Need a stable web ui for sovits4.0 inference

[livid lake]

Does anyone knows a free and stable web ui for sovits4.0 inference?
I don't have a nvidia gpu so I can only go to online training and inferencing.
The google colab script I was using was unable to run inference now (because of the update of colab maybe), so I was searching for a new web ui.
I have look for the hugging face spaces around, but didn't find a space that I can upload my local model and successfully run inference.
The only hugging face space that I can upload my own model showed this error when I press import model after uploading. I wonder it's due to my model or the space.
If you would like to share how you do the inference, please let me know, thanks.

Here are the colab script and hugging face space I've tried but not successful:
https://colab.research.google.com/github/svc-develop-team/so-vits-svc/blob/4.1-Stable/sovits4_for_colab.ipynb
https://huggingface.co/spaces/neuroama/so-vits-svc

edit: I'm trying RVC now

[heavy geode]

-svc

[sick ermine]

https://cdn.discordapp.com/attachments/1159822429137412177/1194008385113301003/svc.gif?ex=65c13ef6&is=65aec9f6&hm=ce1f57bc316793f641c80f9b359ba9db7d1312af6918b1aed9c8468c74035bac&

[ember saffron]

sovits is old

use RVC

[livid lake]

sure i will consider RVC to train a new model, but i need to inference a song immediately, is there any solution?

[ember saffron]

you can use RVC for that too

what’s ur pc gpu

[livid lake]

i don't have gpu so i can only go to cloud training and inferencing

i only have a sovits model trained by myself on my hand

[ember saffron]

Train (make) RVC Models on cloud:

1. Prepare the Dataset
2. Setup RVC:
   Choose a cloud way to use RVC,

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
  • RVCDISCONNECTED (no ui)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI, no guide as of right now)
  • Applio by Shirou (UI, no guide as of right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. Be sure to know about the tensorboard

Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time
If you are looking for the easiest way and for free, is using https://weights.gg which ofc uses RVC

RVC Inference (use models) on pre-recorded audio on Cloud

You can use either:

• Weights.gg: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio (ui)

[ember saffron]

u used cloud for that too?

Btw so vits is really old

[ember saffron]

oh, I’m guessing u already checked

[livid lake]

yeap

how much time it takes to train a RVC model on kaggles on average? is there a chance to make it by today or tomorrow?

[ember saffron]

well it various by dataset lenght, batch size and gpu, I can’t really predict how much it’s gonna take u but a few hours

I don’t even think there’s a stable svc, no one uses it since 2 years

[livid lake]

i have a raw dataset for about one and a half hours

[ember saffron]

Usually peopledon’t train over an hour for RVC, but yeah u should be fine as long as they are good quality

[livid lake]

i'm setting the RVC Mainline Colab on kaggles, when it goes to the theme loader cell it goes like this: Loaded Theme: ParityError/Interstellar
what's going on??

[ember saffron]

that’s just the theme of the webui, don’t worry

[livid lake]

phew thanks a lot

btw i still click the start running below and it has been running for about 10 or 20 minutes and still nothing came out, is that normal?

[ember saffron]

u sure u put ur ngrok token

[livid lake]

if there are only one place to put it on, yes i think so

[ember saffron]

could u show a ss (censor ur token tho)

[ember saffron]

i mean the whoke thing

the whole screen

[ember saffron]

i mean the output too

like the thing that’s under the code cell

[livid lake]

that's the point. there is no output

[ember saffron]

u sure u run it?

be sure the ngrok token is inside the “”

[livid lake]

yes of course

how to ensure that the cell is running? i can only see the blue circle beside the cell spinning but the gpu isn't running though

[ember saffron]

try to stop running it and run it again

[livid lake]

tried for several times

usually how long does it take for the step?

looks like the cell is running but the gpus isn't

[ember saffron]

could you try to restart the whole thing

weird

[livid lake]

i'll try it

omg the urls comes out

thank god

[ember saffron]

@primal spade u might need to check out why it doesn’t appear the first time but needs a restart, not sure if it’s a silly issue

[primal spade]

just tested it earlier

works fine for me training twice

[ember saffron]

maybe it was a silly moment

[livid lake]

when i press the process data button it came out this: start preprocess
['infer/modules/train/preprocess.py', 'dataset', '32000', '2', '/kaggle/tmp/training/logs/Serval', 'False', '3.0']
dataset/serval.zip->Traceback (most recent call last):
File "/kaggle/tmp/training/infer/lib/audio.py", line 37, in load_audio
ffmpeg.input(file, threads=0)
File "/kaggle/tmp/training/.venv/lib/python3.10/site-packages/ffmpeg/_run.py", line 325, in run
raise Error('ffmpeg', out, err)
ffmpeg._run.Error: ffmpeg error (see stderr output for detail)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/kaggle/tmp/training/infer/modules/train/preprocess.py", line 87, in pipeline
audio = load_audio(path, self.sr)
File "/kaggle/tmp/training/infer/lib/audio.py", line 42, in load_audio
raise RuntimeError(f"Failed to load audio: {e}")
RuntimeError: Failed to load audio: ffmpeg error (see stderr output for detail)

end preprocess

is it problem with my dataset?

is there any request forum for wav files?

[ember saffron]

Show what’s inside the zip

[livid lake]

all the wav files forumed like this inside the zip

[ember saffron]

what if u try to put the files in the dataset folder instead of the zip

[livid lake]

you mean directly upload this file to the dataset folder?

[primal spade]

it needs to be a folder not a zip

[livid lake]

oh i see

[primal spade]

there should be an input thing on kaggle

you can just press the copy icon and use that for the dataset on the ui

[livid lake]

where should it be? i didn't see it

[primal spade]

kaggle website

[livid lake]

below the urls?

[primal spade]

late reply but you didnt add the datasets on the kaggle website

idk if its possible to unzip it on the filemanager

[livid lake]

i just restart the session then click the RVC url but it doesn't work

oh it open now

it's fine

[livid lake]

so, should I click it and upload the files?

[primal spade]

yup

upload the zip file there

it will automatically unzip it

[livid lake]

what, zip? i just upload the original floder

[primal spade]

the .zip file

[primal spade]

this one you have

[livid lake]

what should the title be

[primal spade]

just put whatever you want

[livid lake]

understand

still the same error!

start preprocess
['infer/modules/train/preprocess.py', '/kaggle/input/serval', '32000', '2', '/kaggle/tmp/training/logs/serval', 'False', '3.0']
/kaggle/input/serval/serval->Traceback (most recent call last):
File "/kaggle/tmp/training/infer/lib/audio.py", line 37, in load_audio
ffmpeg.input(file, threads=0)
File "/kaggle/tmp/training/.venv/lib/python3.10/site-packages/ffmpeg/_run.py", line 325, in run
raise Error('ffmpeg', out, err)
ffmpeg._run.Error: ffmpeg error (see stderr output for detail)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/kaggle/tmp/training/infer/modules/train/preprocess.py", line 87, in pipeline
audio = load_audio(path, self.sr)
File "/kaggle/tmp/training/infer/lib/audio.py", line 42, in load_audio
raise RuntimeError(f"Failed to load audio: {e}")
RuntimeError: Failed to load audio: ffmpeg error (see stderr output for detail)

end preprocess

wait, there's another serval folder inside the zip

i fix it

[primal spade]

so it should be /kaggle/input/serval/serval

[livid lake]

right

now i have the processed data

[heavy geode]

nvm im not sure the path as I'm using applio, but that's good

[primal spade]

next is feature extract just bellow process data

[livid lake]

yeah i have done that

so now i just click the train model button right?

[heavy geode]

"train index" is the faster one to do rn, so then you can start training the actual model

[livid lake]

but i have click the left one and start training

[primal spade]

you could press both since they can train at the same time

but i personally train index first

[livid lake]

how do i know how many steps does it save a model?

[primal spade]

there is no "sync" for it for the step count and epoch count

[livid lake]

i don't understand what this part say

when should i stop training?

[heavy geode]

it means to adjust the log interval to be same as the steps per epoch, but you dont need to do that in the latest Applio as it already logs per epoch

[livid lake]

it means i can jump the step 3?

[livid lake]

how to operate the inference ui? where to put on the audio i want to convert

[livid lake]

the kaggle website says it's failed to save draft. is that matter?

[livid lake]

the start running cell goes back to this, but the tensorboard still goes on

the gpus are still running too

but the epoch stuck on 99, even though an e100 model has been saved

what does it mean?

the gui shows an error also

what should i do??

the gpus are still running and the tensorboard shows over 10000steps now, there should be an e125_s9875.pth in the weights, but there isn't.

[livid lake]

this one appeared so i guess it's fine?

[livid lake]

but i still don't know what the gui's error means

[livid lake]

I've downloaded the e100 model and use weights to inference and the output is ok so my needs wrere temporarily met.