0:00:15 | but i and i'm looking at automatic accent recognition in the context of forensic applications |
---|
0:00:22 | so forensic speech scientist have access to use a speaker recognition technology so you assess |
---|
0:00:30 | multiple recordings and to see how likely it is that the speech in these recordings |
---|
0:00:36 | were produced by the same speaker |
---|
0:00:38 | other kinds of cases i one might be interested in the speech community and then |
---|
0:00:43 | nine speaker or hunting speakers belonged c and i'm investigating whether we can apply automatic |
---|
0:00:50 | accent recognition c and this kind of problem |
---|
0:00:55 | so i'm |
---|
0:00:56 | a one step towards doing that and this that i'm taking with the place to |
---|
0:01:01 | do that is and see you investigate whether an automatic accent recognition technology can work |
---|
0:01:08 | on and while i'm having geographically proximate accent |
---|
0:01:13 | and the assumption here is that and that's a greater degree of similarity between these |
---|
0:01:19 | different accents eh i'm evaluating five different automatic accent recognition systems |
---|
0:01:26 | an eight corpus or else |
---|
0:01:29 | for accents |
---|
0:01:30 | and |
---|
0:01:31 | which is from the i subcorpus |
---|
0:01:34 | and i've got before locations and within their use pads with only got ten miles |
---|
0:01:40 | sitting between like |
---|
0:01:42 | stereo we do next |
---|
0:01:44 | to find differences between these accent variety |
---|
0:01:47 | and |
---|
0:01:48 | but we expect us differences and well just high degree of similarity study i'm assessing |
---|
0:01:55 | how sensitive and different automatic accent recognition systems are |
---|
0:02:00 | a combination of text-dependent in text-independent systems and the how robust these things can be |
---|
0:02:07 | that this problem features that's challenges these systems in |
---|
0:02:13 | in different steps |
---|
0:02:14 | thank you |
---|