A study on speaker and session variability in speaker recognition of Chinese whispered speech

A new approach to the problem of estimating the hyperparameters which define the speaker and session variability model in joint factor analysis is applied to Chinese whispered speech. Especially, the proposed technique decoupled estimates the speaker space and channel space. The aim of this paper is to compare different recognition methods, which used in speaker recognition of Chinese whispered speech. Experiments are carried on our own database. This corpus consists of 100 target speakers, 80 male and 20 female, in which each speaker is recorded over 8 typical channels. Compared with MAP, Feature Mapping + MAP and SMS systems, the proposed JFA technique does provide superior performance and significant speedup.