Analysis and Optimization of Bottleneck Features for Speaker Recognition

Poster Session 2: Speaker Recognition I

Alicia Lozano-Diez, Anna Silnova, Pavel Matejka, Ondrej Glembek, Oldrich Plchot, Jan Pesan, Lukas Burget, Joaquin Gonzalez-Rodriguez

an initial draft and that would like to well

a given introduction to the to the work done by olivier models on all of

the s two cannot be here

and try to be quick

so of the work is about some analysis homework a sorry forgot the title skull

analysis an optimization of bottleneck features for speaker recognition

and it's basically are about features that are used as an input to the to

the widely used bottleneck features

so first we started by using the of the sr features and try to optimize

the network to give the best phoneme recognition for phoneme error rate basically

and what we see that it is that she try to put in the features

that we know there are best for these speaker recognition which is the mfccs plus

some extra normalisation plus some extra constraint on the on the neural network training et

cetera so the conclusion is that the asr features

are not the best for of speaker recognition using the bottleneck ubm i-vector system

and better phone accuracy does not necessarily mean a better speaker error rates so thank

you once the poster