Language Recognition with Language Total Variability

In this paper, we try to introduce the idea of total variability used in speaker recognition to language recognition. In language total variability, we propose two new recognition systems, Language-independent total variability recognition system (LITV) and Languagedependent total variability recognition system (LDTV). Our experiments show that language-independent total factor vector includes the language dependent information, what's more, language-dependent total factor vector contains more language dependent information. These two systems LITV and LDTV can achieve performance similar to that obtained with state-of-the-art approaches. Experiment results on 2007 National Institute of Standards and Technology (NIST) Language Recognition Evaluation (LRE) databases show LDTV gains relative improvement in Equal error rate (EER) of 23.2% and in minimum Decision cost value (minDCF) of 14.2% comparing to LITV in 30-second tasks, and we can obtain further improvement by combining these two new systems with state-of-the-art systems. It leads to relative improvement of 21.1% in EER and 23.1% in minDCF comparing with the performance of the combination of the MMI and the GMM-SVM systems.