INTERSPEECH 2006 - ICSLP
This paper deals with unsupervised model adaptation for speaker recognition. Two adaptation schemes are proposed, the first one is based on a test by test model adaptation and the second one proposes a batch mode, where the adaptation is performed using a set of tests before computing the decision score for each of them. The experiments are conducted thanks to the NIST SRE 2005 database. This paper shows clearly the interest of unsupervised model adaptation when enough test data is available (batch mode) and the intrinsic difficulty of an online (test by test) adaptation mode.
Bibliographic reference. Preti, Alexandre / Bonastre, Jean-François (2006): "Unsupervised model adaptation for speaker verification", In INTERSPEECH-2006, paper 1554-Wed3CaP.4.