| 0:00:14 | a good afternoon |
|---|
| 0:00:17 | so in this at one |
|---|
| 0:00:19 | allpass in this a new method call between class covariance collection to improve language recognition |
|---|
| 0:00:24 | performance |
|---|
| 0:00:25 | the we conducted our experiments on the nist lre two thousand fifteen corpus though see |
|---|
| 0:00:31 | that the corpus is organized like there are twenty languages and each of them are |
|---|
| 0:00:34 | grouped into six clusters based on their phonetic similarities like you have a rabbit cluster |
|---|
| 0:00:38 | which has all the grabbing dialects of english and french all these clusters so we |
|---|
| 0:00:43 | followed a very interesting thing when we |
|---|
| 0:00:46 | a lot all the i-vectors when the past into the pca and these are the |
|---|
| 0:00:51 | first two dimensions of the first two base of pca so we found that |
|---|
| 0:00:55 | the all these languages are grouped together in the form of clusters and all of |
|---|
| 0:00:58 | these clusters |
|---|
| 0:01:00 | so you can see that all the languages going to the chinese cluster they are |
|---|
| 0:01:04 | grouped together |
|---|
| 0:01:05 | a belong to be i've been a cluster there are grouped together |
|---|
| 0:01:07 | so they are wonderful multimodal distribution |
|---|
| 0:01:10 | and so we so we computed the eigen directions representing this multimodal distribution and we |
|---|
| 0:01:15 | added them to the lda |
|---|
| 0:01:18 | initial some improvement in performance |
|---|
| 0:01:20 | so you wanna |
|---|
| 0:01:21 | no more you in the post animal is once i welcome you all their and |
|---|
| 0:01:25 | to get the more details about it things |
|---|