input of bob.kaldi.plda_train function.

38 views
Skip to first unread message

Chang Pan

unread,
Mar 29, 2018, 2:36:47 PM3/29/18
to bob-devel
Hi,

I have a question regarding to the input of plda_train under ivector.py.
As comment for
 ivector_extract(feats, fubm, ivector_extractor, num_gselect=20, min_post=0.025, posterior_scale=1.0)
 feats should be a 2D numpy ndarray object containing MFCCs.
However, under the test case, the shape of 'data/feats-mobio.npy' is (37, 192, 100). 
So it should be a 3D numpy ndarray. But what should it be? MFCCs should have a dimension of 39.
Could anybody tell me what should I pass as the first input?

Thank you

Chang Pan

unread,
Mar 29, 2018, 3:20:52 PM3/29/18
to bob-devel
To be explicitly, I want to know where does 37 and 192 comes from?

Milos

unread,
Mar 29, 2018, 4:24:48 PM3/29/18
to bob-devel
Hi Chang,

If I remember well, the first dimension is for the speakers, the second one for the utterances, and the third one for ivectors.

Concretely, 37 is the number of speakers in used data, each speaking 192 utterances, and the dimensionality of the ivectors is 100.

Milos

Chang Pan

unread,
Mar 29, 2018, 4:27:16 PM3/29/18
to bob-devel
Hi Milos,

Thank you for your quick reply!
This quite make sense. But does it mean I have to pick a minium number of utterances between all the speakers?

Milos

unread,
Mar 29, 2018, 5:15:52 PM3/29/18
to bob-devel
Well, it is one solution. Otherwise you would need to modify the code.

Best,
Milos

Chang Pan

unread,
Mar 29, 2018, 6:07:31 PM3/29/18
to bob-devel
Thank you for your explaination!
I'll try to modifiy the code then.
Reply all
Reply to author
Forward
0 new messages