The usage of independent component analysis for robust speaker verification

This study employs independent component analysis (ICA) subspace feature selection for the robust speaker verification (SV). ICA subspace provides statistically independent basis that spans the same space and preserves the Euclidean distance measurements. These independent components are applied to a vector quantizer (VQ) SV system. In the feature space modification stage, a batch-mode FastICA algorithm and two adaptive algorithms EGLD-ICA and Pearson-ICA are employed for two-microphone case. As a result, the feature space is modified by a choice of independent component basis to obtain a lower classification error and a better generalization in real environments. The performance of the approach is demonstrated with YOHO database in various noise cases.