0:00:16so we are presenting here our work using what we can have a four gram
0:00:21using require program known and of course for language identification
0:00:28where you know what we it doing with a regular and recurrent neural network is
0:00:32to use phonemes as input to see now when i think with indication
0:00:37begin and that of the number of phonemes
0:00:39and then we have also incorporated in the context information use in a uniform slide
0:00:44function trigrams
0:00:45comparing them and their fusion all of them
0:00:48and so we are proposing the concatenation of this in a descent phonemes in our
0:00:53in our system
0:00:56so this architecture apply to this language and the via some system
0:01:01is based on phonotactic system i prepare landmark detection
0:01:05so we have for its phonetic recognisers in the bruno recognizers we obtain a sequence
0:01:11of phonemes
0:01:12and in evaluations for each utterance we a compute like an entropy metric provided by
0:01:17their like the network
0:01:18and this entropy scores are calibrated than used later
0:01:23we also present a word but don't with funded hubert representations a it used in
0:01:29order to reduce the vocabulary in this a neural network using k-means to group a
0:01:35similar from grants
0:01:37and we have were with this keeper model at the phoneme level and we had
0:01:40a relative improvement of seven percent
0:01:43hence the despicable to read the text
0:01:46also in the work we present like the study of the most role of and
0:01:50their right brown report in that no one of course parameters
0:01:54so here is the list of parameters have been working with
0:01:59here in their results we can see this you have rate in our database used
0:02:04in comparing the nice ones the diphones triphones and then we can see a the
0:02:09fusion of them and the comparison with the work these landed pprlm
0:02:14and a fusion with that and the standard acoustic system based on mfccs
0:02:20c different portions finally where we can see that a
0:02:24this approach also provides complementary information so there are a final improvements in our global
0:02:31that's it