JFA modeling with left-to-right structure and a new backend for text-dependent speaker recognition | IEEE Conference Publication | IEEE Xplore