Fast speaker adaptation using triple diagonal and shared block diagonal transform matrices | IEEE Conference Publication | IEEE Xplore