Speaker recognition using speakerspecific text

Abstract

In any classification task the error rate is substantially related to newlinethe commonalitysimilarity among different classes In conventional GMMbased newlinemodeling technique since the model parameters of a class are newlineestimated without considering other classes in the system features that are newlinecommon across different classes may also be captured along with unique newlinefeatures In other words out of class data is not used to adjust the model newlineparameters that may lead to poorer performance of the classifier Further in newlineconventional GMMbased classifier the performance is to a greater extent newlinedirectly proportional to the amount of data used during testing which is newline newline