cannot use svcg

[kevinrex66]

Describe the bug
The conda enviroment is OK and i infer my own model what i trained before successfully, but i cannot run the svcg

Traceback (most recent call last):
File "/home/anaconda3/envs/so-vits-svc-fork/bin/svcg", line 5, in
from so_vits_svc_fork.gui import main
File "/home/anaconda3/envs/so-vits-svc-fork/lib/python3.10/site-packages/so_vits_svc_fork/gui.py", line 10, in
import sounddevice as sd
File "/home/anaconda3/envs/so-vits-svc-fork/lib/python3.10/site-packages/sounddevice.py", line 71, in
raise OSError('PortAudio library not found')
OSError: PortAudio library not found

PS：The infer result sang out of the tune, how can i improve the result except add dataset?

[Agamemnoid]

sudo apt-get install libportaudio2

Let me know if that works.

[kevinrex66]

sudo apt-get install libportaudio2

Let me know if that works.

It works,but how can i improve the model, it sang out of the tune?

[Agamemnoid]

1:
I personally run the command
svc pre-hubert -fm crepe

BEFORE training the model. Then during inference I use the crepe auto-predict method.

2:
Train longer. Depending on your dataset. I personally still see improvements up to 1 Million steps. It's definitely diminishing returns, but it does exist.

3:
Better quality target. Not your training dataset, but the song you're inferencing onto. Try another one, clearer, make sure you use vocals only with absolutely no background noise. Make sure it's as clear as possible.

Other than that, you're out of options. It's quite possible that with a shitty dataset you will simply never ever have a good model.