EXPERIMENTS WITH LVCSR BASED LANGUAGE IDENTIFICATION
        Tanja Schultz, Ivica Rogina, Alex Waibel
                  <tanja@ira.uka.de>

             Interactive Systems Laboratories 
             University of Karlsruhe (Germany)
             Carnegie Mellon University (USA)
                 
                  published at: SRS 95

Automatic language identification is an important problem in 
building multilingual speech recognition and understanding 
systems. We have developed a front-end LID module based on 
LVCSR to identify English, German, and Spanish language for 
use in spontaneous speech-to-speech translation.

We studied the constitution of different levels of knowledge 
to identify a language, i.e. the phonetic, phonotactic,
lexical, and syntactic-semantic knowledge.

A comparison of LID systems using different levels of 
these knowledge sources is presented. We showed that the 
incorporation of lexical and linguistic knowledge leads to a 
reduction of the language identification error by up to 50%.