Improved phonotactic analysis in automatic language identification

This paper presents a method for phone-dependent weighting within phonotactic models in automatic language identification. Based on statistical analysis of the phonetic-recognizer behaviour, a phone confidence measure is derived and used to weight the bigram probabilities during testing. The confidence corresponds to the expected decoding stability of individual phones. The proposed method was shown to improve the system performance consistently on a three-language task. The best improvement of the error rate was from 8.4% to 1.8% for the 45-second utterances.