0:00:15 | a good morning |
---|
0:00:17 | i'm gonna present |
---|
0:00:19 | in your approach or relatively new approach to extract i-vectors |
---|
0:00:24 | the idea is to extract phonetically compensated |
---|
0:00:28 | i-vectors |
---|
0:00:30 | by using my scenery that is quite close to the g a the jfa must |
---|
0:00:36 | ignore e |
---|
0:00:37 | so |
---|
0:00:40 | it's |
---|
0:00:40 | the nn will be involved a slightly differently than just under two or three ways |
---|
0:00:48 | meaning of a bottleneck approach |
---|
0:00:50 | or the use the use of the d n n's |
---|
0:00:56 | instead of ubm |
---|
0:00:57 | in this case we gonna use the ubm |
---|
0:00:59 | and |
---|
0:01:01 | we consider remote these model as a probabilistic extension to the subspace |
---|
0:01:07 | gmm |
---|
0:01:09 | and the core of the idea is to treat the phonetic variability a side and |
---|
0:01:15 | nuisance variability |
---|
0:01:17 | and |
---|
0:01:19 | we assume for it to do that we assume that at each frame |
---|
0:01:23 | a week this super vector that corresponds to each frame to each observation can be |
---|
0:01:29 | decomposed |
---|
0:01:30 | into an i-vector by corresponds to the combination of speaker and channel |
---|
0:01:36 | and |
---|
0:01:38 | plus |
---|
0:01:39 | and which an analysis variability that captures |
---|
0:01:43 | the phonetic variability |
---|
0:01:46 | k and that's where the d n and games |
---|
0:01:50 | can't and that provides an extra supervision |
---|
0:01:54 | okay so my telling us which seen on is probably |
---|
0:01:58 | i which corresponds to this particular frame |
---|
0:02:02 | probabilistically of course |
---|
0:02:04 | we form of the variational bayes algorithm to train the model |
---|
0:02:08 | and |
---|
0:02:10 | practically estimated to subspaces |
---|
0:02:14 | and |
---|
0:02:16 | as it all you the nn provides these extra supervision |
---|
0:02:20 | which differentiated from the channel factors that we have enough jfa model so please come |
---|
0:02:27 | to the possible to discuss the |
---|