| 0:00:15 | i would have done it's two hundred and i'm going to prison what only spoken |
|---|
| 0:00:22 | detection |
|---|
| 0:00:23 | on the spoofing challenge corpus |
|---|
| 0:00:26 | employing deep neural network |
|---|
| 0:00:28 | this is actually an extension of our previous work for spoken challenge task |
|---|
| 0:00:34 | a where you're given actually and |
|---|
| 0:00:37 | a different spoofing attacks generated using different voice conversion is and speech synthesis technique |
|---|
| 0:00:46 | and |
|---|
| 0:00:48 | five i'm on the dandy spoofing attacks one on a prior it was in actually |
|---|
| 0:00:55 | in the u but evaluation set we had five hundred hz and five one text |
|---|
| 0:00:59 | so |
|---|
| 0:01:01 | in this work in this is what would be in this work to overview of |
|---|
| 0:01:06 | this one |
|---|
| 0:01:07 | who what right here actually tried to train a the nn |
|---|
| 0:01:12 | which we will try to discriminate between school and |
|---|
| 0:01:16 | human this basic and then we try to extract |
|---|
| 0:01:20 | order thing feature representation and the users would with the standard |
|---|
| 0:01:23 | a gmm classifier also we use a |
|---|
| 0:01:27 | tandem feature which is basically concatenation of |
|---|
| 0:01:31 | bottleneck feature and |
|---|
| 0:01:34 | acoustic level spoofing detection features and we try to reduce the damage using pca |
|---|
| 0:01:40 | and then he to use this the gmm classifier |
|---|
| 0:01:44 | so if you want to know more able or walk and the results based on |
|---|
| 0:01:48 | that the poster session was a number six thank you |
|---|