sweetcocoa / deepcomplexunetpytorch Goto Github PK

Hi,

When I ran "estimate_directory.py" with a directory that contains almost 5 minutes long wav files I got a memory error. It ran with smaller wav files, but I need it to run with larger files. Could you please take a look at that? I'll try to fix it as well. Thank you for your good work.

Have a nice day

Hi,
First, thanks for sharing this code. I tried to train a model with my dataset. But it gave "Processing error!" and stop working. Could you please tell me what causes this error?

Thanks.

May I ask the what is the package "PinkBlack"?

Hallo,

I have run this code on same dataset (DEMAND+VoiceBank). But during the training the train_metric(pesq) is much smaller than val_metric, namely, train_metirc is about 2.7 and val_metric is about 3.2. Do you have the similar situation?

if (err_info->Nutterances < 1)
{
/printf ("Processing error!\n");/
/exit (1);/
PyErr_SetString(PyExc_RuntimeError, "processing error!");
}
Log print warnings.warn('Processing Error! return NaN........')
use older version, i see it stop at conditions err_info->Nutterances < 1 processing error! WHY?
but if i change metric="pesq" to metric="" ，training is ok!

Hi, thanks for your code.
It seems the paper has been retracted due to data fold setting error, April 2019.
https://openreview.net/forum?id=SkeRTsAcYm

Is there pretrain model for download?
Thanks!

I see you put 'torch.tanh' directly on 'mask' produced by 'self.linear(p)' in the 'forward(self, bd)' of 'class UNet(nn.Module)' (unet.py, line 119). If I read this right, you implement 'torch.tanh' on an complex number, but it implements 'torch.tanh' on the magnitude of the complex number as published in the 'Phase-Aware Speech Enhancement with Deep Complex U-Net' paper. Are this two implementation the same? Or it should be implemented like the 'mask_processing' in https://github.com/Doyosae/Phase_aware_Deep_Complex_UNet/blob/master/model_module.py (line 73)?