✨│ai-help

[knotty moth]

again please try the recommendation I gave you above and see that it could disprove your obsolete argument

this is not 2023 anymore y'know?

[runic smelt]

Does anyone have a good voice changer recommendation i should use? i have a 5060ti 16 gig. and i need to use certain ai voice models.

[viral mason]

it runs wayyyy better

I don't understand this "fronted" thing

if you're using old models most likely it will not change when using a different voice changer

from how long ago?

i'd suggest retraining it with applio, using the legacy core v1.5 pretrain

[knotty moth]

it's never recommended to have non-ASCII characters in the names

[viral mason]

https://tenor.com/view/pet-rock-rock-erosive-gif-19734322

https://tenor.com/view/speak-gif-9578947274896178319

have u tried vonovox?

fair enough

regular rmvpe is better

if you're on amd ur only choice is onnx

so I assume regular one is better

[knotty moth]

hey @analog obsidian , I want your opinion on this person arguing that old version is better as well as those other points

[viral mason]

please correct anything dumb I've said lyery

[analog obsidian]

onnx only exists for compatibility, regular rmvpe is better

for nvidia use .pth, onnx uses more cpu and might affect pitch accuracy

[knotty moth]

also take a look at the whole convo

[analog obsidian]

same, the f0 accuracy gets affected

it's better on amd

but it tracks pitch worse

[knotty moth]

@analog obsidian also that argument, where 131k extra on old wokada is equivalent to the maxed extra and 16k extra is equivalent to 2.7s on deiteris/tg-develop fork

[analog obsidian]

max extra technically is better since it's closer to local inference (not realtime) but adds more delay

[knotty moth]

also convince them that the performance of 3060 or 4060 isn't that bad, except probably for some competitive gaming purpose where a single digit ms delay difference could determine between winning or losing the game

[analog obsidian]

as a 4060 user im not happy with the performance of realtime with this thing lol

yea better gpus have less delay

[knotty moth]

no, none of AIB 4070 models have x8 lane, only x16

[analog obsidian]

there was a discussion about this in another server, yea actually pcie speed affects realtime delay too, crazy

[knotty moth]

you prob mean 4060 Ti

I dont think any normal users would deliberately hamper the pcie performance from x16 gen 4 to x8 gen 3 or 2, except some crazy reviewers

even the older Z490 motherboards and the AMD equivalent i forgot have pcie 3 x16 lane. forgot how much the performance hit but it'd be mostly cpu bottleneck rather than pcie (that happens on having not enough vram and forcing to load texture data from system ram in gaming)

depend on what pcie gen

[knotty moth]

it seems to make sense that pcie bandwidth difference has more impact on ai workloads than just gaming

[analog obsidian]

ye someone said upgrading their pcie 3 to pcie 4 decreased their realtime delay by about 30ms

random obscure fact (?

[knotty moth]

and even both X870 motherboard and RTX 5070/higher have pcie 5 x16 support, though not sure if it is fully utilized yet

slower gpus are less impacted by pcie difference, e.g. 4060 Ti has some little impact by x8 gen 4 and any slower gpus are less impacted

[hallow thistle]

Because many people here have had "Hide extensions for known file types" enabled on their Windows Explorer, which hides the file extension of the file for cleaner look. When try to open the .zip.001 (or .zip in disguise which is actually .zip.001 in the first screenshot) whether with 7-Zip or WinRAR, the error often occurs, including the corrupted archive error, to much confusing. The only workaround to the root of the cause is to go to Explorer, go to Settings, "View" tab, navigate "Hide extensions for known file types" and then uncheck it, click Apply and then OK. Check your Explorer again, and now every file will show its real file extension respectively, which includes .001. If a file doesn't show its file extension after setting, try press F5 to refresh. This workaround not only fixes the paradoxical ".zip.001" issue, but it can also help you easier to distinguishing each file type (like a fake .pdf file but it actually ends with .pdf.exe), though with a bit messy look for some.

[hallow thistle]

For more pronounceable performance, there's NVIDIA GeForce RTX 4080. The performance can be far little or unnoticable for typical users if either GPU is downgraded to use PCIe 3.0 x16 (or x8), but anything below x4 and eventually x1 (of any PCIe version equals or below 4.0) will have a noticeable to significant performance drop.

[hazy forge]

hello i have a vps and i want to run rvc on it how do it do it is there anyt tutorial out there

[rotund horizon]

What is the very best option for a 5090, I want very accurate and real time voice changing for use on voice chat

[lusty holly]

how can i get it to work on a 5060

[hallow thistle]

Try Tg Develop's W-Okada fork. https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/ https://github.com/tg-develop/voice-changer/releases/tag/b2397 Or Vonovox. https://github.com/dr87/Vonovox/releases/tag/v1.6.9 Older versions like b2332 and v.1.5.3.18a are not compiled for RTX 50 series.

[hallow thistle]

@lusty holly you too as well.

[hallow thistle]

Are some websites blocked in your country's internet? What is your PC GPU? And are you looking for Applio RVC or W-Okada voice changer?

[lusty holly]

thank you ✌️

[lusty holly]

vonovox says "Cannot start voice conversation - No GPU available. You can still access settings and configuration." even though i have a 5060

[knotty moth]

try update the Nvidia driver and make sure to be using the 169 version or you can also try manual setup https://docs.aihub.gg/realtime-voice-changer/local/vonovox/#precompiled-setup-nvidia-on-windows

[lusty holly]

ok

[knotty moth]

also the prerequisite as well

[lusty holly]

it still wont work

are there any video tutorials?

[warm python]

Deepwoken player spotted

[brittle wing]

is a high simple rate better?

or worse?

[graceful tinsel]

Guys how to make a model

helpme

PLZZ

[hallow thistle]

!howtoask

[patent trellis]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 4060 8gb vram desktop)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a legal, safe & ethical community, we will NOT provide help for:

• ANY illegal activities.
• NSFW/Porn.
  Requests for these topics may be ignored, not helped and result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.
• Don't Ask To Ask.

[hallow thistle]

There are options. Weights.com can train a voice model there, but the result would usually lower than that the one in Applio RVC and that's really it. What is your PC GPU?

[autumn cedar]

Inference with Applio 3.6.0 on windows 10. I get errors in console

The .wav file producing that error was 60 seconds long. If I use only a 10-second long .wav file, there is no error. How can I tweak Applio settings, or Windows settings, in order to work with longer files?

[feral sigil]

why is the voice changer app not opening?

[pallid chasm]

what is epocs

[hallow thistle]

See #🧬│ai-chat message.

[full helm]

Bro

[warm python]

?

[true nacelle]

hey uhh.. is downloading isolated converted vocal isn't free anymore or is it a bug?- a week ago it was until i come back again-

[viral mason]

Never suggest weights for training it's bad

[odd harbor]

hey can someone help me whit the voice changer set up? and where to download

[astral jungle]

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[astral jungle]

-train

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[astral jungle]

-train

[odd harbor]

whats all that?, are these voice changers?

[viral mason]

what gpu do u have

[odd harbor]

AMD Ryzen 5 5500

[viral mason]

u can use Wokada tg fork, it's what I use rn it's pretty cool

[odd harbor]

where do i download that

[viral mason]

right here

https://software.muzychenko.net/freeware/vac470lite.zip

https://github.com/tg-develop/voice-changer/releases/download/b2397/voice-changer-windows-amd64-dml.zip

[odd harbor]

i download both?

[viral mason]

the first one is to have the voice changer connect to games and discord ect, second is the voice changer

[viral mason]

ye

[odd harbor]

thanks can i use these voice models then?

[viral mason]

u can use any of them except ones made with refinegan, there aren't many of those at all tho

[viral mason]

did u figure out how it works btw or do u need further help

[odd harbor]

I’m downloading right now if u need help I will tell u

[viral mason]

oki

[latent cypress]

can the audio ‘watashi no koi wa’ also be used as a benchmark for male models?

[analog obsidian]

That audio is mainly used to see how well the model can generalize high notes. For a male voice, I don't think it's very useful

You could make an inference using pitch -6 with that audio if you want, but I would recommend looking for better audio samples to see how well the model generalizes

[viral mason]

what's that

this is good for testing a models range of pitch bc it has both a female singer in it and a male singer

[patent trellis]

❓ Help Guidelines

If you need urgent help, please checkout our AI Hub Docs or ask for help here following the [Guidelines](#1402790586028789830 message)

[latent cypress]

aaa okay okay, thanks man

[latent cypress]

the audio that rvc devs often use to test the models they’ve made (cmiiw)

[latent cypress]

thanks, i’ll use this audio to test my model

[viral mason]

I need the audio to listen to bc I don't have the best memory

[latent cypress]

aaa i see, wait a sec

[viral mason]

np I have plenty of random samples in a folder if u want

[viral mason]

ohh yea that one

[latent cypress]

whoaa, i’ll let you know if i need more samples to use for testing

[analog obsidian]

For benchmarking, I recommend trying different audio files. For example, your model may sound good with the watashi audio file, but it may have other problems. The most common one I see is when they glitch when inferring a sound they didn't learn very well

[latent cypress]

https://tenor.com/view/spongebob-patrick-star-noted-notes-gif-17474838830648097856

[latent cypress]

is testing the weights one by one only for finding a sound that’s not too harsh on the ears, or can it also be used to find results with better pronunciation?

[analog obsidian]

yea can be used for both, i had cases where one early epoch glitched in a part where the latest epoch didnt

there will be a point were the pronunciation is going to eventually stop improving tho

[wild bobcat]

is there a way i can make an ai voice model out of drum toms

and make them actually pitch correct

[analog obsidian]

why not just using a daw rather than making a rvc model

[wild bobcat]

because im curious

[latent cypress]

so is there actually a shortcut to finding the best epoch? like jumping back 10 or 100 epochs. or is the only way really to check them one by one? i want to look at the tensorboard graph, but i’m training a singing model with batch size 4, so the graph is super messy

[wild bobcat]

i can do it in a daw yeah but im curious on if that works

[analog obsidian]

it works

[wild bobcat]

is there an example of this or do you know itll work because they have a f0

[analog obsidian]

them having a f0 is the reason why they works

[wild bobcat]

i see

[analog obsidian]

this guy makes drums models

#1455640521195388969 message

[wild bobcat]

i wonder how these drums are made because theyre not "singing"

theyre atonal just like drums

[analog obsidian]

sadly there is no way to tell, i dont like rvc graphs

rvc graphs are not really that useful

to me the most important one is fm, you dont want that one to have high values

[wild bobcat]

whats fm?

[latent cypress]

haha same, people keep telling me to pay attention to this graph and that graph, but in the end i still end up testing the epochs one by one lol

[analog obsidian]

but the only way to decrease the fm value is to have a big dataset so.. yea

ah and lowering the batch size decreases fm iirc

[latent cypress]

from loss_avg or loss_avg_50?

[analog obsidian]

imo they're really a waste of time

[analog obsidian]

unironically i use the mainline graphs lol

[latent cypress]

https://tenor.com/view/oh-the-misery-everybody-wanna-be-my-enemy-cat-gif-25172445

[analog obsidian]

idk maybe use avg 50 (?

[wild bobcat]

offtopic but i wonder if rvc is ever gonna be continued

we have rvc v2 but nothing of v3 i think

?

[latent cypress]

idk why, but fm from avg 50 is more noisy than the regular loss

[analog obsidian]

rvc is abandoned, development seems to be ceased

[wild bobcat]

oh

so like its just up to the users to "continue" it

[analog obsidian]

both have a very bad trend

[latent cypress]

[analog obsidian]

high values and rising

[analog obsidian]

ye, basically, applio have their own "v3" but i did not liked it at all

[wild bobcat]

whats different about it

[errant sand]

https://tenor.com/view/gitapro3-gitagita-gitapro1-gitajah13-gif-3421945902795778238

[analog obsidian]

a new vocoder and a fusion between mpd and mrd

[latent cypress]

dataset issue?

[wild bobcat]

what about it audibly?

does it pitch weirdly, weird consonants

noise

[errant sand]

@analog obsidian I loved reading your LegacyCore thread in voice-models, it was like a "Making of" documentary.

[analog obsidian]

how big is the batch size?

[latent cypress]

4

[analog obsidian]

xD

[latent cypress]

and 18 minutes dataset

[analog obsidian]

sometimes i type a lot oops

[errant sand]

https://tenor.com/view/the-office-michael-scott-thank-you-bow-steve-carell-gif-5009516783019270794

[analog obsidian]

graphs don't matter anyway, does the model sounds good? thats what matter the most

can we just ignore the fact rvc graphs exists?

they just add fear

lol

i only use them when im training pretrains to monitor stability

[errant sand]

So I had to jump in and ask you a few questions, if you don't mind...

[analog obsidian]

wat

[latent cypress]

real lmao

*299 epoch

[analog obsidian]

this sounds good to me, yea ignore the loss lol

fm raising and having values of 10 is not good tho but at the end the model still sounds good

[latent cypress]

i'm using your 1.5 pretrain btw

[analog obsidian]

having values of 10 in a pretrain is very bad, but for finetuning i dont think it really matters

[latent cypress]

the power of legacy core

[analog obsidian]

no one will know why they chose the worst dataset ever for the og pretrain

lol

m4singer was already a thing in 2023

[errant sand]

I read in the guide that in TensorBoard if I filter by g/total, I can watch til it starts to trend up and then stop training. But I just saw you say that you don't rely on those graphs. Does it mean I should continue training and manually check the epochs after the graph starts going up?

[analog obsidian]

g/total is such a stupid metric because it adds the FM metric, which is not a generator metric

😭

it's a discriminator thing

[errant sand]

OK, maybe that's why I wasn't happy with the results I got, I probably stopped too early

[analog obsidian]

yea dont watch the graphs, they just add confusion

overtraining is something you hear, not something you see in a graph

g/total raises when fm raises, and fm raises when the discriminator got too good

[latent cypress]

yea, train the model and watch netflix (forget about monitoring the graphs)

[analog obsidian]

im sorry for all of the misinformation out there

fm going up and having values over 10 is not good... but at the end you can still have a good model despite having such crazy graphs

[errant sand]

My first training session was with the original pretrain cos I wanted to use it as a quality baseline before I try your improved LC2.5. I stopped training after about 230 epochs cos the graph told me too😢 and the pronunciation wasn't as crisp as I hoped. From now on, I'll listen to all the epochs

[analog obsidian]

The most stable og pretrain is the 32k, i know because i have done multiple tests with all of them

[errant sand]

Darn,I used 40k cos that was my dataset. I thought, the higher the better

[analog obsidian]

it is better (quality wise) but the og pretrains have critical flaws

[viral mason]

honestly just checking is the best, check around after 100 tho

[analog obsidian]

40k and 48k do not work properly with nsf hifigan, they have sibilant and breath problems

[latent cypress]

found it already ✌️ 😁

[analog obsidian]

but if you want to try high sr i'd recommend legacy core 1.5 48k pretrain (even if you have a 40k dataset)

#1453062294446805244 message

[latent cypress]

yk, i’m often torn between choosing 32k or 40k for my singing model. but for some reason, after trying the 32k pretrain, i’m really starting to doubt training a 40k model lol

[analog obsidian]

40k is fine, the quality is there, but it's noisy and the breaths are just worse

[errant sand]

Is LC1.5 still out there? Didn't see it on your HF. I read in that thread that 2.5 is the best?

[latent cypress]

how? won’t the results be messed up if the dataset isn’t 48k?

[analog obsidian]

yep, click the link i sent, the pretrain is there

about which one is the best im not sure... every dataset reacts different to a pretrain

[analog obsidian]

no, who told you that?

[errant sand]

Yep, I thought if you use 48 pretrain, the code auto resamples your dataset to 48?

[latent cypress]

the last time i did that was with the og pretrain, but i don’t know how the results would be with your pretrain

back in 2023 ig

[analog obsidian]

the model just learns the frequency cutoff of the dataset

the sibilants are noisy and the breaths are weird but thats just because hifigan is bad at 48k

lol

[latent cypress]

ooo i see

[analog obsidian]

the quality is good (assuming you have a high sr dataset)

[latent cypress]

what is sr? something like sdr?

[analog obsidian]

sample rate

[latent cypress]

aaa

[errant sand]

Some of my dataset is higher quality (20kH highs according to Spek) but the rest tapers off at around 16k, but I feel it's still good enough for singing cos I usually roll off those harsh highs anyway.#

[viral mason]

is this a good way to add non voice stuff to a dataset without them just being in there, the coughing was cut and put somewhere where it sounds natural instead of being on its own

@analog obsidian what do u think? is editing the audio to be more natural going to give better results for human voices?

[stray copper]

what batch size should i use for 17 minutes of 41.1khz dataset
should I do 6 (3 on kaggle)

[errant sand]

The experts seem to have left, I've read in the docs.aihub guide that if your dataset is less than 30 mins, then go for the batch size of 4. But, the experts here have been debunking that guide all day, so maybe it's not the definitive source I thought it was...🤔

[stray copper]

i used batch size 4

no worries

[errant sand]

MJ forever, anyway👍

[analog obsidian]

i dont think nsf hifigan can generate natural sounding coughs, let me check

edit: it can generate coughs but eh, its not that good

it will not decrease the quality of the model, just in case

also, you're doing it wrong, it should be like this:
cough voice

the sound needs to be closer to the voice, you put the sound too far away

[analog obsidian]

batch 4

[stray copper]

i did

[unique dew]

I downloaded Wokada voice changer but the files :

• voice-changer-windows-amd64-cuda.zip.001
• voice-changer-windows-amd64-cuda.zip.001
  this files is corrupt, what happen?

[analog obsidian]

oh you're mixing 32k data and 40k data in the same dataset and want to train 40k? hmm, I don't know how “good” this is, but I know someone else tried it and had good results. It's not something I would personally do, though xD

[analog obsidian]

I know that for pretrains it's fine because someone else already tried training a 40k pretrain with a dataset that have some speakers with 32k data

I did the same for the 48k version of 1.5
sid0 is 48k, rest varies between 32k, 40k and 48k

[analog obsidian]

https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/ this is what most use nowadays i think

that and vonovox

[analog obsidian]

Perhaps the result will be a model that sounds like 32k despite having a sample rate of 40k (? idk im just guessing

[errant sand]

I simply didn't have enough high quality singing samples. I only had around 13mins which resulted in a subpar first voice model. So I prepared more samples, but they sound like they "lack the high end" and you could see on Spek that they don't have anything over 15k. So, yeah,I was hoping to train my next model using both sets of samples (of course converted to one final sample rate (40k). But still the underlying samples are of different quality. It makes me sad that you think it might not work. I didn't even think of that. And you recommend 15mins minimum dataset for your pretrain.😢

I guess, I could convert ALL samples to 32k and train on your 32k pretrain.

[analog obsidian]

It will work, hmm but since most of the data is 32k maybe the model is going to sound mostly "32k", but i have no clue, i can't predict it

[analog obsidian]

You can try to experiment and train using the 40k pretrain

doesn't hurt to try

[errant sand]

Yep, experimenting is key, I guess. The good thing is that you recommend so few epochs with your pretrain. It won't take too long!😀

[analog obsidian]

also dont worry you can train a dataset below 15 min, however you may have some problems with stability and pronunciation, it depends on how many phonemes there are in your dataset. If there is a wide variety of words, there may not be such serious problems

For example, it is entirely possible to have problems with pronunciation even in a larger dataset

[errant sand]

My high quality dataset (13mins) didn't have enough variety unfortunately. That's why the model struggled with some phenoms.

[errant sand]

Phenoms! lol. Phonemes!

[analog obsidian]

My recommendation is 40-100 epochs, your best model will be around that area

Of course, that also varies. There are cases where others have found their best epoch to be 200

That's another bad thing about rvc, we can't guess when the model will sound “right."

it just randomly happens

[errant sand]

That's great. Oh, wow. Ok, I'll run it for 300 epochs then just in case. And then will manually check.

[analog obsidian]

yeah it's all experimentation

[errant sand]

Also, do you recommend including speech, for a singing dataset. And how much?

[analog obsidian]

If the speech is good (not monotonous), add everything you got

[errant sand]

I feel, most speech ends up being monotonous

[errant sand]

Not theatrical enough, if you see what I mean

[errant sand]

Typical interviews and such sound monotonous

[analog obsidian]

hmm you could add some monotone data to help the model with the pronunciation

I don't know how much would be the “right” amount

[errant sand]

Gemini 3 says 40%, speech for RVC, but do you think it knows about this topic?

[analog obsidian]

it has some knowledge of the inner workings but it doesn't know anything about training models*

[errant sand]

It mentioned that you need speech for pronunciation (just like you confirmed). It said it's very important because sustained singing doesn't really gives that

[analog obsidian]

true, singing alone can't provide good pronunciation

What RVC does is average the audio files of your dataset (I think Gemini knows this too) and create an unique voice model

[errant sand]

Funny story. When I read about your LegacyCore pretrain, I eventually got to the part where someone said (YourLocalWorm) that you are not going to make any more public updates. I was bummed. So I checked your Hugging Face. Got the 2.5 and saw that there was a model you uploaded 1 day ago. (It was a couple of weeks back) The model didn't have a name, but I still downloaded that hoping it's some fresh new thing you've been working on. Sorry for stalking. 😀

[analog obsidian]

I changed my mind lol, but i wont be releasing any new pretrains for the default rvc architecture because is old and bad, but i'll be training a new pretrain for a new architecture

[errant sand]

Is that the next big thing for singing?

[analog obsidian]

imo seems thats the case, the model generalizes singing very well

also 48k reproduction is way better than nsf hifigan

[errant sand]

Are you one of the first who jumped on it? I haven't heard it mentioned anywhere

[analog obsidian]

it's not here, is from another rvc server

[errant sand]

I see, let us know when you have something cool Please.

[analog obsidian]

sure

[errant sand]

Would you say your internal tests blow everything else you had done out of the water?

[analog obsidian]

btw we better gotta talk in #🧬│ai-chat coz we're kinda spamming this channel lol

[errant sand]

Cool. I think I've asked you all the questions I could remember for now. Thanks. Let's switch channels

[unique dew]

I just downloaded from that link too, and i was choosing "Download NVIDIA on Windows" then downloaded that files i told before. I cant extract the file and told me if the files corrupted (sorry for my bad eng)

[craggy bough]

i cant install it or cant open start_http because the smart app controll is blocking it but i dont want to turn it off because i cant turn it on isnt there another way to get it ?

i am trying to install the the voice changer

isnt there another copie of it or another way to install it ?

[viral mason]

old voice changer

what gpu do u have

[craggy bough]

oh

idk i got it from github

[craggy bough]

a good rtx one

MMVCServerSIO_win_onnxgpu-cuda_v.1.5.3.18a

this is the latest one no ?

[viral mason]

do u know the number s of it orr

[viral mason]

nah that's Ancient

ye that is good

you should try vonovox

-rt

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Tg-Develop Fork, with extra features, but it supports only Nvidia GPUs on Windows 10/11 unlike other options that have wider support. and without cloud options GUIDE

• Wokada Tg-Develop Fork

A personal fork (modified version) of Wokada Deiteris Fork, it just adds some Quality of Life improvements to it like supporting Spin Embedder and Audio Effects. Don't expect too much about it since the creator made it originally as a personal project. GUIDE

• Wokada Deiteris Fork

Deiteris' fork (modified version) of wokada that doesn't get updates anymore. GUIDE

⚔️ Wokada Tg-Develop Fork vs Vonovox

For Windows Nvidia, Both Wokada Tg-Develop fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Tg-Develop Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[craggy bough]

yeah but a laptop one tho its not as much as powerful lol

[viral mason]

oh

[craggy bough]

ohhhhhh didnt know

[viral mason]

:(

[craggy bough]

is it made by the same people ?

[craggy bough]

yeah 🙁 wbu what u got ?

[viral mason]

completely different person, but it's a huge upgrade compared to that one u have

still realtime, still ai voice changer

[craggy bough]

oh

[dire cedar]

yo chat so ive done everything correctly and the voice changer works but when i try to speak w the voice changer it just cuts everything out and only makes one sound and it has huge delay

[craggy bough]

it exist in here so i think its legit ? and safe ?

[viral mason]

https://github.com/dr87/Vonovox/archive/refs/tags/v1.6.9.zip

[viral mason]

it is yes

[viral mason]

did you get it off a yt tutorial?

[dire cedar]

yeah

[craggy bough]

alr thx

[viral mason]

get rid of it, it's outdated

what gpu do u have

[dire cedar]

i use intel

[viral mason]

nevermind

good luck

[dire cedar]

😔

[viral mason]

intel is the one thing that just cannot run ai at all

[craggy bough]

btw

[viral mason]

gotta spend for nvidia if u want the best

[dire cedar]

yh i use a old laptop

[craggy bough]

what guid i should use there is 3 here lol

[dire cedar]

i dont got cash

[craggy bough]

the guid gave me a complete diffrent download which one i should be using ?

the one u gave me isnt the same ig ?

also it says its not real time voice changer ?

RVC does NOT mean realtime voice changer. RVC means Retrieval-based-Voice-Conversion.

@viral mason can u help ?

explaining

[viral mason]

the vonovox one

[viral mason]

the one I gav u is the most recent one

[unique dew]

someone can suggest me what voice changer should i use?

[craggy bough]

okay thx i will give it a try

even if there isnt really a tutorial

BRUH

same issue

its blocked

[viral mason]

your pc is weird

[viral mason]

what gpu do u have

[craggy bough]

smart app control is the weird

[viral mason]

idk what that is but I don't use it

should prob turn it off

[craggy bough]

its a windows thing

its exellent to protect a device

[manic seal]

Hi

[craggy bough]

but it cant be annoying sometimes

i think its blocking it because its a bat file from unkown source

however once you turn it off you cant turn it on until you redo your windows

so yeah

[viral mason]

like restart your pc?

[craggy bough]

nope reinstalling windows

[viral mason]

that's uh

really

bad

sounds like a virus

should get rid of whatever that is

[unique dew]

rtx 5070 laptop

[viral mason]

could try out vonovox
https://github.com/dr87/Vonovox/archive/refs/tags/v1.6.9.zip
https://software.muzychenko.net/freeware/vac470lite.zip

[craggy bough]

naaaah bro this exists in every pc that have wibdows how do you dont know about it ?

[viral mason]

I don't use it ig

[craggy bough]

its just stupid microsoft policy

[unique dew]

how to use it?

[viral mason]

run setup, then run start

and for vac lite run setup64

install vac lite first

[craggy bough]

but its exellent tho it works on karnel level technically the best but can be annoying cuz it can block stuff that are not bad

it doesnt get talked about enough tbh

[unique dew]

what is it?

[viral mason]

connects voice changer with games and discord

[viral mason]

I never used it tho, didn't know it even existed until now

[unique dew]

setup or setup64?

[hardy crow]

anyone know how to get exclusive mode to work on vonovox

[craggy bough]

lol good for u ig

its not known that much ig but it always works in the background

unless its off

didnt know much about it too but yeah since it caused me that issue i may let it go too

idk why its on anyway it should be in the normal mode which isnt that much stricted

anyway thx

[viral mason]

64

[viral mason]

Did u pay?

[hardy crow]

no

[viral mason]

What do u mean then by exclusive mode

[viral mason]

Ah

Not sure, would be better to ask in the official server

[viral mason]

Probably should tbh if it's causing you so much issues

[craggy bough]

it didnt until now tbh

i finally manged to start the setup file

however

[unique dew]

what should i change in Vonovox settings?

[craggy bough]

its not downloading just stuck on

• 0.1/2.9 GB 1.3 MB/s eta 0:36:49

its so slow damn

and idk if i can do the same with the start file or not

i hope this actually worth it

can i just cancel it btw ?

[viral mason]

Awesome

[viral mason]

No you have to run setup first then start

[craggy bough]

i mean about the blocking thing

also its stuck

[viral mason]

Oh

Idk how to help tbh

[craggy bough]

or at least slow asf

u sure the software is safe ?

[viral mason]

It is yes

The creator of it is in this server

[craggy bough]

okay

[craggy bough]

ayo

how much storage does it take ?

its taking forever to install it

@viral mason finally the setup is done but the start isnt doing anything for me ???

its been installing for an hour

and now ive got this shit when i try to run start

C:\Windows\System32>runtime\python.exe launcher.py
The system cannot find the path specified.

C:\Windows\System32>pause
Press any key to continue . . .

[viral mason]

I'm not sure how to fix any errors so u should join the official server for vonovox

[craggy bough]

okay

[brittle wing]

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[civic lotus]

why cant i upload models

[odd harbor]

@viral mason hey so I now am in the browser on the voice changer but it doesn’t pass through to discord I don’t even know if It’s working at all

[viral mason]

could u send screenshot?

[graceful vale]

@viral mason so i installed everything, the vc audio cables, and the models and the ai doesnt change my voice at all

[odd harbor]

I send them to u private cause I can’t send pictures here

[viral mason]

did u get it off youtube

[graceful vale]

Get what off of youtube

I did exactly what the video said yeah

[viral mason]

the voice changer

[graceful vale]

I came here and got the model

[graceful vale]

Tup

Yup*

[viral mason]

change cpu to gpu

[viral mason]

it's old then, what gpu do u have

[graceful vale]

5060

But my homeboy jus got it off the same video recently and his works

[odd harbor]

I did now but I still can’t hear anything when I go on dc voice test

[viral mason]

u and him should use vonovox, well actually what gpu does he have

[graceful vale]

not sure about him, but hes 0 help

[viral mason]

0 help?

should ask him, but what u got is outdated, should use smth better

https://github.com/dr87/Vonovox/archive/refs/tags/v1.6.9.zip

to use this just run setup, then run start

[odd harbor]

What setting do I have to use on the vc are mine right?

[graceful vale]

is iit the same process to setup?

[odd harbor]

I changed gpu

@viral mason or could u send me ur setting so i can copy

[viral mason]

run setup, then run start

[graceful vale]

ok how do i add models

[viral mason]

there's some slots at the top u add the pth and then add index at the bottom where it says index

[graceful vale]

then i add vb cable?

[viral mason]

yup, regular mic input and output as vb cable

[graceful vale]

u got good settings?

[viral mason]

crossfade at 0.15 is really all u need to change tbh

[hallow thistle]

"Exclusive mode" is an audio mode of WASAPI and ASIO. WASAPI has both shared and exclusive modes, while ASIO only has exclusive mode. When you have "exclusive mode" on, only one program (like Vonovox in this case) can output audio at the time, while other programs (like Discord) will be muted if they are on the same audio system (e.g. Realtek HD Audio). This mode is not very ideal if you want to hear other programs at the time if they all on the same audio system. When audio API is set to "WASAPI" and you have "exclusive mode" off, the program will use "shared mode" instead.

[hallow thistle]

For "exclusive mode", one of the good uses of this mode is when you're making music on a DAW software, and you prioritize bit-perfect audio over other having to go through system's mixer (like one in Windows) and letting other programs to output audio at the same time.

[void flume]

You're making an assumption. I was just answering threads, where I found answers where none where given. I'm not interested in asking others to do it for now.

There was no need to drag my attention twice either

Multiple people will use search and check threads, not just the OP

Because of that having info in the right place is handy, so I thought

[hallow thistle]

When you go in one's thread that has already been long solved long ago and say anything there, it can considered "invading someone's thread" or potentially "necroposting". While these older threads have already been solved, you better wait for a newer one and then you go answer with your information, or at least have a talk with staff member(s) that could or might open possibilities for Beatrice voice models in "AI Hub by Weights". If you are so confused with why these happening, you better ask a mod or admin to clarify, I'm a Helper and it won't be my duty to abuse or change guidelines into something else.

[void flume]

I could say the same, as in; apparently you are not sure whether or not it is actually considered as such, but you didn't ask the rest of the team before immediately going hostile.
And likely it is not.
The rule reads: Search the Forum: See if a solution to your problem already exists. Be sure to not invade their post.
This is not about sharing information, instead it is about taking over the question with your own.

I should also note that, the website reads:
Contributions
We'll appreciate any feedback, big or small.

Just saying that I do not mean what I said as hostile, simply as feedback.

If it is unwelcome, then I will not continue.

but at the same time, it means that the rules aren't clear enough and require clarity.

[soft flame]

can anyone help me download the voice changer idk how to i search youtube and it aint working
dm me

[hallow thistle]

What is your PC GPU?

[soft flame]

nvidia

kinda old tho

gtx 1070

[hallow thistle]

Try Tg Develop's W-Okada. https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/ https://github.com/tg-develop/voice-changer/releases/tag/b2397

[soft flame]

what do i click when inside

so many stuf

[hallow thistle]

Click "voice-changer-windows-amd64-cuda.zip.001" and "voice-changer-windows-amd64-cuda.zip.002".

[soft flame]

where is that

[hallow thistle]

GitHub.

[soft flame]

oh but it says amd here tho

[hallow thistle]

Don't get confused with the term "AMD64"; it's a CPU architecture where it's the same thing as x86-64 (Intel).

[soft flame]

oh ok

[soft flame]

so i need both ?

[hallow thistle]

Must.

[soft flame]

ok

[hallow thistle]

These two files are split parts of a single zip. Check if "Hide extensions for known file types" is disabled in your Explorer. Use WinRAR or 7-Zip to open and extract the .zip.001 one.

[soft flame]

ok

what abount hte second one

[hallow thistle]

.zip.002 is another part of .zip.001, you don't need to open it. Just open .001 one and the program will temporarily combine them into one.

[soft flame]

but do i need to extract it

[hallow thistle]

[soft flame]

huh

[hallow thistle]

When you ask for every step-by-step step, the progress would be too little to going or ending up in mess. I can focus on settings in the program, not these.

These steps are so simple to follow. You simply extract to somewhere like "D:\MMVCServerSIO" or "C:\Users\your username\Downloads\MMVCServerSIO".

[soft flame]

how do i start it i extracted it

[hallow thistle]

Oh, really?

[soft flame]

wat

how to start

[hallow thistle]

Go inside "MMVCServerSIO" folder, you'll see a file named "MMVCServerSIO.exe". That's the program itself, you simply double click on it to launch the program.

[soft flame]

this ?

[hallow thistle]

Always make sure you read the guide carefully alongside steps, before start asking something too simple.

[soft flame]

ok

so i added a model to it but how do i connect and use it with discord and other platforms

[hallow thistle]

For how to download a voice model from #1175430844685484042, go to #1175430844685484042, go to a post/thread there, spot a link (usually being Hugging Face) and then click it.
Because another person here asked how to download a voice model, yet they deleted their message because the query might sound stupid.

[soft flame]

ik how to download it

but how do i get it to work

[hallow thistle]

This https://cdn.discordapp.com/attachments/1159290139609137264/1456249974525526185/image.png?ex=6959a839&is=695856b9&hm=2fcbbfb436b8185706268db3f3e1b48f3733828e2be264ebb8a053a36daf62a9& and this https://cdn.discordapp.com/attachments/1159290139609137264/1455527099539521568/image.png?ex=6959a9fe&is=6958587e&hm=9df1ee2b38711b83d2ef6259f8829acd3206a5b4c0e8d3711775fa5ed55e4d63& might give you some ideas.

[soft flame]

i dont see it

its not an option

[hallow thistle]

Because you didn't download and install Virtual Audio Cable. Go download the program from this link. https://software.muzychenko.net/freeware/vac470lite.zip

[soft flame]

so after i extract it is should work

?

[hallow thistle]

That's not how it work.

[soft flame]

uh

[hallow thistle]

Unlike "MMVCServerSIO", Virtual Audio Cable needs to be installed in a traditional way. Inside "vac470lite" folder, double click on "setup64.exe" to install the program.

[soft flame]

ok

[soft flame]

ok i installed it and it shows up in discord but i dont hear anything when i speak

when i do the mic test

[hallow thistle]

Screenshot your voice changer.

[hallow thistle]

Chunk: aronud 320 - 400 ms
Extra: 2.1 or 2.7 s
GPU: NVIDIA GeForce GTX 1070
Pitch extraction: rmvpe (not rmvpe_onnx)
Input: Microphone
Output: Line 1 (Virtual Audio Cable)
Monitor: your speakers/headphones.

Always make sure to check the perf number at top right of voice changer; if the perf number is red or yellow, try increase chunk value up until perf green. https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/#finding-my-own-settings-for-chunk-size-and-extra-processing-time

[soft flame]

i set everything as u said and its still not working

so weird

[valid ravine]

hey i also tried to follow along does it matter which version i download on amd gpu?

[soft flame]

yes

but idk what u needd for amd

@hallow thistle pls help :(

[hallow thistle]

What do you mean it's not working? You sure you have clicked on green "start" button already?

[hallow thistle]

What is your AMD Radeon GPU?

[soft flame]

im fucking stupid

[valid ravine]

9060xt

[hallow thistle]

Try Tg Develop's W-Okada fork. https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork https://github.com/tg-develop/voice-changer/releases/download/b2397/voice-changer-windows-amd64-dml.zip

[soft flame]

how do i decrease the ping

[hallow thistle]

See #✨│ai-help message again.

[valid ravine]

what do i do now? i downloaded it and it brought me to rvc voice changer

nvm

i got it, it was being weird it gave me two options of my mic and wasnt working with first one

[hallow thistle]

RVC (retrieval-based voice conversion) and W-Okada voice changer are two different programs, where the latter primarily uses RVC voice models.

[hallow thistle]

Here's your settings if you haven't set anything:
Chunk: around 60 - 70 ms
Extra: 2.7 s
GPU: AMD Radeon RX 9060 XT
Pitch extraction: rmvpe_onnx

Input: your microphone
Output: Line 1 (Virtual Audio Cable)
Monitor: your** speakers**

[fiery wing]

@dire drum

[astral jungle]

helo

[limpid trout]

Alright I need a voice changer for local running

This time I have a good pc with rtx 5060 ti and 16gb (ram goes brrr)

[hallow thistle]

Try Vonovox.

[limpid trout]

Uh how to get it

[hallow thistle]

https://github.com/dr87/Vonovox/releases/tag/v1.6.9

[limpid trout]

Uhhh
This has no tutorial in YouTube

[hallow thistle]

It's because you won't need it. There's a guide doc.https://docs.aihub.gg/realtime-voice-changer/local/vonovox/

[autumn cedar]

I have Applio working, but for inference it can only handle .wav files less than 30 seconds long. Are there any tweaks to allow longer files? I don't mind if the tweaks slow the inferencing down a little.

[autumn cedar]

How do I configure Applio to run in CPU mode? and how do I configure RVC-WebUI to run in CPU mode?

[quartz heart]

yyo

how do i ai voice in realtime

[void flume]

The current recommended method is using Tg-Develop's fork: https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/
Note: the instructions are partly incomplete and partly confusing, but it should get you set up at least. The process is basically very simple:
Download zip, extract zip, run server, click stuff in the browser window that opens up later.

[warm pond]

rvc isnt working but beatrice is

[honest coral]

Hey everyone 👋
I work mostly on the pre-automation side of things helping spot when a workflow or automation is about to amplify a broken decision instead of fixing it.

If you’re dealing with automations that “work” but still create problems, I’m happy to help think it through or look at assumptions before you build.

[hallow thistle]

What is your PC GPU?

[distant panther]

why is my delay so long

i only hear myself like 20 seconds later

[hallow thistle]

What is your PC GPU? And did you follow any tutorial or guide?

!howtoask

[patent trellis]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 4060 8gb vram desktop)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a legal, safe & ethical community, we will NOT provide help for:

• ANY illegal activities.
• NSFW/Porn.
  Requests for these topics may be ignored, not helped and result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.
• Don't Ask To Ask.

[distant panther]

my gpu is radeon 580 rx series

yes i followed the guide

[hallow thistle]

This AMD Radeon RX 580 would struggle even on any W-Okada version. You either try Tg Develop's W-Okada fork (b2397) DirectML https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/ https://github.com/tg-develop/voice-changer/releases/download/b2397/voice-changer-windows-amd64-dml.zip , or go for online option like Kaggle for more performance. https://docs.aihub.gg/realtime-voice-changer/cloud/tg-develops-w-okada-fork-kaggle/

[distant panther]

bruh

[hallow thistle]

[viral mason]

Your voice changer is older than the ancient Egyptians, what gpu does your computer have?

[latent knoll]

?

[sonic pier]

best ai to use for hyper realistic models?

[viral mason]

What even is that

[chilly furnace]

???

[latent knoll]

im tryna run applio on dione

[viral mason]

idk what dione is but applio can be run on kaggle, it's very simple

[swift thunder]

Which do you recommend I use, Legacy Core 1.5 or 2.5?

[rough meadow]

having an issue trying to setup microphone and such

cant send pic.

when I need to

[onyx remnant]

can someone help me.. my voice changer isn't working for some reason..! if someone helps me ill give them nitro

[rough meadow]

got it working last 2 - 3 yrs agp

ago

[viral mason]

your voice changer is super old, what gpu do u have

[rough meadow]

dw I just disconnected my headset and now it works I setup new just now so all is good

thx for the help tho.

[viral mason]

what gpu do u have?

[rough meadow]

anyway since you wanna know my GPU, 4060 Mobile.

[viral mason]

mobile?

[rough meadow]

yeah notebook

mb

[viral mason]

I'm confused

[onyx remnant]

nvidia

[viral mason]

u know the numbers orr

[onyx remnant]

3060

[crimson depot]

@viscid moss

DM me

[viral mason]

U should get Vonovox it's the best current for Nvidia

https://github.com/dr87/Vonovox/archive/refs/tags/v1.6.9.zip

https://software.muzychenko.net/freeware/vac470lite.zip

[waxen notch]

How do I set its parameters?

Batch Size: 8
Dataset Length: 30 Minutes
Hop Length: 64
Pretrain: DMR V1
Precision: FP32
Sample Rate: 32K
@patent trellis

[viral mason]

Those are settings it was trained on you can't set those, also sapphire is a bot

[waxen notch]

How can I use the same on the application?

[viral mason]

On what application?

[waxen notch]

Voice Changer

[waxen notch]

i ping you in ai help forum

[real depot]

Hello everyone, how are you? I’m new here, even though I’ve been around for some time looking for RVC models and pretrains. I finally decided to join in order to clear up some questions I have about this, and I’d like to thank in advance anyone who can help clarify things for me

[viral mason]

What things would you like clarification for?

[real depot]

I use Applio to train my models. At the very beginning, I used it with Colab until I was able to invest in a machine with a GPU. Since then, I’ve become more familiar with the process, using different pretrains, and yesterday I was able to test a model that I retrained using SPIN v2 as the embedder, and the results are quite noticeable in terms of speech clarity and overall quality. In this case, I used Legacy Core 1.5, which has a version designed to work with SPIN v2. My question is: do I necessarily need a pretrain specific to SPIN v2 or any other pretrain I might use, or will it work if I simply choose SPIN v2 instead of ContentVec?

[viral mason]

@analog obsidian did you make a version of legacy core that works with spinv2 I don't remember this

[analog obsidian]

yes

[viral mason]

Ok was just checking thx for the quick response 👍

[viral mason]

If you choose a pretrain made to work with spinv2 it will only work with spinv2 and vise versa

This is why original pretrain doesn't have a spinv2 model as it wasn't made to work with spinv2

[wraith sundial]

busco a un español que sepa generar ads con ai

[real depot]

It’s just that I thought pretrain models and embedding models were different things and that one didn’t depend on the other.

[wraith sundial]

spanish?

[analog obsidian]

they're different things but with a small dataset you cannot correctly finetune embeddings from a different embedder

so your model wont be able to speak unless you train like a 44 hour dataset minimum

but yes it's possible to train spinv2 embeddings with a cvec pretrain, its just that you need a really big dataset for the model to be able to speak

tldr just use a spinv2 pretrain if you want to train spinv2

[real depot]

Certainly, it really was a question I had, since pretrains for SPIN v2 have only started to appear now, and I thought I could use the pretrains that use ContentVec. Thank you for the clarification.

[real depot]

Your pretrains are very good. Congratulations

I used Legacy Core 2.5 before, and I already liked it a lot

[versed hull]

why is the web server

taking so long

[viral mason]

?

[versed hull]

its been waiting for web server

for a while

and it's not loading much

the VCC client

[versed hull]

its still just

loading

[nocturne granite]

Poop

Cundy

[viral mason]

https://cdn.discordapp.com/attachments/1326378532590456943/1419686397475819662/vivlos-uma-musume.webp

[brittle wing]

whats the best was to use realtimevoicechangerclient with vb virtual audio cable on a laptop that only has an i3 and intergrated graphics?

i dont mean voice settings i mean settings so that it runs as smooth as possible even if its delayed by 1 minute or smthn

[brittle bobcat]

anyway to make vovonox more sensitive?

[onyx galleon]

can someone help me the app wont open it keep saying wait web server 0-9999 and then it shows my ip and port

[graceful tinsel]

GUYS

HOW TO MAKE AI MODELS

TEACH ME

[hollow storm]

i have no idea how to download the voice changer at all

help

[azure imp]

yo is there ever a real time thing so i can talk while i play games?

[graceful tinsel]

hey bro how to make models

[hallow thistle]

I've asked you what your PC GPU is before, though I doubt if you still remember it.

[graceful tinsel]

no i dont remember🙏😭

and my gpu is so bad

gt 610🙏😭

we can do with cloud right

[hallow thistle]

NVIDIA GeForce GT 610 is very unlikely for interence or train anything locally. It's best to go with colud/online service.

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[graceful tinsel]

idk what to read

[hallow thistle]

If you don't want to pick one there, see https://docs.aihub.gg/rvc/cloud/applio-kaggle/.

While I can focus on installation and some settings, for preparing dataset it's best to ask a fellow voice model maker.

[graceful tinsel]

hmm

is it hard to make a voice model

and need lots of coding knowledge?

[hallow thistle]

Training one model won't be too hard if you aren't lazy.

[graceful tinsel]

hmmmmmmm like what

[autumn cedar]

I have a possibly helpful tip to share about using RVC WebUI, I have a very weak laptop and Windows 10. RVC kept crashing my system after a few minutes, so I tried this advice:

Increase TDR timeout: Open Registry Editor (regedit), navigate to
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers. Create/add
DWORD values: TdrDelay (set to 60 decimal) and TdrDdiDelay (set to 60M. Restart PC. This
extends the timeout to 60 seconds.

[hallow thistle]

https://media.tenor.com/X8l3uRqRz2cAAAAe/just-let.png

[graceful tinsel]

mine doesnt crash

[hallow thistle]

How can I answer your query? That's an awkward one.

[autumn cedar]

Increasing the TDR timeout worked! Now I can do inferencing on files up to 5 minutes long with no crashes, and the laptop doesn't heat up too much either

I was going to put my laptop in the refrigerator but that's a bad idea, condensation will cause water to appear inside the laptop

[hallow thistle]

This bro's laptop has NVIDIA GeForce MX, by the way.

[autumn cedar]

so take it me from boys, don't put your laptop in the fridge

[graceful tinsel]

i mean which part of it needs me to be not lazy

[hallow thistle]

First of all, go create a Kaggle account.

[graceful tinsel]

thanks I thought to book some tickets to antarctica and live there

[graceful tinsel]

not colab?

why no colab twin

[hallow thistle]

See https://docs.aihub.gg/extra/glossary/#google-colab.

[autumn cedar]

yes i also wish to know why Kaggle is preferred over Colab

[graceful tinsel]

william shakespeare of today💔🥀

[autumn cedar]

it requires a phone number?? Jaysus

[quasi condor]

yes kaggle requires a phone number which it sucks tbh

[graceful tinsel]

oh i forgot i already have an account in kaggle

[autumn cedar]

well then kaggle can suck my balls

[graceful tinsel]

no gif perms

[hallow thistle]

Gay.

[quasi condor]

eh atleast u can use two gpus at once

[graceful tinsel]

two balls💔😭🙏

[autumn cedar]

yes that's worth a few balls I suppose

[hallow thistle]

If you don't like Kaggle or a cloud server, because you refuse to do so, just try train anything on your old laptop and tell me how you feel about it.

[quasi condor]

two gpus with 16 gigaballs of vram

[autumn cedar]

I'm not doing training, I'm only doing inferencing with existing models

[hallow thistle]

Lie.

[autumn cedar]

I tried training once with Applio and the laptop got so hot I was sweating

[graceful tinsel]

and then the laptop gave bro some backshots

[hallow thistle]

Enough of off-topic topics, how about your progress on Applio RVC on Kaggle?

You asked for it, but you don't seem to do anything.

[quasi condor]

is ur laptop have a intergrated gpu or u have an actual laptop gpu?

[graceful tinsel]

mb mb mb im watching some anime wait sometime twin😭💔

[hallow thistle]

[autumn cedar]

tbh I don't know if it's integrated gpu or not, i bought the laptop used

[quasi condor]

oh wait i remember now ur gpu is a nvidia geforce 940mx

[autumn cedar]

how can I tell? let me pull up hwinfo

[hallow thistle]

NVIDIA GeForce MX, still less powerful than those GeForce RTX Mobile GPUs.

[autumn cedar]

yes geforce 940mx -- can that be "integrated" or is it actually a separate component?

[quasi condor]

yea ik its only good for low end games or minecraft

[quasi condor]

its separate

[autumn cedar]

ok

[hallow thistle]

Every NVIDIA GeForce GPU is a dedicated GPU. The integrated GPU is something like Intel UHD Graphics.

[autumn cedar]

Hwinfo says the geforce has 4GB graphics memory, but I read somewhere that might only be the equivalent of 2GB because 2GB might be "shared" ??

[quasi condor]

ur mobile gpu has only 2gb of vram sooooo i suggest u shouldn't use applio locally ;-;

[autumn cedar]

why does it say 4gb?

"shared" thing?

[hallow thistle]

Yes, there's W-Okada voice changer. What is your PC GPU? There's a specifive version you might like to try.

[autumn cedar]

I can get Applio to do inferencing but it heats up my machine MUCH more than RVC WebUi

[quasi condor]

cuz ur reading the shared memory if u look at dedicated gpu memory its 2gb

[autumn cedar]

i can't find anything about shared and dedicated gpu memory in this hwinfo utility, can you suggest another way i can see this information?

[quasi condor]

task manager

[autumn cedar]

k

in task manager performance, is this the tab you mean?

[quasi condor]

yes

[quasi condor]

oh tf o - o

4gb?

that exist?

[autumn cedar]

so it's 4gb dedicated?

[quasi condor]

yes

[autumn cedar]

whoo hoo I am God

[quasi condor]

but the geforce 940mx is a ddr3 gpu so its gonna be slow at inferencing

[autumn cedar]

not bad for a second hand laptop = )

[quasi condor]

ddr3 is outdated and old ;-;

[autumn cedar]

I am also outdated and old

[hallow thistle]

While Applio RVC actually detects your laptop GPU, when your GeForce MX GPU is 4GB variant, the program still struggles to do anything there because almost every GeForce MX wasn't made for AI. Not only that, GeForce MX has fewer CUDA cores than what GTX and RTX ones have.

[quasi condor]

its ddr3

that's also terrible

[autumn cedar]

I may get myself a reasonably good gaming desktop PC this year, what's a suitable NVIDIA card that isn't the most expensive?

my budget is only moderate

namari's circuits are overloading with my complex demands

[quasi condor]

gtx 1080 ti

its old but still good

[autumn cedar]

yes that looks affordable

[hallow thistle]

There are a few variants of GeForce 940MX that use GDDR5 or DDR3, I'm surprised there's a kind of entry-level GPU that uses a type of VRAM similar to the main PC RAM (DDR). But still very slower anyway.

[quasi condor]

there's also the rtx 4060 which its only 300-370 usd

[quasi condor]

.

[autumn cedar]

I assume these suggestions are good for AI as well as gaming?

[quasi condor]

yep

[autumn cedar]

cool

[cosmic spire]

I got my 5060 for just 320€ it's amazing for ai and gaming. (I only use it for AI workloads on my server lol)

[autumn cedar]

when I test inference on the same file in Applio vs. RVC WebUi, Applio heats up the machine much more. Could this be better programming in RVC Webui?

[quasi condor]

damn and hello yui

[cosmic spire]

depends on what runs faster. It could also be better programming on the applio side for using your whole PC

[autumn cedar]

ah yes, perhaps Applio would outperform RVC if I had a proper machine and GPU

[hallow thistle]

For much better performance, the desktop one is always better than a mobile laptop. Those GeForce RTX Mobiles GPUs can be a step down of desktop GPUs (like "GeForce RTX 4080 desktop" has 16 GB GDDR6X while the mobile one has 12 GB GDDR6). Gaming laptops, while capable of training a model if they have dedicated GPU, can be questionable for thermal issues.

[autumn cedar]

yes that makes sense

I'm trying to race Formula 1 with an old Volkswagen Beetle

[cosmic spire]

found this on a car resale website recently. Apparently it still runs

[autumn cedar]

that's me

[cosmic spire]

what CPU/GPU do u have?

[autumn cedar]

intel hd graphics 620, nvidia geforce 940mx (4gb vram, not shared)

[hallow thistle]

7th gen Intel Core.

[cosmic spire]

good enough for Minecraft

[autumn cedar]

heheh

[cosmic spire]

ayo my old PC (still one of my servers because it runs) has a 4th gen i5 without any dedicated gpu

4 cores without hyperthrrading go crazy

[quasi condor]

with low end shaders 🗿

[autumn cedar]

I think I destroyed my last used laptop by putting in the fridge quite a few times while it ran heavy stuff

[cosmic spire]

wasn't talking about shaders. I had a laptop once that couldn't even run Minecraft java without anything

[hallow thistle]

I'm better looking for a laptop that uses Intel Core Ultra CPU or traditional Intel Core i5.

[cosmic spire]

condensation in the fridge goes crazy, otherwise good idea

[cosmic spire]

no inten

AMD

Intel is so far behind in terms of power and efficiency

[autumn cedar]

poor old Intel, they are crumbling

[cosmic spire]

fr

top 13 most efficient CPUs are AMD lol

both in terms of fps/watt and fps/dollar

[autumn cedar]

some billionnaires should prop up Intel, competition is good, if AMD becomes the only game in town, prices will go up and quality will decline

[hallow thistle]

What about AMD Ryzen AI? This specific CPU competes Intel Core Ultra.

[cosmic spire]

also the new core ultra bullshitters are worse than 14th gen intel CPUs. they literally went backwards 😭😭😭

[hallow thistle]

When you say anything shit about Intel CPU, it won't make me to switch for AMD either. I don't need those top-tier performance or efficiency, I simply go for what is working or currently available from shops.

[cosmic spire]

haven't had it myself and haven't seen benchmarks so can't say. But imo still better since Intel is literally struggling with everything except stuff that says "more cores = better, idgaf about how good they are"

[quasi condor]

im just saying :p

[heavy jasper]

guys when i try to hear myself it lags very badly have any of u had that problem

[quasi condor]

u have a bad gpu

[autumn cedar]

we should all hope that Intel recovers

[cosmic spire]

valid. Can't say anything against that. Also Intel will probably get cheaper since they can't compete I think.

[heavy jasper]

what gpu would u need to run it

[cosmic spire]

which one do you have?

[heavy jasper]

one sec

[cosmic spire]

anything above a 1060 means ur config is probably beoken

[autumn cedar]

is training more resource-intensive than inferencing?

[cosmic spire]

always

[hallow thistle]

What is your PC GPU?

[autumn cedar]

that's what i though, just checking

[heavy jasper]

checking rn

well yo can i just dm u cuz its bunz tbh

[quasi condor]

u can just say it here -_-

[hallow thistle]

To check your PC GPU, go to Task Manager, go to Performance tab, see GPU 0 or GPU 1 and say the name any of these.

[heavy jasper]

absolutley not bro majority of the people here got amd nvidea

[heavy jasper]

k bet

[quasi condor]

lemme guess intel?

[heavy jasper]

yes bro..

[hallow thistle]

Intel UHD Graphics or Intel Arc?

[heavy jasper]

arc

[quasi condor]

o not bad

[hallow thistle]

You can't always assume anything like that.

[knotty moth]

exact model? A series or B580?

[quasi condor]

when was i assuming my guy i was guessing

[hallow thistle]

Is it Intel Arc A/B (dedicated) or simply Intel Arc Graphics (integrated)?

[heavy jasper]

uhh where do i check that?

[knotty moth]

in your GPU package or task manager

note that the arc A series may be less optimized than B580

[hallow thistle]

Check this #✨│ai-help message again, but you click on GPU 0 or GPU 1 to reveal its full name in the right panel.

[knotty moth]

if you're using the recommended tg-develop or deiteris fork instead of the original wokada, there's no need to be confused with gpu 0 or gpu 1

[knotty moth]

so you were lying 😭

https://cdn.discordapp.com/attachments/1457287271731429398/1457287503689158760/Screenshot_2026-01-04_011821.png?ex=695b743f&is=695a22bf&hm=408ab207f06445579c65034688cab2fa99178e600290a34faa0612577315984e&

[hallow thistle]

This is not Intel Arc. This is Intel HD Graphics, an integrated GPU.

Intel Arc Graphics (integrated GPU) typically found in certain Intel Core Ultra systems. 8th and generations onwards Intel Core CPU systems use Intel UHD Graphics, older gens below 7th use Intel HD Graphics, though there are some desktop CPU models that do not offer an integrated GPU. Intel Arc "aXXX or bXXX" are a dedicated one, which is a different type of GPU.

[surreal agate]

hi why are my voice so laggy and buggy

[hardy crow]

most likely your chunk size being to small

[hallow thistle]

What is your PC GPU? And did you follow any tutorial or guide before?

[hallow thistle]

You have been asking for the voice changer since last year. Did you follow any tutorial or guide before? Which W-Okada (MMVCServerSIO) version are you trying to use? And what is your PC GPU?

[viral bloom]

yo

where to download

[hallow thistle]

NVIDIA GeForce RTX 4060 Mobile, basically a dedicated laptop GPU. Historically, the terms "notebook" and "laptop" once referred to different types of computers, but today now refer to the same or identical computer type (clamshell-type portable PC with keyboard attached).

[hardy crow]

whats your gpu

[hallow thistle]

What are you looking for?

[viral bloom]

rtx 1650

[viral bloom]

the voice changer:)

[supple quarry]

how do I use the voice models for a voice changer I downloaded, it only lets me use it for musics and stuff

[hallow thistle]

For better working voice changer versions for NVIDIA GeForce RTX GPU, considering for Tg Develop's W-Okada fork (b2397) or Vonovox. If you meant by an older "AMD Radeon HD 7770", this one would unlikely suitable for any AI program.

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Tg-Develop Fork, with extra features, but it supports only Nvidia GPUs on Windows 10/11 unlike other options that have wider support. and without cloud options GUIDE

• Wokada Tg-Develop Fork

A personal fork (modified version) of Wokada Deiteris Fork, it just adds some Quality of Life improvements to it like supporting Spin Embedder and Audio Effects. Don't expect too much about it since the creator made it originally as a personal project. GUIDE

• Wokada Deiteris Fork

Deiteris' fork (modified version) of wokada that doesn't get updates anymore. GUIDE

⚔️ Wokada Tg-Develop Fork vs Vonovox

For Windows Nvidia, Both Wokada Tg-Develop fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Tg-Develop Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[frank cairn]

Ok

[hardy crow]

https://docs.aihub.gg/realtime-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-on-windows

[hallow thistle]

Try Tg Develop's W-Okada fork. https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/ https://github.com/tg-develop/voice-changer/releases/tag/b2397

[supple quarry]

no its more like, I search the voice changer I want, then I go to the link and it puts me on this RVC page where I can only use the voice changer for things like singing music and stuff but not to download it and move it to W okada. or at least idk how to do that

yea but how do I do what I want

[hallow thistle]

Deiteris W-Okada fork (b2332), released in December 2024, is slightly older than b2397 W-Okada, though this version can only be used if b2397 W-Okada doesn't work.

[hallow thistle]

RVC (retrieval-based voice conversion; like Applio RVC) and the realtime voice changer (like W-Okada) are two different programs, even if they both use RVC voice models.

[hardy crow]

someone should prolly update the ai docs to show the latest one

[hallow thistle]

A staff member who's responsible for docs on AI Hub is "Nick", though I think some admins or related mods might have access to them as well.

[hardy crow]

ah, i see

my brain auto translated the tg fork into deiteris fork

for some reason

[surreal agate]

amd something like that im not a pc pro and yeah i watch many tutorials

[hallow thistle]

To check your AMD Radeon GPU name, open Task Manager, go to Performance tab, check GPU 0 or GPU 1.

[surreal agate]

radeon rx 5700

gpu 0

[hallow thistle]

Try Tg Develop's W-Okada fork https://github.com/tg-develop/voice-changer/releases/download/b2397/voice-changer-windows-amd64-dml.zip and its guide https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/.

[surreal agate]

okay thx

[analog current]

I haven’t used Google Colab for a while, and now I’ve come back to use it. I want to create a voice model with RVC v2 disconnected, but when I run the first cell (dependencies), I get an error during the code execution at “installing faireq”.
Is anyone else experiencing this issue, and is there any solution?

[hallow thistle]

The RVC Disconnected has long ended, see #📰│dev-updates message. It's better to go for a more recent RVC fork like Applio RVC instead.

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[hallow thistle]

Running the batch file directly inside Applio RVC folder would be better than running within Dione Launcher environment. The native batch file would show error logs within terminal at one place, while Dione might not. The latest version for Applio RVC is 3.6.0.

[surreal agate]

where do i get voice models

[hallow thistle]

#1175430844685484042

[surreal agate]

okay

[young furnace]

which vovonox should i download? i have rtx 4070

1 7 beta?

[analog current]

Can you send me it's colab notebook for training a model plz?

[hallow thistle]

https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb

For no-UI one, there's this notebook. https://colab.research.google.com/github/IAHispano/Applio/blob/main/assets/Applio_NoUI.ipynb

[analog current]

It seems that not all resources are being used properly. I remember that when I used RVC disconnected, almost all resources were utilized fully, although it took a much longer time to train the model.

[hallow thistle]

If the Applio uses only little of VRAM, try change "batch size" up.

I don't train voice model, but this observation usually indicates that the batch size is set too low, which is why it uses very low VRAM (3.3 of 15.0 GB). Train and inference speed also depend on GPU clock speed too, not just the batch size of the program or GPU VRAM size.

[ruby forge]

Hm, how do I build Hybrid production-grade architecture with constrained scope based on current AI knowledge?
Something like: improved transformer, learning loop, memory systems, limitations / rules / reasoning, interaction layer, etc.

Any good guides / papers / frameworks / tips?

[reef berry]

Hello! sorry for the late reply.
My GPU is RTX 3060 Laptop.

I was wondering what is it that he is using, like if there is any new application, because if i recall correctly, AI Voice changers cannot do cries/coughs realistically..?

And, i am sorry if i was being unclear about anything!🙏

[latent knoll]

im using applio

[versed hull]

im on mac

onnxcpu nocuda

is the one im using

[slender grotto]

how do i train my corpse husband model do i just keep talking with it on or what

[hallow thistle]

Does your Apple Mac uses Intel CPU or Apple Silicon (M2, M3)?

[hallow thistle]

There's a newer W-Okada fork made by Tg Develop. https://github.com/tg-develop/voice-changer/releases/tag/b2397 https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/

Like RVC voice models, the voice model can do some crying and coughing noises, if the model was trained with such dataset audio, but that's about it.

[versed hull]

m2

[void elbow]

french voice or french serv

[dapper socket]

after extraktin the model i only gives me the pth and not pth and index

[unreal wing]

it wont let me talk

and if i am i cant hear myself

idk what i did wrong

[swift creek]

can some1 help me

[limpid trout]

Uh, it failed

What to do

[swift creek]

some1 pls help when i open start http batch file it dosnt launch it opens and closes in 0.1s pls helpp

[south tree]

yo can someone help me with the voice changer when im speaking it talks like 2 or 3 times

[south tree]

@tame oracle

[rose root]

how to make own voice models?

[narrow sail]

How do i get and use vonovox?

[quartz moon]

hi
is there a good lightweight voice changer for linux?

i use rx 570 and i3-8100(yes ik very bad lol)

[latent knoll]

i tried reinstalling applio,
Traceback (most recent call last):
File "C:\apps\Applio\applio\app.py", line 6, in <module>
import gradio as gr
File "C:\apps\Applio\applio\env\Lib\site-packages\gradio_init_.py", line 3, in <module>
import gradio.simple_templates
File "C:\apps\Applio\applio\env\Lib\site-packages\gradio_simple_templates_init.py", line 1, in <module>
from .simpledropdown import SimpleDropdown
File "C:\apps\Applio\applio\env\Lib\site-packages\gradio_simple_templates\simpledropdown.py", line 7, in <module>
from gradio.components.base import Component, FormComponent
File "C:\apps\Applio\applio\env\Lib\site-packages\gradio\components_init_.py", line 1, in <module>
from gradio.components.annotated_image import AnnotatedImage
File "C:\apps\Applio\applio\env\Lib\site-packages\gradio\components\annotated_image.py", line 15, in <module>
from gradio.components.base import Component
File "C:\apps\Applio\applio\env\Lib\site-packages\gradio\components\base.py", line 21, in <module>
from gradio.blocks import Block, BlockContext
ModuleNotFoundError: No module named 'gradio.blocks'

help?

[frigid sapphire]

I'm new to this myself, but I think you'd need to set your Output Device and Monitor Output Device to None.

[gloomy dew]

-uvr

[patent trellis]

UVR: Ultimate Vocal Remover

One of the best free and open source vocal and instrumental isolation tool.

Website

https://ultimatevocalremover.com/

GitHub

https://github.com/Anjok07/ultimatevocalremovergui

Guide

https://docs.aihub.gg/rvc/resources/dataset-isolation/#vocal-isolation--cleaning

[gloomy dew]

is UVR still used? like, is it still the best for vocal isolation locally?

last thing I've heard from it, was that is no longer being updated

[south tree]

when i put output to none it doesnt talk

can someone help me when i talk there is like a echo or sum and its speaking the sentence i said

[quartz moon]

does the kaggle w/odaka thing work on linux?

if yes then how cuz im very dumb

[south tree]

can someone help me when i talk there is like a echo or sum and its speaking the sentence i said

[south tree]

@craggy bough @silk stirrup @viscid moss please help

[slender grotto]

i wanna train this corpse husband model but i dont know where to start does anyone got any realistic ones

[simple ore]

you're trying to use v1 model, there's a small bug with that

[gusty moth]

-colabs

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[south tree]

can you help me when i talk there is like a echo or sum and its speaking the sentence i said

[craggy bough]

dont ping mods for stuff like this

[south tree]

then maybe here should be more helpers or stuff to help people

[south tree]

can u help me or not

[craggy bough]

no.

not with that attitude

[south tree]

wow what a fucking beatiful staff that doesnt even help lol

[craggy bough]

we are all volunteers, be rude and you wont get any help

[south tree]

for what are u even here

[craggy bough]

to moderate the server

[south tree]

ur not even looking here u just came rn

[craggy bough]

yeah cuz you pinged the entire staff team???

[south tree]

yes

its a dead server lol

keep deleting my messages

😂 good staff that doesnt even look in the server and doesnt help

and comes because i pinged him and its his biggest worry

[stiff python]

in vonovox, what is "import" and "export" effects?

[hardy crow]

why did you make an alt account just to fake ask for help then argue with staff

[south tree]

wdym alt lol

its my normal acc dude

[south tree]

i asked for help but like i can see the staff is bad

[quasi condor]

import is u upload a file and export is u download or save a file

[stiff python]

well yea but what type of file

like what effects

[quasi condor]

i dunno i dont use vonovox ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯

[quasi condor]

@viral mason

[brazen spear]

Can i ask what is the best ai for make a cover song?

[viral mason]

no clue I use fl studio for voice effects

[south tree]

ur lit a mod so u know how the changer works so u can help others to fix it or not? a mod does help too?????????

[viral mason]

how are people like that even real lmao

[calm badge]

could anyone help me? why does my mic cut out like crazy? i have a rec of it but cant send in this channel

[finite prism]

did u find fix mine is doing the same

must be settings

[hardy crow]

no idea either, im assuming its an alt from his recent creation and join date

eh who knows

but how can you tell if another account matches his🤔

using linguistic patterns?

i would never reply to my main acc if i was on my alt

then again not everyone is me so

[latent knoll]

how do i fix

[simple ore]

if you want to use v1 models, make this change

use notepad

[hardy crow]

[autumn cedar]

I have run another inferencing test with RVC WebUi, this time a 4.5 minute wave file, on my weak little old windows 10 laptop with 4GB VRAM GPU, after making a suggested registry tweak (increase Timeout Detection and Recovery to 60 seconds). Again success!

But now I would like to experiment with some Python code bits to add to the RVC scripts which are supposed to cause the cpu/gpu to run a bit cooler -- does anyone have any experience in this regard?

[autumn cedar]

omg this looks complicated -- if anyone is interested, see this: https://www.perplexity.ai/search/i-found-some-python-suggestion-LZEbgL_GS3Sr5ajoYytZcw#1

I wonder if I should try to communicate with the people who actually wrote the RVC WebUi python scripts?

[odd flower]

Hello, I am training a RVC model using a 15 minutes worth of audio dataset and the dataset is clean as it is generated using my cloned voice (text to speech), now I am creating a speech to speech model but the results are so bad, I am using the settings mentioned on the official documentation with noraml pretrained weights for a epoch of 4 then when I try to train using epoch 8 the results are worst as compare to batch size 4 (I though it was better).

Any suggestions?

[ivory trout]

what settings do i change for it to stop being a little glitchy

[frank pivot]

someone can help me to improve a RVC Model? like idk how make it sound better / less "lag"?

[slender grotto]

any super realistic deep voice changer's every single one i use sound's very bad

[rose drift]

Which models are used to create the images with the public figures? OpenAI, Gemini etc don't allow it but I see lots of pics edited / modified involving celebrities and politicians.

[thorn kraken]

Im using intel as my gpu and cpu so what do i download

[hallow thistle]

AMD Radeon(TM), a likely integrated GPU for "AMD Ryzen AI", is still not recommended for running AI program like W-Okada voice changer anyway. While an integrated NPU does exist within AMD Ryzen AI system, it's not optimized for running more-intensive AI and neither voice changer compiled for it. The best bet is to looking for a dedicated GPU like AMD Radeon RX, stay for CPU-only which might be faster than a traditional "AMD Ryzen" or go for online cloud instead.

[hallow thistle]

Does your PC has Intel UHD Graphics, Intel Arc Graphics, Intel Arc Axxx/Bxxx or something?

[hallow thistle]

See #1175430844685484042.

[thorn kraken]

How do i check

[hallow thistle]

To check your PC GPU, open Task Manager, go to Performance tab, see GPU 0 or GPU 1.

[thorn kraken]

It say's Intel(R) Arc (TM) A770 graphics

[hallow thistle]

This Intel Arc is a dedicated GPU, so W-Okada voice changer DirectML should work with that one. Try Tg Develop's W-Okada fork. https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/ https://github.com/tg-develop/voice-changer/releases/download/b2397/voice-changer-windows-amd64-dml.zip

[thorn kraken]

I also have voicemod installed will that cause a problem

[hallow thistle]

You can still have that one installed, and while it's better to not use that as your main voice changer, Voicemod can still be useful if you wanna use it for specific audio effects.

[thorn kraken]

do i need to download the VAC (virtual audio cable)

[hallow thistle]

There. https://software.muzychenko.net/freeware/vac470lite.zip

[hallow thistle]

What is your PC GPU? And are you following any tutorial or guide?

[thorn kraken]

Theres one on the site should i downnload that instead

[hallow thistle]

If you mean on official Muzychenko website, look for "lite" one. If you mean by VB-Cable, it's a complete different program.

[thorn kraken]

im talking about the site you said to go download the voicechanger

[hallow thistle]

What site? There are three.

[thorn kraken]

i found the lite version its made by muzychenko

[thorn kraken]

this site

[hallow thistle]

VAC Lite on AI Hub doc, it's the exact same one as https://software.muzychenko.net/freeware/vac470lite.zip.

[thorn kraken]

so it doesnt matter which one i download, and another question do i NEED to download them

[hallow thistle]

If I say "sure", it means you should download the program, so sure. No need to confirm another time.

[hallow thistle]

Do you really think "all" moderators and admins here are supposed to know about W-Okada voice changer? Those staff members with "Helper" role sure know about voice changer (and also RVC) as basics, and one of those Helpers is me.

[thorn kraken]

i downloaded it what now

[hallow thistle]

Use 7-Zip or WinRAR to extract "vac470lite.zip" into a directory. After extract, go inside wac470lite folder, spot "setup64.exe" and double it.

[thorn kraken]

Wait. How do i extract it into a "directory"

[hallow thistle]

Make sure to read guide before you ask anything.

[thorn kraken]

I put it in WinRAR but im not too sure how to double it

[hallow thistle]

A member here once voiced his opinion where "Helper members" aren't supposed to tell overly many steps to other members, but because most of the time some members don't even read the guide doc as reference, and when I have to tell so it feels more like a "forced" strategy. I kind of agreed with that statement, but even if there's such opinion it won't fix anything.

[thorn kraken]

the guide isnt helping me unless im not looking hard enough

[limpid trout]

Alright,
I need a Voice changer besides Vonovox

Mainly Voice changer that has A way to train any voice in

Besides with TTs, I want Just upload audio and change into different voice

Real time voice changer doesn't seem necessary for me,
I need voice actors

[hallow thistle]

W-Okada voice changer doesn't train a voice model. Applio RVC can. Your queries look conflicting, but if you're looking for a local voice actor, you can try hire or commission them.

To request for a voice model, there's #1159289738314919936, and if you want to train one from a user's thread in #1159289738314919936 you might have to get "model maker" role first.