Source-Normalized LDA for Robust Speaker Recognition Using i-Vectors From Multiple Speech Sources | IEEE Journals & Magazine | IEEE Xplore