Language Recognition in iVectors Space

Acoustic Language Recognition, iVectors, Joint Factor Analysis

We have introduced a novel approach for language recognition. Three classifiers (linear generative model, SVM and logistic regression) have been tested in the iVector space, and all outperform the state-of-the-art JFA system. Very simple and fast classifier based on linear generative model provides excellent performance over all conditions. The advantage of this classifier is also its scalability: addition of a new language only requires estimating the mean over the corresponding iVectors. Most of the computational load is in the iVector generation. Hence, as a next step, we will try to obtain iVectors from the utterances and the corresponding sufficient statistics in a more direct way.

The concept of so called iVectors, where each utterance is represented by fixed-length low-dimensional feature vector, has recently become very successfully in speaker verification. In this work, we apply the same idea in the context of Language Recognition (LR). To recognize language in the iVector space, we experiment with three different linear classifiers: one based on a generative model, where classes are modeled by Gaussian distributions with shared covariance matrix, and two discriminative classifiers, namely linear Support Vector Machine and Logistic Regression. The tests were performed on the NIST LRE 2009 dataset and the results were compared with stateof- the-art LR based on Joint Factor Analysis (JFA). While the iVector system offers better performance, it also seems to be complementary to JFA, as their fusion shows another improvement.