ws-choi / ismir2020_u_nets_svs Goto Github PK

Hello, Thank you for your sharing codes.

Could you please tell me how long does it take you to train this model (TFC_TDF_Net (large))?
In my setup, one 2080ti, batch_size=8, it takes 12 minutes for every epoch. It means that I need to wait 17 days for 2000 epochs. It looks like unreasonable. Is this the same as your training time?

Looking forward to you.

Hello, thanks for good example repository!

Is there any way to fix lags in model output? I mean two things:

1. much part of realtive volume is changed. Some pieces of song sounds in increased volume, relative to all track, comparing to original. Some times it's sound very glitchy, so sounds that were very quet in original, now very loud and much louder than rest of track.
2. Some strange sound in output, like glitches. Like some non voice sounds from original track, but in increased volume and transformed sound. Sometimes it sound scarry )))

Tranks in advance!

ImportError: cannot import name 'EvalResult' from 'pytorch_lightning'

Hello, I encountered this error when configuring the environment. How can I solve it?
Traceback (most recent call last):
File "main.py", line 7, in
from source_separation.models.scripts import trainer, evaluator
File "/home/Music/ISMIR2020_U_Nets_SVS-master/source_separation/models/scripts/trainer.py", line 8, in
from source_separation.models.model_definition import get_class_by_name
File "/home/Music/ISMIR2020_U_Nets_SVS-master/source_separation/models/model_definition.py", line 3, in
from source_separation.models.tdc_net import TDC_NET_Framework
File "/home/Music/ISMIR2020_U_Nets_SVS-master/source_separation/models/tdc_net.py", line 8, in
from source_separation.models.dense_u_net import Dense_UNET_Framework, Dense_UNET
File "/home/Music/ISMIR2020_U_Nets_SVS-master/source_separation/models/dense_u_net.py", line 9, in
from source_separation.models.separation_framework import Spectrogram_based
File "/home/Music/ISMIR2020_U_Nets_SVS-master/source_separation/models/separation_framework.py", line 12, in
from pytorch_lightning import EvalResult
ImportError: cannot import name 'EvalResult' from 'pytorch_lightning' (/opt/conda/lib/python3.7/site-packages/pytorch_lightning/init.py)

Hello, I would like to know how to implement multi-instrument source separation, I only want to use this code, thank you