Abstract
For many practical applications of speech recognition systems,it is desirable to have an estimate of confidence foreach hypothesized word, i.e. to have an estimate of whichwords of the output of the speech recognizer are likely to becorrect and which are not reliable. We describe the developmentof the measure of confidence tagger JANKA, which isable to provide confidence information for the words in theoutput of the speech recognizer JANUS-3-SR. On a spontaneousgerman human-to-human...

This publication has 6 references indexed in Scilit: