| 0:00:16 | so we are presenting here our work using what we can have a four gram |
|---|
| 0:00:20 | units |
|---|
| 0:00:21 | using require program known and of course for language identification |
|---|
| 0:00:27 | so |
|---|
| 0:00:28 | where you know what we it doing with a regular and recurrent neural network is |
|---|
| 0:00:32 | to use phonemes as input to see now when i think with indication |
|---|
| 0:00:37 | begin and that of the number of phonemes |
|---|
| 0:00:39 | and then we have also incorporated in the context information use in a uniform slide |
|---|
| 0:00:44 | function trigrams |
|---|
| 0:00:45 | comparing them and their fusion all of them |
|---|
| 0:00:48 | and so we are proposing the concatenation of this in a descent phonemes in our |
|---|
| 0:00:53 | in our system |
|---|
| 0:00:56 | so this architecture apply to this language and the via some system |
|---|
| 0:01:01 | is based on phonotactic system i prepare landmark detection |
|---|
| 0:01:05 | so we have for its phonetic recognisers in the bruno recognizers we obtain a sequence |
|---|
| 0:01:11 | of phonemes |
|---|
| 0:01:12 | and in evaluations for each utterance we a compute like an entropy metric provided by |
|---|
| 0:01:17 | their like the network |
|---|
| 0:01:18 | and this entropy scores are calibrated than used later |
|---|
| 0:01:23 | we also present a word but don't with funded hubert representations a it used in |
|---|
| 0:01:29 | order to reduce the vocabulary in this a neural network using k-means to group a |
|---|
| 0:01:35 | similar from grants |
|---|
| 0:01:37 | and we have were with this keeper model at the phoneme level and we had |
|---|
| 0:01:40 | a relative improvement of seven percent |
|---|
| 0:01:43 | hence the despicable to read the text |
|---|
| 0:01:46 | also in the work we present like the study of the most role of and |
|---|
| 0:01:50 | their right brown report in that no one of course parameters |
|---|
| 0:01:54 | so here is the list of parameters have been working with |
|---|
| 0:01:59 | here in their results we can see this you have rate in our database used |
|---|
| 0:02:04 | in comparing the nice ones the diphones triphones and then we can see a the |
|---|
| 0:02:09 | fusion of them and the comparison with the work these landed pprlm |
|---|
| 0:02:14 | and a fusion with that and the standard acoustic system based on mfccs |
|---|
| 0:02:19 | and |
|---|
| 0:02:20 | c different portions finally where we can see that a |
|---|
| 0:02:24 | this approach also provides complementary information so there are a final improvements in our global |
|---|
| 0:02:30 | system |
|---|
| 0:02:31 | that's it |
|---|