In cooperation with NTNU, SINTEF is running an initial project with the aim of developing a general large vocabulary speech recognition system for Norwegian. The project is funded by the Norwegian Research Council.

We are using previously proprietary speech, text and lexical resources which now have been made available. The development is based on freely available source code, thereby providing a good foundation for further research on large vocabulary speech recognition for Norwegian.

Today, there is no such open system available for Norwegian. Also, commercial systems do not exist for general dictation in Norwegian. Systems exist for specific and limited domains such as hospital usage, however these systems are closed and therefore not suitable for further research and development in the open domain.

There are many applications of this technology, where dictation is a prominent one. Also within the telecommunication area (e.g. self service terminals), dialogue systems (e.g. the BRAGE-project), media database search, and means for the disabled the applications are widespread. An example of the latter is subtitling of live TV programs for the hearing impaired.

Even within limited domains where the user is allowed to speak using natural language, large vocabulary speech recognition has a great potential.

The main challenges are found in the development of acoustic and statistical language models for Norwegian in order to obtain high recognition scores for real time applications.

Our goal is to end up with a first version of an online demonstrator during the first quarter of 2008 with a vocabulary of more than 20000 words for general dictation in Norwegian.

Contact:
Erik Harborg
Tel.: +47 73 59 31 39