OH THATS FUCKING WHY
#✨│ai-help
brittle wing
please select RMVPE here
it still sounds weird
rare gobletBOT
Ayo? @brittle wing level 2 !!! 
half cove
id recommend it
brittle wing
it still sounds weird
Give me a link to the model and your audio and I'll test it for you
it could very well be the model being dodgey
but I am willing to test it, got nothing better to do on this Sunday night.
brittle wing
Give me a link to the model and your audio and I'll test it for you
ok
The audio send in DM though because copyright wacky
Sometimes badly trained models can sound awful.
So narrowing it down will help
there
One moment pls 
fickle delta
Can I get some help on this? It sounds kind of robotic at parts. is there any way to somewhat perfect the sounds?
like bits where the voice will try to dip lower and sound really like
broken
brittle wing
there
Ok so try Crepe
RMVPE sounds a bit more robotic in this case than Crepe for some reason.
But the model doesn't sound that clean in general.
fickle delta
who are you talking to?
rare gobletBOT
Ayo? @fickle delta level 2 !!!
brittle wing
The one who I replied to.
fickle delta
ah
brittle wing
Can I get some help on this? It sounds kind of robotic at parts. is there any wa...
Is this an infer or a voice recording from Okada?
brittle wing
Ah okay, better to go to #🔍│help-w-okada for that then if you want advice for live inferring through microphone.
fickle delta
i dont really care too much about it being live
would using inferrence be higher quality?
brittle wing
Since it's not being done live, it can to some extent be less artifacty.
Because your GPU is being fed data live whilst trying to infer it, it can cause a heck of a lot of artifacts, but that is fixable.
fickle delta
ah
brittle wing
Ngl I haven't touched Okada in ages LMAO, could you give me a screenshot and I will try to point you in the right direction if you wanted to use it?
I know the setting to play with, just not what it's called
😭
fickle delta
tried out inferrence
it seems better but I had to record straight from the voice recorder app in windows lol
azure marshBOT
RVC Guides (How to Make AI Cover)
brittle wing
guys, how can i train a model to be open-souce? cuz if i clone someone else's voice it will be bad right? like if me ppl get angry at me 😦 , i'm new to this so idk how this kind of thing works
half cove
How did u know i was on the channel
half cove
which one for singing?
Rmvpe
brittle wing
The chunk I believe is what I am thinking of
Also set the F0 Detection thingy to RMVPE, it will use less resources @fickle delta 
RMVPE_gpu I think is what it's called in Okada.
half cove
Okada it's just called rmvpe if I'm not mistaken
Might be wrong
Rmvpe and for directml rmvpe-onnx
brittle wing
Rmvpe and for directml rmvpe-onnx
rmvpe-onnx is probs just rmvpe-gpu
i haven't touched okada in long
west lark
i dont think i get it
brittle wing
Have you installed python?
And did you click the "add to path" thing I told you to at the end of installation?
west lark
yeah
west lark
yeah
rare gobletBOT
Ayo? @west lark level 2 !!!
brittle wing
Okay so have a look at the answer there and add PYTHON to the path
installer didn't do it for you it seems
which is odd.
west lark
yeah i dont get it
im just trying to get to the app, didnt know it would be this hard 😂
brittle wing
It shouldn't be, IDK why your python installer didn't add the thing to PATH tho, usually works fine for me.
glossy tartan
i tried setting up tensorbaord but then i got this
PS C:\AI\RVC1006Nvidia> python -m venv venv
PS C:\AI\RVC1006Nvidia> .\venv\Scripts\activate
.\venv\Scripts\activate : File C:\AI\RVC1006Nvidia\venv\Scripts\Activate.ps1 cannot be loaded because running scripts is disabled on this system. For more information, see about_Execution_Policies at https:/go.microsoft.com/fwlink/?LinkID=135170.
At line:1 char:1
- .\venv\Scripts\activate
-
+ CategoryInfo : SecurityError: (:) [], PSSecurityException + FullyQualifiedErrorId : UnauthorizedAccess
help?
rare gobletBOT
Ayo? @glossy tartan level 1 !!!
west lark
It shouldn't be, IDK why your python installer didn't add the thing to PATH tho,...
il uninstall and try again
this, right?
brittle wing
i tried setting up tensorbaord but then i got this
PS C:\AI\RVC1006Nvidia> pyth...
Launch Powershell with admin and type this command Set-ExecutionPolicy Unrestricted
brittle wing
this, right?
Yep
brittle wing
Okay @west lark in lieu of just pip install -r requirements have you tried python install -r requirements?
wait wtf
west lark
ill try
sand iris
i tried setting up tensorbaord but then i got this
PS C:\AI\RVC1006Nvidia> pyth...
It linked you a Microsoft article
Contains everything you'll need
brittle wing
It linked you a Microsoft article
I gave em the command
sand iris
Oh whoops
brittle wing
all good lol
brittle wing
here, right?
Where are all the files 
west lark
ill re-download it again tomorrow
half cove
Where are all the files <:PepeHands:1015671622176342106>
It's cuz...
They are on a folder on applio
west lark
ill re-download it again tomorrow
itll work probably
brittle wing
Applio's file structure is bloody weird
brittle wing
ill re-download it again tomorrow
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI try this instead tomorrow.
Mainline is awesome
brittle wing
Mainline has it too, you just have to add it yourself which isn't too hard.
Just two files you download from huggingface which I can link you in just a moment.
half cove
i wanted to get on applio only cuz of rmvpe
Mainline has rmvpe
west lark
okay then
half cove
Tf 😭
brittle wing
half cove
Weird
brittle wing
https://huggingface.co/lj1995/VoiceConversionWebUI/tree/main this is the assets folder, you also need to get the pretrained_v2 stuff as well for some weird ass reasion
after that tho, off to the races
sand iris
look under /tools/ in the RVC repo
brittle wing
Yea I did
sand iris
There's three download scripts, two shell ones (unix via sh, Windows via bat) and a python one
For the models
brittle wing
oh dayum
sand iris
They don't store it on the git because it's bad practice lol
brittle wing
mfw I didn't realise those were there 
sand iris
just gotta make sure the current folder is the root of the RVC repo, but it's best to launch the python downloader instead of the shell ones
glossy tartan
this is a graph for one of my models, the graph is going up again so that means i am overtraining it? should i just go with the lowest point?
brittle wing
Turn up the smoothing pls
glossy tartan
glossy tartan
its a finished one, i did 1000epochs with this model
brittle wing
When I train models it's mostly just trial and error
but yours seems to have plateaud.
So your model should sound ay-okay.
Chef gotta taste the food sometimes
glossy tartan
but yours seems to have plateaud.
is that a good thing? hahah
lilac pilot
say you train a model with 500 epochs and it finishes but it hasnt been overtrained. can you resume training it?
brittle wing
is that a good thing? hahah
It means that the ai has learned all it can from what it has been given, if it goes up by a dramatic amount it means that it's overtraining.
brittle wing
say you train a model with 500 epochs and it finishes but it hasnt been overtrai...
Yes, just increase the epoch amount and click "start training".
It'll just start from the checkpoint it left off at.
glossy tartan
It means that the ai has learned all it can from what it has been given, if it g...
aaah okay, so this model is as good as it can be? i can train it more by giving it more samples?
brittle wing
It's as good as it can be, if you're going to give more samples I'd suggest just making a longer dataset and training again.
Try the model though, as I say chef's gotta taste the sauce.
You'll know if it's had too much time in the oven or if it's cooked perfectly.
glossy tartan
It's as good as it can be, if you're going to give more samples I'd suggest just...
alr alr thx! this is already with 1.5 hours of data so ig its good enough, it sounds good so yea
brittle wing
If it sounds good then it's great :)
glossy tartan
thx for the help
lilac pilot
It'll just start from the checkpoint it left off at.
thank you for the help. I am training my first model so its been a steep learning curve for me
brittle wing
Good luck, I hope it turns out good :)
lilac pilot
Good luck, I hope it turns out good :)
so lets say you are training a model with a 1000 epochs and you notice its overtraining at 700, how do you stop at 700?
rare gobletBOT
Ayo? @lilac pilot level 2 !!!
brittle wing
There's 3 ways to answer this:
1: Some RVC forks have "stop training" buttons so you can use that.
2: RVC has the ability to save a model at a set checkpoint (every 50 epochs for example) so you can save your model at each step.
3: I wouldn't go full ham on 1000 epochs straight away, I personally do it in increments.. 250 as a base and if all is well, increase to 500, and so on.
bleak nymph
need rvc v2 huggingface for trainign
brittle wing
-colab
brittle wing
HFv2 I think has training.
But it only uses CPU
I'd recommend using Kit Lemonfoot's colab.
lilac pilot
There's 3 ways to answer this:
1: Some RVC forks have "stop training" buttons s...
thank you, that seems like the best method. just keep increasing the epochs until you notice overtraining.
lilac pilot
so does that mean after it's done training you can go back to the earlier epochs?
like it is saving every 10 epoch out of 500 so i will have 50 files to choose from?
sorry if these questions seem naive. i'm a baby trying to learning how to speak lol
lilac pilot
I'd recommend using Kit Lemonfoot's colab.
luckily this is the one i am using 🙂 Are there any good options for training on MAC locally?
brittle wing
Mac training is a bit fickle but it can be done
brittle wing
luckily this is the one i am using 🙂 Are there any good options for training ...
this guide should help you out if you have M1 silicon :)
bleak nymph
I'd recommend using Kit Lemonfoot's colab.
where can i find one
rare gobletBOT
Ayo? @bleak nymph level 3 !!!
silent sun
I downloaded the app on Mac OS and it won't open do I need to download any program to make it work?
vital mortar
Is there any way of making a custom voice? Like making a combination of voices into one but not exactly cloning the exact likeness of someone's existing voice because I want to make an original voice on my own so I don't have to worry about copyright. For example, I want to make a happy-go-lucky jolly clown voice but I don't want to copy like for example Pennywise or Ronald Mcdonald's voice.
glad zealot
@brittle wing you think I should make mangio gui to work again on colab? Since people want to pause training sometimes and that's not on mainline
brittle wing
Give it a go honestly <3
short pecan
meanwhile how to use Mangio to dub audio with AI? Only used it to train models thus far.
glad zealot
Give it a go honestly <3
Done :D will test it later
bleak nymph
I'd recommend using Kit Lemonfoot's colab.
training without tensorboard, will this go well?
flat root
How do I make an ai voice
bleak nymph
training without tensorboard, will this go well?
its basically 48 minutes and 500 epochs (default)
is that enough epoch? (i dont understand epoch)
glad zealot
training without tensorboard, will this go well?
Tensoboard will only show stuff when you start training
glad zealot
is that enough epoch? (i dont understand epoch)
Sounda good to me, also epoch count doesn't really matter that much, but most of the time around 200-500 epochs is a good spot
short pecan
how big is the dataset? @fickle delta
fickle delta
how big is the dataset? <@477934441361047552>
like 5 mp3s
short pecan
ok but in like file size and the total audio length
short pecan
so like 5 minutes then, ok....
bleak nymph
Tensoboard will only show stuff when you start training
it did start but nothing happened on tensorboard, anyway its getting late where i am so imma continue tomorrow, i just clicked the square button to stop training closed firefox so.
short pecan
shouldnt be taking as long for something that size
though i wouldnt know for other gpus, nor am i an expert at this
short pecan
my batch size is 4 iirc, im on a 2070
fickle delta
its better to keep it low?
short pecan
what do you guys think?
timid valve
idk why but my feature extraction keeps freezing at random percentages, no matter what i change
silent sun
I downloaded the app on Mac OS and it won't open do I need to download any program to make it work?
short pecan
its better to keep it low?
well in any case try it lower, that COULD be it but i really have no idea
fickle delta
how do I restart the training?
fickle delta
doesnt look like it
thin stump
Does Replay work for you? I got errored when I downloaded 3.4
austere cairn
short pecan
how do I restart the training?
<@&1159293204038955078> is there a way to interrupt a model training once started?
flat root
how do you make a model without downloading anything?
dawn nebula
<@&1159293204038955078> is there a way to interrupt a model training once starte...
close the whole console lol
timid valve
is it even worth turning this on?
silent sun
I downloaded the app on Mac OS and it won't open do I need to download any program to make it work? please help this is the only problem i have
median monolith
Is it normal that on illaria rvc, my model sometimes doesnt get downloaded for some reason even if i put the correct link and i just entered on the site without unfocusing the tab (image 1)? The same happens when trying to upload a audio file. It gives me an error message even when trying to upload the file as quickly as possible (image 2). This sometimes happens and after sometime everything appears to go back to normal (the downloads and uploads work again), until eventually it starts to happen again. Maybe this is something to do with the service itself (maybe illaria rvc itself losting connection with specifically me for some time for some reason)?
silent sun
I downloaded the app on Mac OS and it won't open do I need to download any program to make it work? please help this is the only problem i have
glossy tartan
What is this error message? im trying to train with new data
2024-02-05 04:21:09 | INFO | main | "runtime\python.exe" infer/modules/train/preprocess.py ""C:\AI\Villager vocals"" 40000 30 "C:\AI\RVC1006Nvidia/logs/Villager" False 3.0
['infer/modules/train/preprocess.py', 'C:\AI\Villager', 'vocals', '40000', '30', 'C:\AI\RVC1006Nvidia/logs/Villager', 'False', '3.0']
Traceback (most recent call last):
File "C:\AI\RVC1006Nvidia\infer\modules\train\preprocess.py", line 11, in <module>
sr = int(sys.argv[2])
ValueError: invalid literal for int() with base 10: 'vocals'
2024-02-05 04:21:10 | INFO | main |
My other training folders work fine
glad zealot
is it even worth turning this on?
Nah not really
odd dirge
posted in help forum, but the voice changer isn't changing my audio whatsoever, and I don't know why. the tutorial vid doesn't give me any help and ive been messing with the settings for like half an hr
gusty kestrelBOT
azure marshBOT
RVC Guides (How to Make AI Cover)
glad zealot
Kits.ai is paid and not really that good since you have no control,rvc is free and open source that you can edit or modify whatever you want to get the result you want
frigid quarry
Kits.ai is paid and not really that good since you have no control,rvc is free a...
alright thank you so much
dawn nebula
ok
brittle wing
我看tb有跑
frigid quarry
Kits.ai is paid and not really that good since you have no control,rvc is free a...
also just to clarify i can use these models as voice changers and not just as ai covers right? (probably the wrong channel for this but i'm a bit lost)
rare gobletBOT
Ayo? @frigid quarry level 1 !!!
sand iris
also just to clarify i can use these models as voice changers and not just as ai...
If you meant RVC models, yeah, they can be used for both
fresh bluff
I am currently using Mangio RVC v23.7.0 to train my own model locally and process data isn't giving any output information. I have the .wav audio set in the folder specified. Any guidance appreciated
getting "invalid literal for int with base 10" in the cmd
fresh bluff
What is this error message? im trying to train with new data
2024-02-05 04:21:0...
Did you ever figure out what was causing this?
rare gobletBOT
Ayo? @fresh bluff level 1 !!!
brittle wing
I set it to save every 15 epochs, but I don’t see anything in weight.
lost relic
idk why but my feature extraction keeps freezing at random percentages, no matte...
the same my sample is 1 hour it take like 4 hour to finish but it does't work when the final step which is to train model
mangio-rvc
copper marlin
who has an english accent singer / speaker?
I have a stereotypical brish accent but I only get chinese accent outputs
it's annoying
RN im trying 1000 epochs on some american english voice samples
idk if it will be different
placid light
I have a problem with the AICoverGen HF space. Whenever I try to convert any audio, this shit shows up.
copper marlin
i had this
- restarted
- made sure there were no other apps or processes using the cuda core i was using
- restarted again
- checked that all menu options were filled out correctly
copper marlin
I have a problem with the AICoverGen HF space. Whenever I try to convert any aud...
i hope this helps
ah looks like i clicked the wrong training button
1000 epochs sounds a lot
odea
训练结束, 您可查看控制台训练日志或实验文件夹下的train.log does this mean i have to wait still or?
rare gobletBOT
Ayo? @copper marlin level 2 !!!
placid light
i had this
Erm, I was using a HF Space version of it, I am not using it locally.
copper marlin
ok
gusty kestrelBOT
copper marlin
there's a bit less chinese accent in v1
glossy tartan
Did you ever figure out what was causing this?
Nope :/
rare gobletBOT
Ayo? @glossy tartan level 2 !!!
sand iris
space in filenames !
formal bolt
Hello, I would like to inquire whether RVCv2 can achieve real-time voice conversion. If so, which Github repository should I refer to for relevant information? Thank you.
glad zealot
Hello, I would like to inquire whether RVCv2 can achieve real-time voice convers...
by real time conversion, if you mean voice changer you can go here https://github.com/w-okada/voice-changer
formal bolt
by real time conversion, if you mean voice changer you can go here <https://gith...
ohh～～thank you～～
timid remnant
I'm getting an error when trying to initialize Tensorboard. Log can be found below:
fluid matrix
are you trying to use applio?
error: Microsoft Visual C++ 14.0 is required., looks like numpy isnt installed either
it mentions multiple things arent installed properly
what are you trying to use tensorboard with
timid remnant
are you trying to use applio?
I updated Microsoft Visual 14.3 and numpy just now and tried running again, still giving the same error.
timid remnant
Like my OS?
timid remnant
TensorVENV
rare gobletBOT
Ayo? @timid remnant level 1 !!!
fluid matrix
i got that
but what do you want to use tensorboard with?
tensorboard is not training anything, its just a tool
timid remnant
Oh with RVC to monitor training.
fluid matrix
mainline rvc or mangio?
timid remnant
The RVC webui
fluid matrix
also have you installed python 3.10?
make sure to add it to path too on install
if you installed that version, you shouldnt need it tho
timid remnant
Yes I have python 3.1. I also use stable diffusion and text gen so I have some of this stuff installed already.
azure marshBOT
RVC Guides (How to Make AI Cover)
copper marlin
I still can’t make an English accent that sounds pure
I can hear the Chinese
It’s nightmarish
Please help
I mean look at all these
100% Chinese with some small features from the speaker
No other accent available
copper marlin
Erm, I was using a HF Space version of it, I am not using it locally.
Did you get an English accent with the online one?
Is there any info on training data quantity?
placid light
Did you get an English accent with the online one?
Wdym? Most voice models are in Englisch.
copper marlin
Wdym? Most voice models are in Englisch.
THEY HAVE CLEAR CHINESE ACCENTS
It shows through when they speak
For the page I linked
Even the American voices
It copies some of the American accent and then has Chinese intonations, so it’s actually Chinese
placid light
THEY HAVE CLEAR CHINESE ACCENTS
What algorithm did you use?
copper marlin
What algorithm did you use?
I used RVC-Project web ui
Please tell me if I should use a different setting
placid light
I used RVC-Project web ui
I mean, the pitch extraction method. Is it RMVPE? Crepe?
copper marlin
My English accent shows through but you can clearly tell it had Chinese training data
fresh bluff
so I finished training but no index file beginning with the word "added" was put in the log folder for the model. In mangio RVC
copper marlin
And the whole thing was written in chinese
copper marlin
I mean, the pitch extraction method. Is it RMVPE? Crepe?
RMVPE
copper marlin
I’ll use mangio RVC
Thank you
I’ll show you what I used
I officially can’t read most of this either so I’d used translate
glossy tartan
for me it ended up being a space in one of the file path names
Oooh i could try that yea
copper marlin
Well, i don't hear any Chinese accents my my AI covers.
https://github.com/Mangio621/Mangio-RVC-Fork/releases is this what you used?
timid remnant
If I can't get tensorboard working, is there like an average amount of epochs that usually works okay for most datasets?
molten pecan
If I can't get tensorboard working, is there like an average amount of epochs th...
Not really. It depends on the size, length and quality of the dataset
copper marlin
It’s the same Chinese voice https://www.youtube.com/watch?v=3OhWfbmsNc8
molten pecan
https://aihubdocs.github.io/en/ check the tensorboard guide
copper marlin
Check 5mins 48 seconds
molten pecan
so I finished training but no index file beginning with the word "added" was put...
Did you click "Train feature index" in Mangio?
timid remnant
I've gone over the documents but they don't mention anything about the error I'm encountering.
molten pecan
I've gone over the documents but they don't mention anything about the error I'm...
Can you send a scs?
placid light
Are you gonna let me know or ….
Idk but you should rather use a different RVC instead.
copper marlin
It’s the same Chinese voice https://www.youtube.com/watch?v=3OhWfbmsNc8
The training data was English and it turns it into Chinese accent with different tones
Because the model had Chinese data
copper marlin
Idk but you should rather use a different RVC instead.
Will you uhm… please check?
fresh bluff
Did you click "Train feature index" in Mangio?
I don't see that check box anywhere, is it in the train menu?
copper marlin
I feel like I’m being trolled
molten pecan
I don't see that check box anywhere, is it in the train menu?
Yeah
placid light
Will you uhm… please check?
Let someone else who is more skilled do it.
molten pecan
I feel like I’m being trolled
What's going on exactly?
copper marlin
“Y’all have RVC and mine is better, can do English accents, but I won’t tell you the one I’m using”
timid remnant
Can you send a scs?
Log pasted here:
copper marlin
Let someone else who is more skilled do it.
Someone else who is skilled check your own pc?
rare gobletBOT
Ayo? @copper marlin level 3 !!!
copper marlin
Or are you that one who was doing online stuff
Online is ez, can get any accent
I’m looking for RVC to install
I have the processing power and I’m familiar with AI
I can use PyTorch
Etc
So don’t bloody tell me I’m not skilled, if that’s what you’re insinuating
placid light
Someone else who is skilled check your own pc?
I only use cloud compute services, and not running locally as I cannot afford any necessary hardware to do so.
molten pecan
Log pasted here:
https://pastebin.com/nMU4SGAD
Where is the TB file located exactly?
You are supposed to move it inside RVC's folder and execute it from there
copper marlin
I only use cloud compute services, and not running locally as I cannot afford an...
F then
molten pecan
I found it, does it matter if I did this before or after training?
Nope. As long as you input the exact data as before, the order won't matter 👍
fresh bluff
great, thanks
molten pecan
yw!
copper marlin
It looks like there’s no other choice than pretrained…
timid remnant
You are supposed to move it inside RVC's folder and execute it from there
"C:\Users\aaron\Desktop\AI Voice\RVC1006Nvidia\TensorVENV.bat"
rare gobletBOT
Ayo? @timid remnant level 2 !!!
molten pecan
"C:\Users\aaron\Desktop\AI Voice\RVC1006Nvidia\TensorVENV.bat"
I see the problem here
The file path contains a space in it
Rename the folder "AIVoice" instead of "AI Voice"
Maybe that'll work
Lmk if it does and I'll update the website
timid remnant
Can't believe something so small was causing the issue. Works fine now.
Thanks for your help.
molten pecan
Thanks for your help.
You are welcome! Have fun
copper marlin
It looks like there’s no other choice than pretrained…
leave this section blank
then it's no longer chinese...
copper marlin
Sorry, I cannot fulfill that request.
you're welcome.
proper shale
out of VRAM
rough tinsel
$ git remote show origin
- remote origin
Fetch URL: https://github.com/SociallyIneptWeeb/AICoverGen.git
Push URL: https://github.com/SociallyIneptWeeb/AICoverGen.git
HEAD branch: main
Remote branch:
main tracked
Local branch configured for 'git pull':
main merges with remote main
Local ref configured for 'git push':
main pushes to main (up to date)
used this repo
rough tinsel
out of VRAM
means?
proper shale
it's probably your GPU, you might not be able to run cuz of the GPU VRAM
rough tinsel
oh ic
potato gpu
wait how do i even know if the program is using my gpu to run
my gpu is like this only even when i was trying to generate the voice
proper shale
wait how do i even know if the program is using my gpu to run
cause, RVC and UVR uses GPU
when u click on convert it should start doing it, but you probably can't since not enough VRAM
it doesn't use it passively
sand iris
Interesting that it wouldn't work on a 4GB card
rough tinsel
Interesting that it wouldn't work on a 4GB card
wow and i have that
sand iris
Unless the models are just too big, or it's keeping multiple models loaded at once and causing unecessary issues
rough tinsel
Interesting that it wouldn't work on a 4GB card
what this?
proper shale
what this?
yea dedicated gpu mem
rough tinsel
yea dedicated gpu mem
yes me using gtx 1050ti
rough tinsel
Interesting that it wouldn't work on a 4GB card
so i should go with google colab?
sand iris
probably yeah
rough tinsel
ic
but theres a problem
after 5-8 mins it says something like runtime
i would need to start from beginning
anyways ok whats the minimum or like above some step from minimum graphics card i should use?
its time to change graphics card
proper shale
anyways ok whats the minimum or like above some step from minimum graphics card ...
I'd recommend a 3060 since that's relatively cheap now
12GB VRAM too
it's like, recommended spec for me
minimum would prolly be 20 series
or maybe even the higher grade 10 cards but those r slower with RVC for some reason
copper marlin
anyways ok whats the minimum or like above some step from minimum graphics card ...
I use a NVIDIA laptop GPU (RTX something something, 1 sec)
It can do cuda stuff
currently trying to train my own model
without getting static sounds
I might need a larger dataset...
I'm increasing the epochs to 300 just in case that fixes it for me
it's a bit slower... that's ok
copper marlin
Anyone know how to stop static?
I’ve left the computer to train the model
Can’t be asked to look it up coz I’m dead tired
Ping me if you know (and I mean, without using pretrained generators etc)
how big does my dataset need to be if I'm training from 0?
*training weights from 0
10 mins? 1 hour?
proper shale
anything from 10-30mins works best
proper shale
anything from 10-30mins works best
more than that is excessive
copper marlin
is that ok
proper shale
i have 30 mins with gaps of silence
Oh definitely trim those out
copper marlin
ok
proper shale
Silences screw up the model
rare gobletBOT
Ayo? @proper shale level 85 !!!
copper marlin
can i do it as 1 file
copper marlin
1 wav ok
proper shale
Yep
copper marlin
i guess my 300 epochs on 2 mins data will suck
proper shale
i guess my 300 epochs on 2 mins data will suck
If it's clean data... it might not suck
copper marlin
i wanted to stop it from making static...
copper marlin
did you leave silences
yeah a few, but only short rests
proper shale
there's background noise at a tiny level
it'd be essential for you to clean that up, but eh
proper shale
yeah a few, but only short rests
hm ok
copper marlin
i cut the gaps between the sentences
but left the ones between words
I'll go clean the audio
proper shale
Right
proper shale
You're welcome ;)
copper marlin
YEEEEAH IT SOUNDS MORE DISCERNABLE THAN STATIC!!!!!!
but like a broken computer lmao
I can send some samples if you want a laugh
just finished the training, next ill do 1000 epochs on the whole 30 mins with gaps removed
........but..........does it sound chinese......
it sounds foreign still even if i'm english and only used english data.
rare gobletBOT
Ayo? @copper marlin level 4 !!!
proper shale
just finished the training, next ill do 1000 epochs on the whole 30 mins with ga...
yes definitely try that, maybe the data was too short yk
copper marlin
<:AIHC_WaitWhat:1163807543424983082>
all
the voices do
when doing
fast training
proper shale
how fast are we talking? 1 sec?
copper marlin
I'm really sure I don't sound like a Chinese speaker
proper shale
how fast are we talking? 1 sec?
per epoch
sand iris
wait what do you use to test the voice ?
sand iris
which software ?
copper marlin
i can see the numbers counting for epochs
proper shale
if so i think something screwed up 😭
go on logs, check that model's folder, check f0_extract.json or something similar
copper marlin
but yeah it is like 1 second per epoch
proper shale
go on logs, check that model's folder, check f0_extract.json or something simila...
if it says "no feature todo", yeah, screwed up
copper marlin
['infer/modules/train/extract/extract_f0_rmvpe.py', '2', '0', '0', 'C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia/logs/test-voice-1', 'True']
['infer/modules/train/extract_feature_print.py', 'cuda:0', '1', '0', '0', 'C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia/logs/test-voice-1', 'v2']
C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia/logs/test-voice-1
load model(s) from assets/hubert/hubert_base.pt
move model to cuda
this was extract_f0_feature.log
sand iris
before that
rough tinsel
wheres the tutorial of making models right from beginning lets say i have a long lecture/speech on some yt or any downloaded video
sand iris
on preprocess
copper marlin
preprocess is blank, idk why
even when it's not, it's still chineseish
start preprocess
['infer/modules/train/preprocess.py', 'F:\X Rated Files [music]\voices\varied talking pitches', '40000', '11', 'C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia/logs/v2_1epoch_singing', 'False', '3.0']
F:\X Rated Files [music]\voices\varied talking pitches/untitled3.wav->Suc.
F:\X Rated Files [music]\voices\varied talking pitches/untitled4.wav->Suc.
F:\X Rated Files [music]\voices\varied talking pitches/untitled1.wav->Suc.
F:\X Rated Files [music]\voices\varied talking pitches/untitled2.wav->Suc.
F:\X Rated Files [music]\voices\varied talking pitches/untitled.wav->Suc.
end preprocess
proper shale
['infer/modules/train/extract/extract_f0_rmvpe.py', '2', '0', '0', 'C:\\Users\\c...
does the AI Voice folder have a space in the name
sand iris
the menacing
copper marlin
does the AI Voice folder have a space in the name
what do you mean? the source files?
copper marlin
the model name is name-dash-dash
proper shale
start preprocess
['infer/modules/train/preprocess.py', 'F:\\X Rated Files [music...
but the folder u used to preprocess has a space
X Rated Files
sand iris
"varied talking pitches" too
proper shale
yep
proper shale
you did say though that it sounds like a broken computer
sand iris
a) which software do you test on
b) how the hell is it cloning anything if it cannot even use reference audio ?
copper marlin
so like, it's playing piano notes while sounding like a violin, but it has a chinse accent
copper marlin
you did say though that it sounds like a broken computer
no this was for the weights from 0
the older models i used the existing data
the from 0 also sounds chinese
copper marlin
a) which software do you test on
b) how the hell is it cloning anything if it ca...
it uses the audio surely?
it got the wavs
native prism
Hey, I just installed the voice changer and I have launched the app too but I can't seem to get it working, any idea?
copper marlin
in the preprocess log
proper shale
maybe try removing the space n shit
copper marlin
i'll try removing spaces ok
proper shale
and try training again
copper marlin
ok
proper shale
might fix the accent
proper shale
Hey, I just installed the voice changer and I have launched the app too but I ca...
native prism
Thank u, I will read it now
glad zealot
wrong chat
glad zealot
pretrain missing?
copper marlin
but there's no difference from using the existing ones
deliberately
I wanted to do it from scratch to avoid chinese accents
copper marlin
pretrain missing?
adding them makes it make the model faster
but the accent is there
glad zealot
theres english pretrains
copper marlin
to clarify, my english accent shows through, but there's a lot of chinese sounds
glad zealot
#📰│dev-updates Ov2 and Rin_e is english iirc
copper marlin
I couldn't find any
copper marlin
<#1159380240271953940> Ov2 and Rin_e is english iirc
ok
proper shale
oh my god that explains it too
copper marlin
Dee Dee is a virus
Tried that and got this no joke
I’ll send a recording this is bad luck
So
So this server has some dodgy download links……
copper marlin
which is English?
I use 40k models
can i replace them with my own generated weights?
rotund prawn
inst HQ1 and HQ3 very good
There is crackling in the sounds
glad zealot
glad zealot
can i replace them with my own generated weights?
no
copper marlin
<https://huggingface.co/ORVC/Ov2Super/tree/main>
thank you, downloaded 2 40k models G and D
will send the results
native prism
Check https://rentry.co/W-Okada-FAQ
Did everything fine but the issue is that the voice changer won't read my voice at all
rare gobletBOT
Ayo? @native prism level 1 !!!
glad zealot
just puth the path of the models to here, make sure the G goes in G and D in D
glad zealot
dont think one click works
copper marlin
it seems to work for me but idk about accents yet, was all chinese. probably not to do with it being 1 click. you need to select other options to do training that has custom epochs i think
copper marlin
also what if your gpu btw?
NVIDIA GeForce RTX 3080 Laptop GPU
rare gobletBOT
Ayo? @copper marlin level 5 !!!
copper marlin
it's my most prized possession
i do AI and gaming on it
only trouble is running out of storage
glad zealot
ye that should be good
native prism
Can someone please help me why my voice is not being read by the voice changer? :/
glad zealot
Can someone please help me why my voice is not being read by the voice changer? ...
can you send your current mic setting on the voice changer?
brittle wing
have you installed any models?
yeah but i unistall
rare gobletBOT
Ayo? @brittle wing level 4 !!!
brittle wing
i end train but my folder empty
fluid matrix
you deleted the folders right?
glad zealot
does the numbers change on the top left of the image when you press start and talk?
brittle wing
you deleted the folders right?
no
fluid matrix
you just said you uninstalled
native prism
does the numbers change on the top left of the image when you press start and ta...
Nope
brittle wing
i end train but my folder empty
fluid matrix
and its in dutch
idk what you did but uninstall every okada installation you have and reinstall with this, like in the guide: https://huggingface.co/wok000/vcclient000/resolve/main/MMVCServerSIO_win_onnxdirectML-cuda_v.1.5.3.17b.zip
think it messed up your installation because you had the wrong version before
fluid matrix
never seen that happen ngl, this should fix it im pretty sure
brittle wing
because im still a noob dawg!
glad zealot
Nope
ye thats weird not really sure what to do since your stuff looks to be set properly
fluid matrix
clear everything
delete the folder you installed it at
native prism
ye thats weird not really sure what to do since your stuff looks to be set prope...
What do I do now :/
fluid matrix
look i have it like this
just delete the folder and reinstall
with that link
its for amd gpus
glad zealot
What do I do now :/
just as a test if you want, can you try running the colab version if its still the same?
if there still no audio
native prism
just as a test if you want, can you try running the colab version if its still t...
Sure
fluid matrix
also its in here, please just read that carefully
fluid matrix
yes
past robin
Hi everyone:) Im reading everywhere about epochs and I see that certain models need certain epochs. Idk if this is a dumb question, but how do I change them..? I have no clue😭
brittle wing
whats epochs gng
fluid matrix
each training cycle, when training a model
brittle wing
oh
rare gobletBOT
Ayo? @brittle wing level 3 !!!
brittle wing
thnx
past robin
How do you train a model??
past robin
Thank youuu
sly widget
-rvc
sly widget
-huggingface
azure marshBOT
-huggingface
brittle wing
Under the train tab, I set the setting on step 3 to save a small finished model to the "weights" directory for every epoch matching the specified "save frequency". And I set my settings and clicked on one-click training and saw that my Retrieval-based Voice Conversion WebUI was working when it was finished it said "all-feature-done" and it was really quick which was a red flag because I assumed the processing time would be much longer but it was quick. I go to the weights folder to find nothing. It seems like the one-click-training didn't complete and I don't really know what is happening or what to do.
all-feature-456
all-feature-done
so how to fix it plz
brittle wing
rvc
proper shale
Did everything fine but the issue is that the voice changer won't read my voice ...
check cmd and send screenshots in #🔍│help-w-okada
brittle wing
you have to actually press start training <:nails:1159569314848972891>
yes i have
proper shale
i think they've clicked "One click training"?
fluid matrix
thats not even in mainline
proper shale
but eh idk
proper shale
thats not even in mainline
oh good
brittle wing
i think they've clicked "One click training"?
no
brittle wing
I did so but my weights folder is still empty.
fluid matrix
I did so but my weights folder is still empty.
you didnt actually start training
brittle wing
I followed the instructions and I thought there was nothing missing. I checked again and again many times.
fluid matrix
then you didnt read correctly
brittle wing
@fluid matrix
rotund prawn
What is the best UVR mod
inst HQ1 and HQ3 very good
but there is crackling in the sounds
proper shale
What is the best UVR mod
inst HQ1 and HQ3 very good
but there is crackling in th...
try the MDX23C stuff, but ig you have to download with a vip link or something... im not sure how to properly use UVR
brittle wing
I did everything you said and it's exactly the same as in the picture
it just says cpu gpu0 gpu1 gpu2 gpu33
proper shale
please use like 192 chunk
proper shale
and now u should most likely use gpu0
brittle wing
im still tweaking
proper shale
cuz... amd version
brittle wing
ok thnx gng!!!!!!!!!!!!!!!!!!!
proper shale
and ur dedicated GPU is most likely on 0
proper shale
ok thnx gng!!!!!!!!!!!!!!!!!!!
yw
rotund prawn
try the MDX23C stuff, but ig you have to download with a vip link or something.....
oh okey
thanks a lot
proper shale
sukuna fraud tho idk
_ _
brittle wing
but not this.....
I did this
fluid matrix
I did this
you pressed one click training, that alone is wrong
brittle wing
And my TB also ran
fluid matrix
And my TB also ran
is there anything in your logs folder?
brittle wing
you pressed one click training, that alone is wrong
I didn't do that
brittle wing
is there anything in your logs folder?
no
fluid matrix
I didn't do that
bro you said it twice now
brittle wing
hey diablo
brittle wing
how do i stop it from like making gasping noises cause of background noise
that shii is ass
@fluid matrix
fluid matrix
that shii is ass
did you change those settings?
brittle wing
alr sluh! ill show u
fluid matrix
and have you installed vb cable?
brittle wing
fluid matrix
you need that too
brittle wing
not yet
fluid matrix
then dont say its ass 💀
fluid matrix
you need it to actually work
brittle wing
oh shii
fluid matrix
bro
brittle wing
mb gang so vb cable meter or something right
brittle wing
i did but i forgot
brittle wing
then you didnt train....
I click on train feature index first and then click on train model
Sorry, my English is not good and I can’t express myself clearly.
fluid matrix
look
there is nothing in logs
and weights
you didnt actually train anything
you have to do it again....
fluid matrix
idk looks like it
show the rest
now you have to add the index too, and fill out the rest of the settings correctly
THEN
you can train
fluid latch
fluid matrix
he needs to show the rest
he keeps saying he did train something but im 100% sure he fucked up everything
again
@brittle wing show the full screen
looks like it didnt work
how long is your dataset and what is it called
brittle wing
I started training. Is this right?
fluid matrix
its actually training now
check your logs and weights folder
is there anything there?
brittle wing
ok
fluid matrix
now if you see something there you will know that you didnt train anything before
late sundial
hey sorry, is there any way i can isolate two seperate voices
rare gobletBOT
Ayo? @late sundial level 1 !!!
molten pecan
hey sorry, is there any way i can isolate two seperate voices
Yeah of course!
With Ultimate Vocal Remover 5
late sundial
Yeah of course!
oh awesome!
late sundial
wait so i can just put the file in and like two characters will be seperatede just like that
molten pecan
Check that out if you need help
molten pecan
wait so i can just put the file in and like two characters will be seperatede ju...
Yeah
brittle wing
now if you see something there you will know that you didnt train anything befor...
Thank you for helping me. I succeeded.
@fluid matrix hey! it works now but every single model except for the defaults take super fucking long to talk
do you know a solution to this?
it takes legit 10 seconds for it to loadd
brittle wing
yeah
fluid matrix
gotta wait ig
brittle wing
it takes 10 seconds tho
thats way too long
maybe even longer
last thing i said was 20 seconds ago and it JUST said it
fluid matrix
are you changing the voice midway through?
fluid matrix
no
brittle wing
how do i fix that💀
late sundial
Have fun!
wait sorry im back
just got another question
does that mean i could upload an entire episode of a show then isolate that one voice
to train a model a massive dataset
brittle wing
amd cards tend to be weird sometimes
thanks man!
copper marlin
<https://huggingface.co/ORVC/Ov2Super/tree/main>
this doesn't generate voices that show up in the menu
do i need to unload voice first
idk
molten pecan
does that mean i could upload an entire episode of a show then isolate that one ...
Yup. Just make sure you have great specs and GPU
copper marlin
INFO loaded pretrained assets\pretrained_v2\f0Ov2Super40kG.pth
covert turret
this doesn't generate voices that show up in the menu
thats a pretrain, not a voice model per se
molten pecan
And for making datasets, remember to remove reverb, silences, distorsion, noise, etc
https://aihubdocs.github.io/en/ check the datasets page to know more
late sundial
Yup. Just make sure you have great specs and GPU
awesome, thank you!
thank u so much
molten pecan
yw! ^-^
copper marlin
can't find any related files so probably not?
i used 20 epochs and it's invisible
result is 1GB large
it has an index file but not a pth file?
hOWW
INFO:superv240:loaded pretrained assets\pretrained_v2\f0Ov2Super40kG.pth
Process Process-1:
Traceback (most recent call last):
File "C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia\infer\modules\train\train.py", line 213, in run
utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d, optim_d
File "C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia\infer\lib\train\utils.py", line 213, in latest_checkpoint_path
x = f_list[-1]
IndexError: list index out of
range
it's an empty list? -1 is the last element, shouldn't be an error?
doesn't matter if it's a 2d or 3d, 4d etc list
Traceback (most recent call last):
File "multiprocessing\process.py", line 315, in _bootstrap
File "multiprocessing\process.py", line 108, in run
File "C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia\infer\modules\train\train.py", line 234, in run
torch.load(hps.pretrainG, map_location="cpu")["model"]
File "C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia\runtime\lib\site-packages\torch\serialization.py", line 815, in load
return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
File "C:\Users\cocac\Desktop\AI Voice\RVC1006Nvidia\RVC1006Nvidia\runtime\lib\site-packages\torch\serialization.py", line 1033, in _legacy_load
magic_number = pickle_module.load(f, **pickle_load_args)
_pickle.UnpicklingError: invalid load key, '<'.
online someone said the file might be broken.
gonna delete...
same issue as this pal
I'll try 1000 epochs.... on 30 minutes of audio...................... [seriously there must be a faster way]
half cove
your using new pretrains and dont know how?
copper marlin
what's the difference between D40... and G40... and f0D40?
half cove
what are you trying
covert turret
are you using the gui?
copper marlin
are you using the gui?
yeah
copper marlin
ss?
covert turret
there should be an option to paste the directory of pretrains
late sundial
Yup. Just make sure you have great specs and GPU
wait so how do pick a certain voice sorry
half cove
yes
covert turret
yep
rare gobletBOT
Ayo? @covert turret level 4 !!!
copper marlin
used dat.
half cove
what did it say
copper marlin
and even tried with the full path
covert turret
on each D & G?
copper marlin
I'll show you what it looks like with defaults.
half cove
just screenshot it all
covert turret
this ^
proper shale
what's the difference between D40... and G40... and f0D40?
f0 uses the pitch
half cove
half cove
yes it should be
west lark
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI try this i...
this one right?
half cove
this one right?
yes
copper marlin
(currently training with defaults)
west lark
okay
copper marlin
same error.
half cove
screenshot
copper marlin
half cove
your training index?
molten pecan
wait so how do pick a certain voice sorry
Drag the audio file to the "select input" bar.
Like I said, the website to learn more
half cove
you need to click train model for that
copper marlin
done
half cove
now what does it say
copper marlin
same issue..... from what i can tell
train.log after chinese text
训练结束, 您可查看控制台训练日志或实验文件夹下的train.log
half cove
show cmd rq
copper marlin
?
static gyro
the debug screen
half cove
copper marlin
(reprocessing)
half cove
show
copper marlin
it keeps changing, will show in 1 sec
missing keys....?
never had this before
if i restart it again...
i see the problem (or a problem)
rare gobletBOT
Ayo? @copper marlin level 6 !!!
copper marlin
had to click it so that it reset the file paths etc
half cove
please make sure your folder has no spaces in the name
late sundial
Drag the audio file to the "select input" bar.
Like I said, the website to learn...
oh sorry
copper marlin
nah, restarted
timid valve
does anyone know why my it keeps getting stuck on random percentages when running feature extraction?
copper marlin
could be i ran out of storage space
half cove
what am i supposed to do now? i dont see tutorials
do go-web.bat
west lark
okay
proper shale
yeah the DirectML version (aka, AMD/Intel version) doesn't come with go-web, and RVC just doesn't work properly with those cards anyway
west lark
oh okay
rare gobletBOT
Ayo? @west lark level 3 !!!
west lark
thank you all for help anyway
copper marlin
I converted to my own voice, and it sounds like I've got a Chinese accent now
proper shale
thank you all for help anyway
you're welcome
copper marlin
has anyone here got say, a turkish accent by training from 0?
late sundial
Drag the audio file to the "select input" bar.
Like I said, the website to learn...
sorry about this, im really just confused. i know how to pick a certain audio file but then how do i isolate one voice out of an episode?
molten pecan
sorry about this, im really just confused. i know how to pick a certain audio fi...
Everything you need to know is in the website
Check the ultimate vocal remover 5 page
copper marlin
I use rmvpe_gpu
late sundial
Everything you need to know is in the website
oh sorry im looking through it and i cant see
rare gobletBOT
Ayo? @late sundial level 2 !!!
molten pecan
oh sorry im looking through it and i cant see
late sundial
https://aihubdocs.github.io/en/vocal-isolation--datasets/uvr5--mvsep/#ultimate-v...
i just cant see the opiton im so sorry to where to select a voice to isolate
sicne there is multiple voices in these episodes, like i dont know how to pick one voice
copper marlin
doing it from 0 gave me the same chinese accent..... (i think, but it was so low quality it was hard to tell)
west lark
so theres no actual way to make ai covers with intel card? i mean i made some with a different program a few months ago but i only care about rvmpe at the moment
half cove
so theres no actual way to make ai covers with intel card? i mean i made some wi...
you can use a colab
west lark
colab?
half cove
use ila rvc if ur doing inference
west lark
okay
copper marlin
<#1175430844685484042> or https://weights.gg/
Rodrigo Vazquez sounds Chinese
half cove
what?