0:00:08 | marianne the representing the school of and electronic electrical and computer engineering this afternoon and |
---|
0:00:16 | the title of head tool it's |
---|
0:00:19 | towards better automated for systems |
---|
0:00:22 | modeling accents for automatic |
---|
0:00:25 | two recognition |
---|
0:00:28 | i want you to remember all those times you had to use an automated phone |
---|
0:00:33 | system |
---|
0:00:34 | how was your experience how did you feel |
---|
0:00:39 | a bit you feel frustrated by don't number of times you have to repeat which |
---|
0:00:43 | is that older an over again |
---|
0:00:46 | even then you might have been transferred to the wrong department |
---|
0:00:50 | but no that's is over you have to wait and then long q listening to |
---|
0:00:54 | a web had to solve and a voice that |
---|
0:00:58 | you're called is extremely important to us ulysses us assume that the next agencies available |
---|
0:01:05 | despite all this stuff estimation that we go through using these speech enabled automated phone |
---|
0:01:11 | systems rather than the traditional types to one |
---|
0:01:15 | the save an average cost and set up to three million pounds manually |
---|
0:01:20 | by two thousand and seventeen this industry could be more up to two billion pounds |
---|
0:01:27 | after all there seems to be no way to avoid you it's speech enabled voice |
---|
0:01:32 | jails |
---|
0:01:33 | but why and these systems understandable be stated |
---|
0:01:39 | and they are not that clever human |
---|
0:01:42 | they cannot adapt then you speakers back then |
---|
0:01:46 | in fact |
---|
0:01:47 | a few months ago you mean intensity con still had to stand eleven million pounds |
---|
0:01:52 | an automated phone system |
---|
0:01:55 | that we couldn't work to be the local accent |
---|
0:01:59 | the goal of my phd just all these you mean and i can't robust automatic |
---|
0:02:04 | speech recognition system |
---|
0:02:07 | so |
---|
0:02:08 | i have this massive database of recorded speech from people be different regional accent |
---|
0:02:15 | then an unknown speaker talks to my system it recognises their accent it then selects |
---|
0:02:21 | the data from a there is speakers with similar accent in order to create a |
---|
0:02:26 | more personalised model out there is speech |
---|
0:02:30 | using the is really improve the performance of my system incredibly |
---|
0:02:35 | in fact |
---|
0:02:36 | the error rate will be reduced by up to fifty path and for scientific contact |
---|
0:02:41 | them |
---|
0:02:41 | such as |
---|
0:02:43 | lexington |
---|
0:02:45 | you can't result of this is that customers |
---|
0:02:48 | we have more productive time to use during their day instead of a tuning call |
---|
0:02:54 | center cues and how companies |
---|
0:02:57 | thus have cu were number of angry customers |
---|
0:03:01 | in the next page of my clear i would like to address the issue space |
---|
0:03:05 | by my now ready to fight as children elderly people with the speech this so |
---|
0:03:10 | the thank you |
---|