Robust text-independent speaker identification using short test and training sessions

In this paper two methods for noise-robust text-independent speaker identification are described and compared against a baseline method for speaker identification based on the Gaussian Mixture Model (GMM). The two methods proposed in this paper are: (a) a statistical approach based on the Generalized Gaussian Density (GGD), and (b) a Sparse Representation Classification (SRC) method. The performance evaluation of each method is examined in a database containing twelve speakers. The main contribution of the paper is to investigate whether the SRC and GGD approaches can achieve robust speaker identification performance under noisy conditions using short duration testing and training data, in relevance to the baseline method. Our simulations indicate that the SRC approach significantly outperforms the other two methods under the short test and training sessions restriction, for all the signal-to-noise ratios (SNR) cases that were examined.