A global covariance matrix based principal component analysis for speaker identification

Speaker recognition using PCA-based feature transformation

“…as Gaussian Mixture Models (GMM)s can be fitted assuming diagonal covariance matrices (Rao and Koolagudi, 2013). This method of defining global principal components was proposed for dimensionality reduction in speaker identification using the GMM-UBM system in (Seo et al, 2009). It was also found to outperform concatenation of features by Sarkar et al (2014).…”