have run some to learn from university of east and women today same here two
presents the last it binds
in using the planning phone number and its applications
so how would you simple we chuckle reflects a hand graph feature in
in direct approach
by our deep-learning that will
so what we use be plantings first
you given any feature itinerary in use a added to considering so we try to
understand how the effect of convolution on the signal by reconstruct the signal from the
art is no messy go very small print a tree trees so signal for via
kind of read effects and fall be but
become comparison is like glow the signals
and yes the
subsets in speech recognition this
or you modify must in
frame within we depending
and next week
deep-learning in bottleneck feature we improve the performance of language identification system a lot
so this is a forty size of i was system
is the end-to-end approach from audio file we extract
no mfcc or filter bank feature and feeding to the network we custom probabilities for
each language
so far
jennings this deep network
well we need to address how overnight in well how to better if it from
multiple watching take the design how to change the artistic the
efficiently using early stopping regularization
am optimization techniques
and how to address the computation noise you with deep network
and this is the results so we get improvement compared to single system and our
republic system that improvement
by using more advanced technical and what's normalization dropouts
and fall we sees as the imbalanced dataset happen or negative interface i on the
model so we use modify the corpus and using by just imposing to be
or sampling and score calibration
we closer to the bottleneck feature but
still plenty of room for improvement