Abstract

In this report we will describe an experiment to test whether we could improve the acoustic model of VIOS by augmenting it with domain knowledge found in the OVR log files of one year. For this we simulated the behavior of VIOS by a Lernout & Hauspie speech recogniser. Although the experiment shows an improvement when the system is augmented with domain knowledge, the improvement was not as big as was hoped for. Nevertheless, there still is a clear improvement. Fine adjustment of the parameters involved, although computationally demanding, gives plenty of opportunity for further improvements.