research programs for developing speech based dialog systems TABOR and SPODIS. Descriptions in Norwegian can be found at TABOR_N and SPODIS_N.

Further our group has recently headed a project (founded by the Norwegian Research Council) which has ended up with a recommendation for a national Norwegian Corpus for language engineering. The report (in Norwegian) decribing the recommendations can be found here.


Last modified: Sept 25 2001 by Magne H. Johnsen mhj@iet.ntnu.no