| 0:00:15 | the data may be run and my name is still from the middle of that |
|---|
| 0:00:20 | and today i write it is worth |
|---|
| 0:00:29 | i think is that is high salaries keynote speech the in their everybody always racist |
|---|
| 0:00:35 | and i spoofing attacks because of cs program |
|---|
| 0:00:39 | and the a lot of groups have proposed a of beer |
|---|
| 0:00:45 | approach is to improve |
|---|
| 0:00:48 | s |
|---|
| 0:00:50 | on the other hand |
|---|
| 0:00:52 | we've all got |
|---|
| 0:00:53 | almost looking at a rate eer out speakers |
|---|
| 0:00:57 | and we propose that we will always that the nist detection frame before |
|---|
| 0:01:04 | voice that it was entirety |
|---|
| 0:01:05 | we usually you |
|---|
| 0:01:08 | can say that the goal is to classify an input sample rates at the by |
|---|
| 0:01:15 | our actual human or a sprint be about the speaker |
|---|
| 0:01:19 | two realizable in the framework we can well known scandal now everything is a vector |
|---|
| 0:01:26 | q one |
|---|
| 0:01:27 | we already proposed to pop noise dictionary it's and this time we propose a fundamental |
|---|
| 0:01:37 | from a in extreme in from experience there is that's a the conventional way to |
|---|
| 0:01:44 | be most around a space is then improve |
|---|
| 0:01:48 | robustness |
|---|
| 0:01:52 | thank |
|---|