| 0:00:15 | hello everyone so |
|---|
| 0:00:19 | this looks we but in this presentation i'm just presenting how to use i-vectors for |
|---|
| 0:00:25 | speech activity detection |
|---|
| 0:00:28 | so actually this is the what was done in the context of the nist open-set |
|---|
| 0:00:31 | channels that use actually very tough later that probably many of your nose |
|---|
| 0:00:37 | and all right speaker on rats data |
|---|
| 0:00:40 | so the proposed technique is actually to make i-vectors work in the scenario so what |
|---|
| 0:00:46 | i did is that i first cluster the data in it to unsupervised manner one |
|---|
| 0:00:53 | of them is based on what they should clustering mainly a k-means plus gmm clustering |
|---|
| 0:00:58 | and the other one is actually based on |
|---|
| 0:01:00 | more accurately clustering after some kind of segmentation |
|---|
| 0:01:04 | and the so then the output of this clusters idea is i can classify them |
|---|
| 0:01:10 | using different kind of a classifiers i tried the i-vector extract i-vectors of agenda and |
|---|
| 0:01:18 | and the results are quite the promising |
|---|
| 0:01:22 | probably bit complex system but the output of this l this class that can be |
|---|
| 0:01:26 | used not only for speech detection but also for other tasks like their decision about |
|---|
| 0:01:31 | or what you event detection or and so please the zero by posted to your |
|---|
| 0:01:37 | interest like |
|---|