| 0:00:14 | so high and gregory usually and form in c and i'll be presenting the work |
|---|
| 0:00:21 | we did with a bow capture research in preparation for than nist language recognition evaluation |
|---|
| 0:00:27 | of to some fifteen |
|---|
| 0:00:31 | so what would it is we just to did for different systems and phonotactic one |
|---|
| 0:00:36 | us an i-vector system |
|---|
| 0:00:39 | a long short-term memory recurrent neural network and the lexical couple component |
|---|
| 0:00:45 | and the main results |
|---|
| 0:00:47 | that can sure but i will be happy to discuss more a new of the |
|---|
| 0:00:51 | poster |
|---|
| 0:00:54 | on that the l s t m r and then can lead to a lower |
|---|
| 0:00:57 | and lower language error rate than i-vectors |
|---|
| 0:01:02 | still the phonotactic system is the most robust with a method especially when you data |
|---|
| 0:01:08 | are available for language |
|---|
| 0:01:11 | and when facing are very strong mismatch between training and testing which was the case |
|---|
| 0:01:17 | for the lre or fifteen |
|---|
| 0:01:20 | and a what's worse really interesting for ice is that the phonotactic system and the |
|---|
| 0:01:26 | l s t n r and then really combined their combine really well in that |
|---|
| 0:01:32 | the combination of the two system lead to an important a language or rate reduction |
|---|
| 0:01:38 | and if you want to know more about the few euros here |
|---|
| 0:01:41 | i you have to come and see the poster in speaker and me thank you |
|---|