STFT.inverse() could not work correct under batch operations

[mcw519]

Hi,

I tried the STFT and its inverse function found that could only be worked correct under batch size is 1.
Looks like more than 1 batch size, the batch audio would be merged together.
Is any idea for this issue?

Thanks.

[mcw519]

Simple code to reproduce it.

[mcw519]

Hi,

I found that the error happen inside the extend_fbins()
I changed this line:
X_upper = torch.flip(X[:,1:-1],(0,1))
to
X_upper = X[:,1:-1].flip(1)
and it worked!

[KinWaiCheuk]

Hi,

I found that the error happen inside the extend_fbins()
I changed this line:
X_upper = torch.flip(X[:,1:-1],(0,1))
to
X_upper = X[:,1:-1].flip(1)
and it worked!

Oh! Good job finding and fixing the problem. I am also trying to figure out what's wrong just now, haha.
Do you mind making a pull request?

[mcw519]

sure. ^^