vitrioil / speech-separation Goto Github PK

Hi,
Really nice project!
Would you like to tell more about your project?

Please, join our slack,
https://join.slack.com/t/catalyst-team-core/shared_invite/zt-d9miirnn-z86oKDzFMKlMG4fgFdZafw

Hi,
When you do this op

You are commiting a mistake.
You are doing input.view(B, -1, 298, 1) and that is not correct.
In pytorch the reshaping op is ordered from right to left.
You have to do a permutation
permute(0,2,1,3)
and then the reshaping
view(B,298,-1)
Basically this way the values
[0,0,0,0:257]-->[0,0,0:257]
[0,0,0,1:257]-->[0,0,257:257*2]
and so on.
Whay you are doing is mixing the data
You are putting
(B,8,298,257)
[0,0,0,0:257] --> [0,0,0:257]
[0,0,1,0:257] --> [0,0,257:298], then [0,1,0:298-257] and filling the reshaped tensor in a bad way

Hi there,
@vitrioil Just want to ask have you overcame the overfitting problem that you reported in README?
Thanks, Do you have any idea of your overfitting? and any idea to overcome it?
how much data you train on? thanks

In the dowload process, I thinks you must return the variable "indx" and not the variable "df.shape[0]" in this line:

Otherwise you may get pathes with the same name for the val and train.csv that override each other.
Thanks.

This line is not working:

Help please!