although an afternoon
it's a problem i mean every to mister technology later to present my paper this
conference
so what the problem is a sampling to address the problem of variability in the
i-vector space you to the acoustic content of the
speech
and the language is the main one of the main now
source of this variability
the probabilistic linear discriminant analysis while going to model the kind of source variability but
it cannot some model this variability using multilingual without a multilingual apartments that's for each
speaker
so
there is a one a method called language normalized w c n
which is designed to model this variability by extending the source normalized mfcc
no
this is done before a prior to the p lda training camp
so what i am going to post is to
propose a purely training algorithm
we would be built to reduce this language in fact
so by estimating the speaker and channel subspace stuff from multilingual utterances your we can
be appealed it can be able to work independent
so when evaluated on the nist sre two thousand eight core condition
we were able to you know
reduce the fact
we use the russian spanish arabic and mandarin in addition to english
utterances
so in comparison with the baseline system we use double system was we were able
to
we choose the language effects by ten percent the equal error rates
so that's it
okay