search

prophesier / diff-svc

Singing Voice Conversion via diffusion model
GNU Affero General Public License v3.0
2.62k stars 800 forks source link

AttributeError: 'HifiGAN' object has no attribute 'device' #8

Open betterftr opened 1 year ago

betterftr commented 1 year ago

on windows, anaconda

betterftr commented 1 year ago

modified line 64 in network/vocoders/hifigan.py fromdevice = self.device to device = torch.device("cuda") now it says: AttributeError: 'HifiGAN' object has no attribute 'model'

betterftr commented 1 year ago

Using this hifigan pretrained model from : https://github.com/MoonInTheRiver/DiffSinger/releases/download/pretrain-model/model_ckpt_steps_1512000.ckpt Can this cause issues?

prophesier commented 1 year ago

check if you have excuted ‘set PYTHONPATH=.’ in the root directory of this project

InDragonR commented 1 year ago

Check the folder name in your yaml file, "vocoder_ckpt:" should match the folder name. Also, mine has an additional ptn model. Not sure if that is the problem.

dillfrescott commented 1 year ago

I am getting this error as well...

dillfrescott commented 1 year ago

Fixed it. In the vocoder_ckpt: folder, there needs to be the config.yaml file and the checkpoint from the 0109_hifigan_bigpopcs_hop128.zip file under DiffSinger releases...

Both files are in that zip file!

misdom commented 1 year ago

Where is the vocoder_ckpt folder? network/vocoders? i dont see that anywhere and i have the config.yaml and checkpoint files for nsf_hifigan. I keep getting: AttributeError: "nsfhifigan" object has no attribute "h"

ayanesakura commented 1 year ago

Where is the vocoder_ckpt folder? network/vocoders? i dont see that anywhere and i have the config.yaml and checkpoint files for nsf_hifigan. I keep getting: AttributeError: "nsfhifigan" object has no attribute "h"

so you need to create the folder and set vocode_ckpt arg in config.yamlto it. it's not a already-exist folder.