✨│ai-help

[graceful obsidian]

because it doesn't work like that.
If your voice is too different from the target ( model ) you gotta put in something from yourself to make it work

• changing transpose ( pitch ) of the model is one thing.

another one is:

• speaking in similar tone or manner

• ( alternatively: speaking in similar pitch to model's native / learned range
  (( if you can, that is )) but If that's the case, transpose won't be as relevant )

[tame mural]

So what you're saying is, to achieve a similar clarity, one should try to imitate the sample's tone and speech pattern?

[graceful obsidian]

no, ignore the samples

those usually don't reflect model's native range or potential

you just simply have to match your pitch, tone with the model or at least tweak the pitch / transpose to fit your use

a lil bit of impersonation works too

[graceful obsidian]

as for clarity, if your mic is truly tragic ( which I don't think is the case ) the model should do just fine ( unless the model is poorly done, then you can't do much with it

[tame mural]

Those are very useful pointers, I truly appreciate and will keep them in mind.

I'm still on the lookout for a good, free model, which I thought (Shylily) was, but according to my own recordings, in accordance to your notes, is still not good enough.

[graceful obsidian]

you can link the model

I'll review it's fidelity and point out if it's the model or something else

if you want

[tame mural]

It's here
#1212177800455266446 message

[graceful obsidian]

alr, gimme few mins

gotta finally leave the bed lol

[tame mural]

Haha! take your absolute time, please.

[graceful obsidian]

@tame mural Alright, testing done

First, familiarize with the audio I used on input;
' Quality tester ' is SynthV AI clean input I typically use to diagnose models, it's a female voice
' Codename vocals ' is, well, my own vocals. Use it to test models' generalization ( and gender-switch ) capabilities.

and here are the outputs of the lily model ( inference using rvc, not rvc's native realtime or w-okada, but there's not much difference as real-time voice changing is nothing else than just real-time / live inference

Pretty much, the model is just not of the greatest quality if I had to be honest with you

and she definitely struggles with lower ish male vocals.
aka; needs pitch adjustement in voice changers + a lil bit of impersonation wouldn't hurt ( because you ideally shouldn't use the faiss index (( feature index )) )

[fathom crater]

hey

where does rvc save the finished trained thing

yo

so it is done training

where did it save

[graceful obsidian]

And for the record, no, not all female models or even high pitch female models struggle with low or low male voices. A good model should be able to work out through quite a nice range. Here's my own Kurisu model in testing ( she's a perfect example of a truly HQ model with good generalization

[graceful obsidian]

All training / model's training files end up in:

rvc's folder / logs / ur_model / here

including tensorboard file, config, logs, generator, discriminator, features and index
.

models themselves ( those you use ) are in:

rvc's folder / assets / weights / here

[fathom crater]

Thx

[fathom crater]

wait

how do i load it in into rvc gui

i put the files in a zip file

and the model wont show up

it just says extracting files

[graceful obsidian]

well, I can't help with notebooks / huggingface spaces or such as I don't maintain nor use them
you should ask others really - they'll know more on that

@fathom crater

[fathom crater]

i meant the rvc gui on pc

[graceful obsidian]

but like, gui? you mean web ui?

[fathom crater]

no

[graceful obsidian]

Dunno that fork then

[proud elbow]

rvc-gui is outdated, not even have rmvpe which is the best currently

[graceful obsidian]

oh huh

[graceful obsidian]

welp

that seems legit old lmao

[fathom crater]

how do i do it on that rvc web thing

[graceful obsidian]

I'm currently working on my fork's updates so, it's wip and can't recommend it but
perhaps this will do for you:
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI
( original rvc )
or applio:
https://github.com/IAHispano/Applio/tree/main
( custom rvc (( fork )) and some other things, utils n so on )

[fathom crater]

I already have the one with nvidia

[graceful obsidian]

then why bother with gui variant then

also, 1006 release is outdated

[fathom crater]

Because i know how to use it

[graceful obsidian]

their repo on the other hand, is up to date

I mean, there's nothing difficult about using rvc
you select the model, index, audio and done

that's all there is to it ^

• f0 method choice
• index slider ( and picking it )
• picking the model
• picking transpose
• audio

[tame mural]

I apologize for the late reply as I just saw your messages right now.
So from my understanding, based on your extensive testing, the voice (Shylily) isn't quite compatible with my own natural voice, due to poor implementation and/or weak range of handling various tunes.
Correct?

[rare goblet]

Ayo? @tame mural level 3 !!!

[graceful obsidian]

nono, it's just that she's rather on the lower quality side, meaning it's less ( the model ) forgiving for voices far from her own ( in style, pitch etc

so you gotta play with pitch ( transpose ) to match your voice

and maybe even impersonate / modulate your voice a little

as for fidelity / clarity of the voice, yeah.. that's just what it is. Matter of the model

[tame mural]

Got it! That makes this much more understandable.
Is the voice you posted (Kurisu) available to test somewhere? and if not, may I ask for a recommendation for a freely available voice that has a wider range?

[graceful obsidian]

Unfortunately, no, she's ( along with my other models ) a private model

[tame mural]

It's perfectly fine if the answer to both questions is (No). As you've helped quite a bit already. which is totally appreciated.

[graceful obsidian]

As for wider-range models.. yeah that's difficult becaus the way weights.gg is doing the models' demos is a bit wonky

there's no freedom to pick own demo samples or such

always default ones for all models so, it's kinda, yea, difficult to tell good models from bad apart ( and tell if it's a matter of demo sample or not

but I'd say, try to search for some RVC covers on yt and find a model that does well in high / low pitch

[tame mural]

Then that's my next step, alongside everything else you pointed out to me. I'll fiddle around some more, and hopefully find something that works good enough for my goal.

[graceful obsidian]

btw, you aim for a female voice model?

[tame mural]

A believable one, Yes.

[graceful obsidian]

it's pretty difficult to get it to believeable levels if i had to be honest with you

it's still AI and those more into it or with sensitive ears, can spot it

model would have to be perfectly trained + you'd possibly have to use some quality degradation effects on the "mic's" ( model's voice ) output

[tame mural]

This is a sample I posted earlier, and something on my end 'eats' syllables in every voice I tried, unfortunately.

[graceful obsidian]

can you show me your settings for the voice changer?

also, how is your mic's threshold ( windows or per headset? sensitivity

[tame mural]

That's my RVC setup, based on the guide from this server. If there are any other settings I need to adjust, please let me know.

[tame mural]

I'm sorry, how to check that value?

[graceful obsidian]

oh, that

at first I thought you use w-okada

lemme see

btw, what's your gpu

[tame mural]

To be perfectly honest, I tried both, and they both result in the same choppy, unclear voices.
My gpu is Nvidia 3060 12gb.

[graceful obsidian]

Try to change the sample length to around 0.5 or 1.5

[proud elbow]

have you also tried on okada?

[graceful obsidian]

also, extra infer time I'd keep at 1 or 2.5

if that won't help, it means there's something up with your mic

it might be it incorporates some noise-gating

or some issues with sensitivity exist

Alternatively, if changing the settings doesn't help, try 'dml' version
and again, if that doesn't help, it must be your mic or some noise-gating

[tame mural]

I'll post a voice test momentarily with those settings.

[tame mural]

I did, and while it's a little bit clearer than RVC, the end result still sounds non-believable enough for me.
I can post a sample from that too if it's ok.

[graceful obsidian]

Generally, I'd say rvc's native voice changer has more potential for stable and smoother voice changing

it doesn't really require onnx model conversion for performance boost ( which degrades models' accuracy due to pytorch and onnx differences )

and has more tweaking options but ye, it's a matter of tweaking the settings, if it's smoothness you care about
If it's about " oh, this must be a legit person and not an AI! " that's almost impossible to do nowadays as people already know rvc, w-okada, so-vits and all of that

as mentioned, effects or degradation filters would be required ( to make the voice sound as if coming from cheap laptop mic or cheap headphones

also a clean as hell model's a must

[tame mural]

Actually, this sounds a bit more clearer than before, thanks to the sliders that you suggested. The voice cut out when I was asking (so what do you think?)

[tame mural]

I think the more I test and listen, the more that this possibility appear to be my main issue.

[graceful obsidian]

basically:

• the sample length is your voice's stream ( your voice audio data going in through the mic ) divided into chunks fed to the changer which are then inferenced on the model

• fade is the fade ( cross-fade ) inbetween those inferenced chunks ( best kept at .10 or .15 as you have )

• extra infer is a lil " overhead " buffering ( 0.5 to 2.5 is good but increases delay so, I'd say 0.5 to 1 is good )

as for the other things I noticed in your case, the characteristics of how your model's output sounds, it's the thing with mic definitely

you should play around with your mic or win settings to increase the sensitivity / turn off any noise-gating or cancelling if you have such

[tame mural]

That actually makes a lot more sense now, as it's almost impossible that every voice that people had almost zero issues with, doesn't work as clear and believable enough in my case.

[graceful obsidian]

also, given shylily is an eng speaking vtuber, perhaps you can use a bit of index to hide your accent ( if that's your goal too )

[tame mural]

I just delayed that possibility out of thinking that my mic was - maybe - good enough.

[graceful obsidian]

it could be, indeed, but some mics have hardware noise-gating

or noise-suppression

[tame mural]

Would you kindly suggest a value for this slider?

[graceful obsidian]

which can't be turned off ( some can, depends on the device )

[graceful obsidian]

perhaps you can try 0.3 or 0.5

[tame mural]

I'll screenshot all the info you posted, for future refernce.

[graceful obsidian]

it's gonna take shylily's own features such as accent and pronunciation

and mix it with yours in 30% or 50% ( kinda wonky way of explaining it but, it works i guess

[tame mural]

Noted. Posting a sample momentarily.

[graceful obsidian]

got it

[fleet sleet]

Do you know how to remove accent?

[rare goblet]

Ayo? @fleet sleet level 5 !!!

[wispy lodge]

Onnx conversion does not reduce the model quality. It's the code that does not work correctly with onnx inference in w-okada voice changer

[graceful obsidian]

It does degrade the accuracy of the model on technical level

but it's typically not audible to you

current way of exporting pytorch to onnx for rvc uses static tracing

if you were to use dynamo exporting, heck, that'd work

The main advantage of this approach is that the FX graph is captured using bytecode analysis that preserves the dynamic nature of the model instead of using traditional static tracing techniques.

and rvc's models are indeed very dynamic in activations and all of that

[graceful obsidian]

as in, masking your own?
or removing model's accent?

[fleet sleet]

Model is in English and when I speak in Portuguese it has a accent

[graceful obsidian]

current pytorch to onnx could be kinda compared to models' pruning

[wispy lodge]

I mean, that's debatable because "audible" is not a valid criterion

[graceful obsidian]

but you still get what I mean

[wispy lodge]

It depends which model you convert and which infer_pack you used

[fleet sleet]

American accent

[graceful obsidian]

tho, avg users shouldn't really worry about it but, my approach includes letting people know, so

[graceful obsidian]

thennn, you'd want to rely on model's index

but results can very really, worth playing around it.
I'd say 0.5 for index is where it starts really kicking in

[fleet sleet]

Already did that and didn't work it

[graceful obsidian]

can you post demo of your output at 0.0 index and 1.0 index?

might be model's issue

[tame mural]

Settings and sample.
Also, I can't help but feel it's getting closer and closer to my goal, minus a few still present caveats, especially when I uncontrollably produce the "ahem" sound.

[fleet sleet]

Unfortunately no

[graceful obsidian]

Then I can't really help if I don't hear it

[graceful obsidian]

Already sounds much better!
input and output noise reduction will cause more harm than good sadly, can have some chopping here and there so I'd advise having it off and playing with response threshold ( lower it is, more it picks up from your mic )

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[graceful obsidian]

also, loudness factor is something you can play with ( it's kinda rms scaling (( think of it as rms normalization

it'll even out the dynamic range ( kind of )

[tame mural]

It does, thanks to you. And based on our fruitful convo, I think I have enough info to fiddle around with.

[rare goblet]

Ayo? @tame mural level 4 !!!

[tame mural]

I'll try to achieve my goal with a few more tinkering. I can't thank you enough for all the notes that you provided.

[graceful obsidian]

Don't mention it - hope it works well for you

[pastel oak]

about the reduction, i like output reduction, always had good results with it. input has been rather terrible

[hot lagoon]

I’m not sure how converting back vocals even works because every time I try the pitch sounds so off

[graceful obsidian]

ohhh, it's been accurate for you?

for me it sometimes mistakes my models' ( which tend to be on the softer tone end ) sounds or end-breaths n such for 'noise'

might be my specific-use case then

[pastel oak]

its weird, i have noisy output on rvc without output reduction ; on wokada for those same models i dont have that

maybe specific use case indeed

[graceful obsidian]

yuh, could be

I could have a look into how w-okada handles it maybe

and potentially port that approach, dunno yet

currently busy with my fork so

[pastel oak]

extra inference time is still confusing me a lil bit

when i set it to a lower value i dont really get the results i want, while at the higher range it sounds exactly like i want it

is there some sort of use case aswell or how does the overhead buffering apply

[graceful obsidian]

cause you gotta compensate the pitch for the model

[graceful obsidian]

if the model is rather low pitch ( naturally ) and input is high pitch, you gotta lower the transpose
say, by -12 or -6 ( but tweaking it to non 12 or -12 results in half octave move, meaning you then gotta be adapting the instrumental and so on

[wispy lodge]

It doesn't. Output is not post-processed (except it could be padded) in w-okada's code

[graceful obsidian]

I believe it's just delaying the whole conversion by whatever you set ( can't tell for sure what metric they use ) and inferences it in memory, sparing some of weaker gpus that can't maintain a specified chunk size, so, sample length
(( from logical point of view

[graceful obsidian]

oh, then that's that, rip ✊

[pastel oak]

and is sample length at a value like 1.5 not increasing the delay? you explained what it does in your convo just now but seems odd at that high value

i dont wanna put misinformation on my realtime guide

[graceful obsidian]

maybe I can draw it, hold on

[graceful obsidian]

@pastel oak this is how I see it personally:

I suppose, extra infer should be scaled up with your gpu's / device's performance.
weaker it is, worse " keep up " pace it'll have and could stutter or whatever

Overal, that's what I've gotten from testing and it seems pretty logical to me but, if I am wrong, anyone can freely correct me ( would be appreciated ofc )

also a note, reason why sample length in shorter values can and usually cut when you use your model for pitch-modulated-heavy content is most likely because it's easier to hide from one's ears and tame longer samples than super short ones that are patched with simple cross-fades

and given speech typically isn't as pitch-modulation dependent ( unless you're a tsundere lol / jk ofc ) then ye, it'll work pretty fine with shorter ish sample lengths

But ye, no idea how much of it applicates to w-okada as there seems to be 1 or 2 tweaking options less

[pastel oak]

alright so this is roughly what i thought aswell, aside from the singing - i did not consider that at all

[graceful obsidian]

👁️ 👄 👁️

glad to hear

[pastel oak]

from my experienes, everyone was able to max out extra and have great sounding quality cmopared to a lower extra value. so i suppose for faster results you can use lower extra

like explained for dumbies

[fleet sleet]

Do you know how to contact RVC-boss, yxlllc?

[pastel oak]

without being technically accurate

[graceful obsidian]

extra is just like a video's buffering

ye

lower it is, lower the latency

[pastel oak]

thats a good analogy

[fleet sleet]

Anyone in here knows who created RVC AI???

[graceful obsidian]

but can be problematic for weaker hardware

[graceful obsidian]

Rvc's team

[fleet sleet]

Do you know?

[graceful obsidian]

fumiama, ftps, rvc-boss

and few others

• tons and tons of contributors ofc, they also added stuff from themselves

[fleet sleet]

How to contact with them?

[graceful obsidian]

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/graphs/contributors

contact the devs? you'd want to visit their devs ( rvc devs ) server

[slate compass]

Low response threshold picks up literally everything, higher response threshold starts causing stuttering and stammering (only -40)

[pastel oak]

yep

rvc sucks at noise gating unfort

[wispy lodge]

Extra is basically a history (or context), not a buffer. It does not add up to the output latency, but just increases computational complexity since you have to process the current chunk + past N seconds

[pastel oak]

so u need steelseries sonar, voicemeeter or anything else that can control it

[graceful obsidian]

oh, that's new then

thanks for clarification!

[wispy lodge]

The way I see it, the longer the extra + your chunk, the longer your inferred sample, hence the output is more coherent

[graceful obsidian]

well, that doesn't change for sure

but I think using the " buffer " analogy is just easier to grasp in general so

but ye, thanks for technical insight

[graceful obsidian]

@pastel oak that one seems to be more accurate

would use that one instead

[pastel oak]

emojikage the madman

so knowledgeful

[graceful obsidian]

other than that, sometimes I hope they'd all devs more properly name their stuff

normally you'd think extra is an extra overhead / buffer

lel

[slate compass]

GOAT TY

[rare goblet]

Ayo? @slate compass level 6 !!!

[slate compass]

AYOOOOOOO

[wispy lodge]

You don't want to see the code for inference then lmao. I mean, guys did great job making it work, but design decisions and the way it's written sometimes are puzzling at best

[waxen spindle]

DEBUG:matplotlib.pyplot:Loaded backend agg version v2.2.
DEBUG:urllib3.connectionpool:Starting new HTTPS connection (1): huggingface.co:443
DEBUG:urllib3.connectionpool:https://huggingface.co:443 "GET /TexX/GPT-SoVITS-Models/resolve/main/Mash_Burnedead.zip?download=true HTTP/1.1" 302 1214
DEBUG:urllib3.connectionpool:Starting new HTTPS connection (1): cdn-lfs-us-1.huggingface.co:443
DEBUG:urllib3.connectionpool:https://cdn-lfs-us-1.huggingface.co:443 "GET
/repos/c3/fc/c3fc12bf1e1a7f93dc8374d26338badae7e7d3549f171b2ce15f964d5b38f2e6/22d0b0731322a83988661a4d9d3dbe0ec32b88504a563f1289d329dd6d2504dd?response-content-disposition=attachment%3B+filename*%3DUTF-8%27%27Mash_Burnedead.zip%3B+filename%3D%22Mash_Burnedead.zip%22%3B&response-content-type=application%2Fzip&Expires=1715082643&Policy=eyJTdGF0ZW1lbnQiOlt7IkNvbmRpdGlvbiI6eyJEYXRlTGVzc1RoYW4iOnsiQVdTOkVwb2NoVGltZSI6MTcxNTA4MjY0M319LCJSZXNvdXJjZSI6Imh0dHBzOi8vY2RuLWxmcy11cy0xLmh1Z2dpbmdmYWNlLmNvL3JlcG9zL2MzL2ZjL2MzZmMxMmJmMWUxYTdmOTNkYzgzNzRkMjYzMzhiYWRhZTdlN2QzNTQ5ZjE3MWIyY2UxNWY5NjRkNWIzOGYyZTYvMjJkMGIwNzMxMzIyYTgzOTg4NjYxYTRkOWQzZGJlMGVjMzJiODg1MDRhNTYzZjEyODlkMzI5ZGQ2ZDI1MDRkZD9yZXNwb25zZS1jb250ZW50LWRpc3Bvc2l0aW9uPSomcmVzcG9uc2UtY29udGVudC10eXBlPSoifV19&Signature=eWK9iqhMXWW0Squ1pr-cp9oHGNyvk3oiuJgDpBbvh3Tb4dbqMn44BAnjry8cO4~BJEn7FMyPZN9kQwbESivYRQt1l6vnQo6RTcfbt9HOs-yC5F5QygWsx8RGFRpUd32kmgvkLY-RZYWriDhAUPMM~aayrrRGMXBIrswGZVAnCAdqb3uczyCc8dGgZo1RC2Pn8YPHPECjnXe~3n2UV4PgyVF7CHGs4ZcxX~BKLXigrlI2CyWOZYr~cQ6ELHBdvO1-UFsiYbxCbivWFGuWu-IPNvFhQGYiEaa9xwzLK7Y5ATaO-V93CfpZeVIWPaNSw1~VgXnw428UU4nlpvfKGKp6Jg__&Key-Pair-Id=KCD77M1F0VK2B HTTP/1.1" 200 223339203
100% [..................................................] 223339203/223339203  

Proceeding with the extraction...
DEBUG:matplotlib.pyplot:Loaded backend module://ipykernel.pylab.backend_inline version unknown.
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/gradio/queueing.py", line 495, in call_prediction
    output = await route_utils.call_process_api(
  File "/usr/local/lib/python3.10/dist-packages/gradio/route_utils.py", line 232, in call_process_api
    output = await app.get_blocks().process_api(
  File "/usr/local/lib/python3.10/dist-packages/gradio/blocks.py", line 1561, in process_api
    result = await self.call_function(
  File "/usr/local/lib/python3.10/dist-packages/gradio/blocks.py", line 1179, in call_function
    prediction = await anyio.to_thread.run_sync(
  File "/usr/local/lib/python3.10/dist-packages/anyio/to_thread.py", line 33, in run_sync
    return await get_asynclib().run_sync_in_worker_thread(
  File "/usr/local/lib/python3.10/dist-packages/anyio/_backends/_asyncio.py", line 877, in run_sync_in_worker_thread
    return await future
  File "/usr/local/lib/python3.10/dist-packages/anyio/_backends/_asyncio.py", line 807, in run
    result = context.run(func, *args)
  File "/usr/local/lib/python3.10/dist-packages/gradio/utils.py", line 678, in wrapper
    response = f(*args, **kwargs)
  File "/content/program_ml/core.py", line 392, in run_download_script
    model_download_pipeline(model_link)
  File "/content/program_ml/rvc/lib/tools/model_download.py", line 323, in model_download_pipeline
    file_name = item.split("_nprobe_1_")[1].split("_v1")[0]
IndexError: list index out of range

[rare goblet]

Ayo? @waxen spindle level 1 !!!

[waxen spindle]

Model link that I was using #1220290728941457408

[odd shale]

nice to see you getting along with other staffers.

Keep up Cody.

[waxen spindle]

I can't seem to be able to find a working link to be able to mak ai covers

[proper shale]

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

should work

[calm jackal]

-rvc

[azure marsh]

Documentation

📚 All-In-One English documentation

https://docs.aihub.wtf/

❔ English FAQ

https://docs.aihub.wtf/faq

Credits: AI HUB FRANCE

✨ More guides

Full AI Voice Model Training Guide (Local)

Link: YouTube
Credits: Christopher Villanueva

Model training with Mainline RVC

Link: Rentry
credits: Raven (ravencutie21)

AICoverGen Colab Guide

Link: Google Docs
Credits: Eddy (Spanish Helper)

Create a model with RVC disconnected (colab)

Link: Google Docs
Credits: Angetyde

How To Make an AI Cover With Ilaria RVC

Link: Rentry
Credits: 👽 Julia (ailen2091)

[calm jackal]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[cursive scaffold]

hey for the real time RVC does it really require a RTX card or would a GTX 1080 do the job?

[proper shale]

eh, probably

i wouldn't game with it running tho

[cursive scaffold]

as in probably could get away with a 1080?

gotcha ill try it out

[rare goblet]

Ayo? @cursive scaffold level 1 !!!

[proper shale]

it will probably work

dunno how fast

[cursive scaffold]

it should be faster than a 3060 at least

[proper shale]

not AI-wise

[cursive scaffold]

was more concerned if RTX was needed because of like driver stuff

[proper shale]

its more like a recommended setting

[proud elbow]

RTX cards are more optimized on AI because of tensor cores

but yeah GTX ones still work

[brittle wing]

do i need an index file aswell for it to not echo?

[rare goblet]

Ayo? @brittle wing level 1 !!!

[proper shale]

index is for accent, so no

if you're having echo on W-Okada just enable sup2 and echo

[brittle wing]

when i use one of the models from here it just echoes over and over

[brittle wing]

alr it worked thanks

[proper shale]

yw :)

[cursive scaffold]

i cant seem to run the realtime GUI

i get stuck on loading

any idea how to fix it

okay nvm i got it to work now

[proper shale]

ah

[brittle wing]

whats the most realistic sounding female model?

what r ur recs

[proper shale]

try this one idk

#1236217350345330719 message

also, search and listen to the samples provided

[cursive scaffold]

why does my gui look wrong

[proper shale]

you can always find good models ere

[proper shale]

ayo wait

that's the

built in voice changer

i thought you were gonna use W-Okada

[cursive scaffold]

i thought its supposed to look like

whats the dif?

[proper shale]

oh this seems old

[proper shale]

W-Okada is easier to use tbf

[pastel oak]

are you running the mangio voicd chanher

[cursive scaffold]

i thought okada was abandon ware

[proper shale]

yeah seems like it

[proper shale]

it wasnt abandoned

only the guide for it technically

[cursive scaffold]

oh

i guess i got the wrong one

[finite cobalt]

this is happening every time i try to download a pth to the rvc v2 colab, help pls?

[proud elbow]

JSONDecodeError means the colab is outdated

[proper shale]

old colab

[finite cobalt]

where's the new colab

[proper shale]

use this, see if it works: https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb

there's other colabs and guides at https://docs.aihub.wtf/

[finite cobalt]

thx

[flint elm]

hi, I'm new, I don't really understand how AI cover songs work, so what I need here is how to cover an English song into a Japanese song using AI?

[proper shale]

you'd have to get somebody to sing japanese lyrics and then convert with RVC, there isn't translation AI for music yet

ooor synthV

@trail iris

[flint elm]

i see i see, thank you

[wispy burrow]

And if you get that error locally?

[quaint beacon]

Hey. I've been using the models from this server, but when I input them and try to use them, I get the "pipeline create failed. check your model is valid" any clue?

[proper shale]

make sure it isnt a GPT-SoVITS model

[quaint beacon]

its not

[proper shale]

wait what are you using in the first place

W-Okada?

[fleet sleet]

The way that "not MJ" said is right but the song will have accent

[quaint beacon]

Thats the one for AMD or Nvid?

[rare goblet]

Ayo? @quaint beacon level 1 !!!

[proper shale]

yeah this one

https://rentry.co/VoiceChangerGuide

[proud elbow]

from the screenshoot it must be colab

[trail iris]

Yes, you can.

[fleet sleet]

Do you know how to contact with creators of RVC?

[trail iris]

It's open source so it's hundreds of people

[proper shale]

not really

[fleet sleet]

Do you know how to remove accent from a model, the model is in English and when I speak in Portuguese it has American accent.

[proud elbow]

otherwise it would be more likely just you having messed up some code

[trail iris]

I use a voice because I want to use that accent on purpose. That's the whole appeal of my favorite voice is the strong accent

[wispy burrow]

I've been wondering this for a while.
Why aren't RVC Realtime and W-Okada included in this guide?
https://docs.aihub.wtf/

[proper shale]

good question ngl

it should be there but... idk 😭

[fleet sleet]

Do you know how???

[quaint beacon]

yup im using that one

[proper shale]

id suggest a reinstall ngl

[trail iris]

Remove the accent? Sometimes I tune to use the accent to my advantage

[proper shale]

havent seen that error ever

[quaint beacon]

im on the latest 18a version. Maybe a diff version? I'll try the reinstall first tho

[fleet sleet]

I'm talking about me, I would like to remove accent from a model that I created(American English). when I speak Portuguese it has the American accent.

Did you get it?

[wispy burrow]

Yeah...I get that error on Ilaria RVC Beta...
I can't train the model.
No solution was found.

[fleet sleet]

????

[trail iris]

You're only going to get out what you put in for training data

[fleet sleet]

So there is no way to solve that?

[trail iris]

That's how it works. Use the voice with an accent you want.

[fleet sleet]

I'm trying to dubbing a English YouTube channel to Portuguese

[trail iris]

So use a Portuguese native voice

[fleet sleet]

It would be awesome if it didn't have accent

[trail iris]

If you use a Chinese voice. It's going to have a Chinese accent.

[fleet sleet]

I'm a native

[trail iris]

I use a specific Japanese voice because I love the accent.

[trail iris]

You made the model of your own voice recordings?

[flint elm]

check DM ^^

[fleet sleet]

No, I made a model through audios of American youtuber

[fleet sleet]

I speak Portuguese fluently and I'm trying to translate his videos Into portuguese.

Did you get it?

[trail iris]

The model will keep the accent of the voice.

On purpose

[fleet sleet]

Isn't there a way to resolve this issue?

[trail iris]

Are you trying to dub it with the original speakers voice.

You can adjust the index ratio for a little difference

[fleet sleet]

Yeah

[fleet sleet]

It still has the accent

[trail iris]

All voice ai software will keep the accent of the voice. Some people like myself prefer it this way for music.

[timid ruin]

whats a good website to remove background noises?

[proper shale]

https://mvsep.com

create an account btw

[timid ruin]

thank you!

[proper shale]

for bg noise and stuff, use MVSEP Demucs DNR or Bandit Plus, see what works best for you

[timid ruin]

<3

[proper shale]

yw :)

[fleet sleet]

Using this website it takes a ton of time to do the thing.

[proper shale]

if you have an account n stuff, it has a small queue

[fleet sleet]

Free account or paid account?

[proper shale]

free

[fleet sleet]

Ok

[stark tendon]

How do I make It take percentage from the GPU instead of CPU, i've downloaded the CPU version and the GPU, both doesnt let me select my GPU

[rare goblet]

Ayo? @stark tendon level 1 !!!

[proper shale]

if you're using the AMD version (onnx_directML), the CPU will still be used despite anything you do, but you can convert to onnx and make performance better

Read this section of the guide: https://rentry.co/VoiceChangerGuide#uploading-models-amdintel-gpus-and-fixing-its-laggy-issue

[stark tendon]

What if I download the GPU?

It will still use the CPU?

So I cant use AMD graphics card for that?

[proper shale]

As I said - even if you use the right version, it will still slightly use the CPU

just make sure you're using the RIGHT version though

[brave garnet]

Voicechanger Download (Okada)

⠀
Download for Nvidia GPUs
Version 18a cuda

Download for AMD GPUs
Version 18a directml

Download for Intel GPUs
Version 18a directml

Download for Mac
Version 17b Mac
⠀

Made with love by Ilaria 💖

[proper shale]

^^

[stark tendon]

ty!

[proper shale]

Also do this part of the guide, you need to convert the models u download to onnx:

https://rentry.co/VoiceChangerGuide#uploading-models-amdintel-gpus-and-fixing-its-laggy-issue

[fleet sleet]

Can I create a new voice(not cloning) through RVC AI?

[proper shale]

ehhhh, kinda

you can mix 2 voices and technically create a new voice

[fleet sleet]

So if I use two different voices in a dataset, it will create new voice, right?

[proper shale]

no

that's the thing, it's not abt the dataset

you can mix .pth's

[fleet sleet]

How to mix?

[fleet sleet]

Can I do that using RVC disconnected- Google colab?

[proper shale]

no

iirc you can only do that on mainline RVC

which is kinda... stuck on local

[fleet sleet]

I see, did you tried that? Did created a new voice?

[proper shale]

yeah sadly not yet

[fleet sleet]

What happens if a use different voices in my dataset?

[proper shale]

frankly... idk

might sound weird

[fleet sleet]

I see

[proper shale]

there's written guides, which are honestly much better (because they arent filled with misinfo, like some videos are)

https://docs.aihub.wtf/ -> voice conversion, not realtime

https://rentry.co/VoiceChangerGuide -> realtime

https://rentry.co/RVCRealtimeGuide exists too

[proper shale]

yw :)

?

monitor option, if you're using W-Okada

set that as ur headphones

[rare goblet]

Ayo? @brittle wing level 2 !!!

[proper shale]

oh

well shit 💀

i don't think you can

but you can always use the discord "hear yourself" thing

it's better sometimes

or that too

you're welcome! lmk if you have any more questions

[fleet sleet]

Do you talk in the chat?

[proper shale]

?

[fleet sleet]

In the voice chat

[proper shale]

yeah no

[fleet sleet]

In the voice chat

[fleet sleet]

Why not?

[proper shale]

i prefer texting, for the most part

[fleet sleet]

I'm in a voice chat right now in AI hub

[proper shale]

yeah i saw

[fleet sleet]

They speaking in Turkish

[proper shale]

oh 💀

[fleet sleet]

Hahaha I'm in my quest to find a way to remove accent 🤣

[proper shale]

good luck on it man

[sour dagger]

I have a questions before i go waste my time, but does rvc work on amd cards?

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[proper shale]

kinda

for inference (converting voice, including realtime), yes

[fleet sleet]

Thank you, do you earn money using RVC AI?

[proper shale]

but for training you'd need to use ROCm

[proper shale]

not really

[fleet sleet]

Really

[sour dagger]

Well thats good to know and im sure there are models for almost anything. Also where can i post/share the outputs?

[fleet sleet]

Not even a cent?

[proper shale]

i mean, you can, but i havent really done that in a while

[fleet sleet]

How?

[proper shale]

by taking commissions, but for that you need model master

[proper shale]

I wouldn't really recommend sharing outputs here (if they're covers, for example)

what's your GPU

uhm

oh, so it's integrated

[sour dagger]

i wasent planning on here i meant like places like youtube or social media. oh boy there are alot of people talking to you, hopefully im not being a pain.

[proper shale]

well you'll have to use w-okada colab

https://colab.research.google.com/github/hinabl/voice-changer-colab/blob/master/Hina_Modified_Realtime_Voice_Changer_on_Colab.ipynb

[jaunty shale]

quick question, for 5 minute dataset, which pretrain type do you prefer?

[proper shale]

ohh

yeah u can share em on yt

[proper shale]

go for original or itaila ig

[sour dagger]

would i have to make an alt account by chance?

[proper shale]

it's cause you have integrated graphics, it wont work yeh

[proper shale]

if you're scared of strikes, yeah maybe

[jaunty shale]

haven't tried the rest apart from original and OV2, is there any difference between those?

[proper shale]

yeah use the colab

[sour dagger]

Thank you for the help, and hopefully i wasent a pain in your hair.

[proper shale]

see, itaila is pretty good for short datasets and can apparently deal with noise, unlike ov2 which collapses with noise

original is just good enough for most things tho

yeah use #🔍│help-w-okada

[jaunty shale]

alr noted

[fallen grotto]

does anyone recommend a google colab for ai covers, the one i currently used doesnt have public url anymore idk why

[proper shale]

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

or https://colab.research.google.com/drive/1mHKTGH5e3SAyDSBss1KtiYRbDdQzwSMs

[fallen grotto]

thankss

[proper shale]

your pick

[proper shale]

yw

:)

[proper shale]

you arent a pain dw

;)

[stark tendon]

Which one i have to set it to, to consume my GPU?

[proper shale]

check task manager performance tab, it should have ur gpu number

and then select that

[fallen grotto]

is it just my laptop because it doesnt show a public url

[proper shale]

is gradio down again

try the second one

[fallen grotto]

oh wait nvm i got it

[rare goblet]

Ayo? @fallen grotto level 1 !!!

[fallen grotto]

yepp i already got it thx tho

[sour dagger]

Thanks, also sorry for the semi-late reply.

[solemn fable]

guys what do i do

[proper shale]

oh dw bout it

[proper shale]

...use txt? for timestamps? not sure

[solemn fable]

great.

[calm crown]

Is there a link for applio where i can make ai covers?

[rare goblet]

Ayo? @calm crown level 2 !!!

[proper shale]

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

[solemn fable]

no way there is 41 people queued for astra labs

i thought the numbers would go down

[calm crown]

Is it working?

[proper shale]

should be

afaik

[calm crown]

Ok ty

[stark tendon]

its not working, i've tried the number on my task manager and the other ones, gpu0, 1, 2 and 3, i've tested each of them and they only take cpu, none of the GPU

[proper shale]

you do also need to convert models to on x

onnx*

[stark tendon]

did that

[proper shale]

its still gonna use ur cpu slightly

as ive said

[stark tendon]

ik

the problem is not that

the problem is, its not using ANY of my gpu

and 100% of my cpu

[proper shale]

some people have said that increasing s thresh reduces the cpu usage? but i dont think its 100% true

try it anyway

[stark tendon]

nothing changed

[proper shale]

as i expected

well... idk, to be honest

[frosty osprey]

salutaions im new here , i would need to ask for help, i just started and my model sounds very robotic which parameter could i change to make it sound less buggy, chunk over 640 starts to not listen to my mic input

[proper shale]

what's your gpu?

[frosty osprey]

NVIDIA Geforce RTX 3060 ti

[proper shale]

what are your current settings

[frosty osprey]

can i post screen shots ?

[rare goblet]

Ayo? @frosty osprey level 1 !!!

[proper shale]

now you can

[mossy nova]

I keep getting this error.........JSONDecodeError Traceback (most recent call last)
<ipython-input-5-2ae6516e3f4b> in <cell line: 31>()
31 if os.path.exists(config_path):
32 # File exists, proceed with creation of creds and client
---> 33 creds = Credentials.from_service_account_file(config_path, scopes=scope)
34 client = gspread.authorize(creds)
35 else:

5 frames
/usr/lib/python3.10/json/decoder.py in raw_decode(self, s, idx)
353 obj, end = self.scan_once(s, idx)
354 except StopIteration as err:
--> 355 raise JSONDecodeError("Expecting value", s, err.value) from None
356 return obj, end

JSONDecodeError: Expecting value: line 1 column 1 (char 0)

[rare goblet]

Ayo? @mossy nova level 1 !!!

[frosty osprey]

should i choose an other ai voice or is there something i could do to improve the one i chose

[proud elbow]

crepe is too slow, change it to rmvpe

[proper shale]

use rvmpe as ur f0

[proper shale]

outdated colab, use the ones linked in https://docs.aihub.wtf/

[frosty osprey]

oki but it still sounds a bit robotic honestly idk how to improve it

[proper shale]

increase extra, mess with tune and index

[frosty osprey]

oki thank youuu

[mossy nova]

i cant find the ai site link on the page, maybe im just dumb and not seeing it

[proper shale]

go on the RVC > Cloud section and then pick whatever one you want to use (Applio Colab, Ilaria RVC), click on that, and it'll have a link to it along with a guide

[mossy nova]

gotcha thanks

[proper shale]

glad to have helped :)

[stark tendon]

im thinking, it might be that?

[rare goblet]

Ayo? @stark tendon level 2 !!!

[stark tendon]

or nothing related? my gpu is gpu 0, even when i select the 0 or others it just doesnt consumes gpu at all, only cpu

[proper shale]

try that

but idk it doesn't seem to be that

[stark tendon]

yea nothing changed

so what should i do now 💀

it doesnt let me use my gpu

[proper shale]

man idk

[stark tendon]

searched all up in the internet

nothing there

[proper shale]

all thanks to the negligence of support to AMD stuff

anyways

https://rentry.co/RVCRealtimeGuide maybe try this

[stark tendon]

what cpu would be good to deal with the ai voice thing, i actually have a ryzen 3 3200g but my graphics card is a RX 6600 8gb

[proper shale]

I don't think you should upgrade your cpu tbf

[stark tendon]

its just cuz i really wanted to use the ai voice changer

[proper shale]

if you wanna upgrade anything think about a GPU upgrade for a NVIDIA GPU

[stark tendon]

the gpu i have matches with a 3060

[proper shale]

well, not for AI stuff

see, gaming wise, yeah

but Nvidia stuff works for AI, so

[stark tendon]

and theres no way to optimize the consumption of the cpu?

[proper shale]

not really, sadly

but, before you think of upgrading

[proper shale]

try this

[covert glen]

What am I supposed to do? https://prnt.sc/rBSiYy7t4snc

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[covert turret]

Hi I'm trying to use Ilaria RVC in paperspace and when I 'make run-ui' this error shows up :/
'python infer-web.py --paperspace --pycmd python
Traceback (most recent call last):
File "/notebooks/Ilaria-RVC/infer-web.py", line 26, in <module>
import faiss
ModuleNotFoundError: No module named 'faiss'
make: *** [Makefile:56: run-ui] Error 1'

it also happens when trying to run mangio rvc, not just Ilaria's

[proper shale]

i think you need faiss.... have you done the installing of requirements.txt though?

pip install -r requirements.txt

[covert turret]

i'm trying to haha

i thought doing 'make install' did install all the requirements, apparently not lol

[proper shale]

oh

might be that then

[stark tendon]

it is working but now its echoing

for some reason

[proper shale]

that might be mic bleed

decrease ur system's volume

[proper shale]

old colab

use the newer ones

also enable "Input noise reduction"

@stark tendon

[stark tendon]

its on alr

i noticed when i said "echo" it only reproduced "co" out of my mic its eating sylabs

[proper shale]

increase response threshold too give it some extra inference time too

im not really a realtime user so

[covert turret]

'Traceback (most recent call last):
File "/notebooks/Mangio-RVC-Fork/infer-web.py", line 31, in <module>
from fairseq import checkpoint_utils
File "/usr/local/lib/python3.11/dist-packages/fairseq/init.py", line 20, in <module>
from fairseq.distributed import utils as distributed_utils
File "/usr/local/lib/python3.11/dist-packages/fairseq/distributed/init.py", line 7, in <module>
from .fully_sharded_data_parallel import (
File "/usr/local/lib/python3.11/dist-packages/fairseq/distributed/fully_sharded_data_parallel.py", line 10, in <module>
from fairseq.dataclass.configs import DistributedTrainingConfig
File "/usr/local/lib/python3.11/dist-packages/fairseq/dataclass/init.py", line 6, in <module>
from .configs import FairseqDataclass
File "/usr/local/lib/python3.11/dist-packages/fairseq/dataclass/configs.py", line 1104, in <module>
@dataclass
^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 1230, in dataclass
return wrap(cls)
^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 1220, in wrap
return _process_class(cls, init, repr, eq, order, unsafe_hash,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 958, in _process_class
cls_fields.append(_get_field(cls, name, type, kw_only))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 815, in _get_field
raise ValueError(f'mutable default {type(f.default)} for field '
ValueError: mutable default <class 'fairseq.dataclass.configs.CommonConfig'> for field common is not allowed: use default_factory
make: *** [Makefile:59: run-ui] Error 1'

[proper shale]

oh no 😭

[covert turret]

lmaoo

i have no idea what the hell is going on

i used to be able to use it just fine

[wispy lodge]

iirc, that's due to python 3.11. Fairseq does not work with python 3.11

[covert turret]

makes sense ig

so i think i should then downgrade python?

[wispy lodge]

Yeah, 3.10 should work

[wispy lodge]

And this can change soon

[covert turret]

now i'm trying to figure out how to downgrade python in paperspace lol

[nimble siren]

is there a way to synthesize text with python with applio? Thanks!

[stark tendon]

you mean they dont negligence?

[proper shale]

don't think so

[pine hull]

sorry to interupt does anyone know if the google colab thing still works? i've been trying to use it and no luck, keep getting JSONDecodeError: Expecting value: line 1 column 1 (char 0)

[nimble siren]

so the only way to interact with applio is through a gui?

[rare goblet]

Ayo? @nimble siren level 2 !!!

[nimble siren]

this is ridiculous

[wispy lodge]

I'm working on support unofficially

[proper shale]

yeah

at least tts wise

[proper shale]

old colab

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[proper shale]

heres the new ones

[pine hull]

you are a hero of our time

thank you!

[nimble siren]

I need to synthesize a lot of text. by a lot i mean audiobooks. In paragraph chunks. Do you have any suggestions?

[proper shale]

yw! :)

[nimble siren]

currently using aws polly and I want to KMS

[proper shale]

u can also check for guides on https://docs.aihub.wtf

[proper shale]

honestly, not sure how you'd do that rn because you'd have to use only 30k words per inference, and it will take a long while

not sure about the 30k limit tho

[nimble siren]

Its silly because I can definitely do it with all of the provided options. Just by hand. I need to hook up my pdf parse - > text chunks - > voice synth -> hundreds of .mp3/wavs -> video editing software

i dont need a high limit

i need to generate hundreds of time

i can generate it per sentence

[proper shale]

oh then applio could do that then

would just take some time

[nimble siren]

yeah im not worried about time. Ill leave my computer on all day if i have to. The problem is I can only generate one file at a time. I can set up a script to do it for me if i had access to load models and generate tts via a programming lang

[proper shale]

oh... yeah idk if you can set up a script

@peak tusk can you do tts on RVC CLI?

🤔

[stark tendon]

I MADE IT

ty so much @proper shale

[proper shale]

YESS

[stark tendon]

ty so much for all ur special attention

[proper shale]

yw

sorry i couldnt help that much

[stark tendon]

i figured it out

why it was lagging and duplicating my voice

if someone comes by complaining about the voice chopping

tell them to lower sample length

[rare goblet]

Ayo? @stark tendon level 3 !!!

[proper shale]

ohhh

makes sense

[stark sable]

how to make ai cover on linux

[proper shale]

https://docs.aihub.wtf/essentials/how-to-make-ai-cover/

use cloud stuff oooor maybe use local

[covert turret]

Does anyone using paperspace know of an RVC fork I can use at all?

[violet heron]

Download mainline RVC and open the sh file

[cursive dawn]

what is the best settings ?

[zealous void]

hey so, ive wanted to install rvc and it gave me an error, saying: Python could not be found. Run the shortcut without arguments to install from Microsoft Store or disable this shortcut at
D:\RVC-GUI-Windows-pkg>
(translated from german)

any help?

-help

[azure marsh]

Command has expired

[covert turret]

seems like you need to install python

[proper shale]

for what

[zealous void]

so just install it from the microsoft store?

[wispy burrow]

RVC-GUI is quite outdated though...

[zealous void]

and like do you know which version? or is everyone fine

[covert turret]

https://www.python.org/downloads/

[zealous void]

ty

[covert turret]

I use 3.11.5 if that helps

[zealous void]

idk, its an old link, do you have a newer one?

[zealous void]

okay ty

[covert turret]

@odd shale I'm trying to get RVC up and running in Paperspace to train some models but I just can't seem to get it to work somehow, I used to be able to use it like 2 months ago but apparently somethings different or I am messing something up. I've already tried to follow the guide but to no avail :/

[wide basin]

how much does audio quality matter when it comes to making a voice model

i have a 40 minute video for the dataset

and my friend has an 8 minute video that i want to add to the dataset

but its noticeably lower quality

[odd shale]

Did you make sure to clean up that dataset?

[wide basin]

yea

[odd shale]

Just use those 40 mins.

[wide basin]

but im wondering if adding the 8 minute video thats lower quality audio will affect the outcome badly

[covert turret]

can you post a snippet?

[wide basin]

alright

[odd shale]

No dataset posting plz.

[covert turret]

i think it might, i just tend to stick with using the highest quality

[odd shale]

Don't use those 8 mins.

40 mins is more than enough.

If those 8 mins are lower on quality, scrap them.

[odd shale]

Please, explain your issue with more details.

Also, it's not necessary to ping helpers.

[wispy burrow]

https://docs.aihub.wtf/rvc/local/mainline/
You can choose Applio, Mangio or Mainline.
the links are in the guide.

[zealous void]

okay thank you

[rare goblet]

Ayo? @zealous void level 1 !!!

[covert turret]

Yup, so I start the notebook, create the 'install.py' script, clone and 'make install' mangio rvc, essentially follow everything in the guide with no apparent issues, until I do 'make run-ui' I get an error 'Traceback (most recent call last):
File "/notebooks/Mangio-RVC-Fork/infer-web.py", line 26, in <module>
import faiss
ModuleNotFoundError: No module named 'faiss'
make: *** [Makefile:59: run-ui] Error 1'

and I just can't seem to troubleshoot it

[odd shale]

Hmmm, i never got that issue.

[covert turret]

I never either, until today lmao

[odd shale]

I used Mangito and never got that problem.

[covert turret]

[keen heart]

Why does it train twice?

[proper shale]

this is just the percentage of the epoch in a given moment

not training twice

[keen heart]

ahhh okk

Is there a way to deactivate this

It takes soo long

[stiff goblet]

tutorial in the pinned message is really complex to understand, written with many terms to make it more complicated. I feel really bored while i'm trying to understand what the hell he's talking about LOL (rvc metrics, overtraning etc). It could be much simpler with clear , brief explanations.

[proper shale]

nope sadly

[proper shale]

best tutorial you'll ever read though

just take your time and reflect

[stiff goblet]

I mean , i really like AI and as a beginner , it's too complicated

[stark sable]

---

Exception Traceback (most recent call last)

<ipython-input-9-74d578b798b3> in <cell line: 21>()
23
24 else:
---> 25 raise Exception("No GPU detected; training cannot continue. Please change your runtime type to a GPU.")
26 gpus = "-".join(i[0] for i in gpu_infos)

Exception: No GPU detected; training cannot continue. Please change your runtime type to a GPU.

rvc v2 disconnected

help

[stiff goblet]

Did you connect a Tesla T4 GPU with Colab?

Check your runtime

[stark sable]

?

[proper shale]

your gpu time ran out

[stark sable]

what i need to do

[rare goblet]

Ayo? @stark sable level 1 !!!

[normal karma]

How do I do this stuff

How do I make the ai voice

https://tenor.com/view/rojo-dick-figures-d-figures-llorar-llorando-gif-15690892931836191766

[calm crown]

Why isnt it giving a public link?

[stark sable]

help

[normal karma]

Ayo send me a link to that

[stark sable]

help

[calm crown]

U mean the link im using?

[normal karma]

Yes...I need a tutorial or something. Idk how to make these things

[hazy magnet]

I dont know a lot about ai models, i would say i'm a begginer and i have a few questions:
What is a pretrain
What is titan, how does it work and how to use it

[calm crown]

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

What is the ip password?

[normal karma]

Is there any tutorial videos on this

[calm crown]

Idk, u just click the two steps and hopefully it gives u a public link where u can make covers

[normal karma]

Ok

[stiff goblet]

@proper shale hey , i've got a little question. Should I export my dataset audio as 32k sample rate from Audacity for training? It's cutoff frequency is 16k. Need for 44100hz exporting?

[calm crown]

Someone help pls what is the ip password, or where do i get it from?

[proper shale]

sure, you can do 32k export then

[stiff goblet]

thx

[covert turret]

it's right there on your screenshot

[thin stump]

Pretraining is trained on hours of audio data and varies depending on the purpose for it. Titan pretrain comes from a research dataset described here with 11 hours of data https://huggingface.co/blaise-tk/TITAN

You can find the links here to use titan 32k as a custom pretrain #✨│ai-help message

[hazy magnet]

Sorry but i dont understand whats the point of using it😭

[covert turret]

to train better models when you have a limited dataset

[stark sable]

---

Exception Traceback (most recent call last)

<ipython-input-9-74d578b798b3> in <cell line: 21>()
23
24 else:
---> 25 raise Exception("No GPU detected; training cannot continue. Please change your runtime type to a GPU.")
26 gpus = "-".join(i[0] for i in gpu_infos)

Exception: No GPU detected; training cannot continue. Please change your runtime type to a GPU.

rvc v2 disconnected

[hazy magnet]

So should i use it when my datasets are longer than 10 minutes?

[stark sable]

how to fix

[hazy magnet]

I think that you ran out of time on colav

[proper shale]

your gpu runtime ended

[thin stump]

The original pretrain had average audio so it's not as clean. Other pretrains are more suited for italian, korean, japanese etc. You can check here https://docs.google.com/document/d/1j9J8A8Oop9bMOHmCs3jDXzPujuD6TQ0Q396rJ0MyuIc/edit

[hazy magnet]

*colab

[hazy magnet]

Oh okay ty

[stark sable]

what it means

[proper shale]

yeah

[stark sable]

[stiff goblet]

I suggest using multiple Google accounts to prevent interruption of training

[stark sable]

thx

[calm crown]

Ok ty

[hazy magnet]

Alright so i have a dataset with 2 languages mixed: korean and english and which dataset should i pick? Sorry for bothering yall but im slow in terms of ai

[thin stump]

Try different pretrains. Rigid pretrain isn't out yet and needs people to test it

[hazy magnet]

Is it gonna sound bad If i use the korean pretrain on the datasets?

[rare goblet]

Ayo? @hazy magnet level 3 !!!

[thin stump]

It all depends on how clean your dataset is for the best results for every pretrain

[hazy magnet]

One last question. How to use the pretrain in google colab?

[thin stump]

You need the G & D huggingface links to use a custom pretrain

[zealous void]

idk why but whole thing doesnt work, because of python or sth help =(

[hazy magnet]

OKAY TYSSSMMMMMM

have a nice day/night!

[thin stump]

yw, these aren't stupid questions at all. No one asks much about pretrains and I assume they have it all figured out

[hazy magnet]

im kinda new to the whole thing and im trying my best

[stiff goblet]

Is increasing of loss/g/fm OK ? The other ones : g, mel and kl are decreasing.

[brittle wing]

can someone help me rvc

does anyone know the best way to trim audio for my dataset without it losing any quality? what program should i use and what format should i export it in?

[violet heron]

What do you need help with

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[brittle wing]

Each time i use it its sounds so bad

and i don't even know if im using the right one

[bold canyon]

does anyone know any apps or anything rlly that helps make the rvc output more realistic or buggy??

[thin stump]

Are you ok with 🏴‍☠️ ⛵ ? Because model creators uses RX10

[brittle wing]

yea

any specific export settings i should be using for that?

i just need to trim audio files down to remove parts with no sound

and remove parts with harmonys

[thin stump]

yeah i got mines for $5. After downloading that, you can refer to this guide https://rentry.org/RVC-dataset-RX-updated

[brittle wing]

because i have stems

[thin stump]

and for stem separations, use MVSEP

with an account

[brittle wing]

i dont need seperations

[rare goblet]

Ayo? @brittle wing level 1 !!!

[brittle wing]

i obtained actual stems

[thin stump]

Oh, but you said the harmonies need to be removed right?

like people talking over each other

[brittle wing]

ill just cut out the harmony part its unseperable and quality > quantity

i have enough stems to use only the single vocal parts of the song

[thin stump]

rx10 isn't good for separation by the way. It only helps clean datasets and make smooth cuts because you can work on the spectral waveform

[brittle wing]

ik i dont need seperation just denoising and stuff

is it required to convert your dataset to mono before training?

[thin stump]

oh my bad, I missed a part about what you wanted. You should convert to mono when your cleaning the dataset either way. RVC will also convert it for you

[brittle wing]

do you think i should still clean my dataset if its pure acapella stems from the artist?

[thin stump]

Studio sessions can still be noisy even if you think it's HQ

the areas selected are the noise

it's also in the guide I linked

[brittle wing]

yea that is true sometimes theres even instrumental bleed if their headphones were loud enough

thankyou

[thin stump]

Then you can use MVSEP's Bs Roformer for that. It might do the trick https://mvsep.com/en , np

ofc this doesn't apply to your stems

[brittle wing]

it does kinda in like 5 seconds of one of the recordings there is instrumental bleed from their headphones lol

[rare goblet]

Ayo? @brittle wing level 2 !!!

[thin stump]

If you cut it out smoothly, then you don't have to use BS roformer. options, options

[flint seal]

how do you train an rvc model?

is there a youtube link, or a colab link?

[worn sphinx]

Can someone give me link to google collab with rvc2 plz? The one I have saved doesn’t work anymore((

[violet heron]

Send screenshot

If your settings

[thin stump]

https://docs.aihub.wtf/essentials/how-to-make-voice-models/

@worn sphinx https://colab.research.google.com/drive/1XIPCP9ken63S7M6b5ui1b36Cs17sP-NS#scrollTo=ZodNcumpg-JM

[worn sphinx]

Thx!

[flint seal]

thank yu

[rare goblet]

Ayo? @flint seal level 2 !!!

[flint seal]

ok, so i've seen that the dataset must be in seperate audio files, how long does each audio file have to be? is there a minimum? and is there a maximum?

[violet heron]

No it doesn’t

RVC cuts them up for you

That was only needed in the SVC days

You can if you want but there’s not much of a point unless it doesn’t work

[flint seal]

oh alright, because i saw a tutorial and there was multiple audio files

[hot citrus]

so my g/total graph isnt improvingmuch but my mel and kl graphs are still going down somewhat, will it benefit any more from more training?

[median monolith]

i appreciate the help, but as far as i have searched, this method looks too complicated to me. also, isnt the opus file a lossy file format? if so, then doesnt it make the video, and as consequence, the audio, lose some of its original quality? again, i appreciate the help, but i founded too confusing and complicated to set up and use everything correctly with the videos i have seen.

[fallow ivy]

Hi, I'm using v.1.5.3.18a for windows, framework ONNX(cpu,cuda), PyTorch(cpu,cuda).

I've noticed that my F0 has no crepe (it says crepe(N/A)). How can I switch to crepe for my voice changer?

[thin stump]

I would download different epochs and test them out. It's true that it can keep training if kl and mel goes down, but sometimes the model can sound static-y

So download the lowest point of g/total for sure and the rest you can test out

[hot citrus]

aight ill test out the lowest point and the final model

thanks!

[thin stump]

right, just saying it can keep on training but we don't know how long that goes on for

because g/total is just the average of kl, mel, d/total, fm

[hot citrus]

yeahh i wasnt sure if i should just rely on the toal graph or if i could train it more based off of the kl and mel graph

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[hollow raven]

Is there any fix that RVC dont lag in games like Valorant or Content Warning or such things? Because many use it in Valorant without problems

[hot citrus]

i wanna ask, how do i train a model from a specific point in it's training?

[olive cliff]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[worn spade]

Guys I use easyGUI and i make AI model, but when i put it to drive i get error message ''Access denied with the following error:

 Cannot retrieve the public link of the file. You may need to change
the permission to 'Anyone with the link', or have had many accesses.''

[brittle wing]

So I have a problem where the voice changer echos when people speak.

So my mic picks it up when they speak even though am far away

I fixed it last time but forgotten how

[north niche]

What does this mean, im feeling hella slow.

JSONDecodeError Traceback (most recent call last)
<ipython-input-13-87685dcb27d0> in <cell line: 31>()
31 if os.path.exists(config_path):
32 # File exists, proceed with creation of creds and client
---> 33 creds = Credentials.from_service_account_file(config_path, scopes=scope)
34 client = gspread.authorize(creds)
35 else:

[violet heron]

Is the drive link public

[worn spade]

Yes

[spare sleet]

where to find assets? trying to place a pth file (apologies for the dumb questions im very new to this

[thorn abyss]

can y'all send me some workflows for uvr5?

[thorn abyss]

are you using rvc locally?

or colab?

[spare sleet]

locally

[rare goblet]

Ayo? @spare sleet level 1 !!!

[thorn abyss]

i don't really remember but it should be on rvc/weights

just search weights on the rvc folder

open the folder

and search for assets

[livid frigate]

Where's the hugging face ai voice in github again. I clear my chrome data for no reason

[wide parcel]

i been out of this for a while

Traceback (most recent call last):
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\gradio\queueing.py", line 495, in call_prediction
    output = await route_utils.call_process_api(
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\gradio\route_utils.py", line 230, in call_process_api
    output = await app.get_blocks().process_api(
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\gradio\blocks.py", line 1590, in process_api
    result = await self.call_function(
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\gradio\blocks.py", line 1176, in call_function
    prediction = await anyio.to_thread.run_sync(
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\anyio\to_thread.py", line 56, in run_sync
    return await get_async_backend().run_sync_in_worker_thread(
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\anyio\_backends\_asyncio.py", line 2144, in run_sync_in_worker_thread
    return await future
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\anyio\_backends\_asyncio.py", line 851, in run
    result = context.run(func, *args)
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\gradio\utils.py", line 678, in wrapper
    response = f(*args, **kwargs)
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\core.py", line 58, in run_infer_script
    infer_pipeline(
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\rvc\infer\infer.py", line 278, in infer_pipeline
    get_vc(model_path, 0)
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\rvc\infer\infer.py", line 231, in get_vc
    cpt = torch.load(person, map_location="cpu")
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\torch\serialization.py", line 1028, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "F:\applio\NEW NEW NEW NEW NEW\ApplioV3.1.1\env\lib\site-packages\torch\serialization.py", line 1246, in _legacy_load
    magic_number = pickle_module.load(f, **pickle_load_args)
EOFError: Ran out of input

on applio 3.1.1 precompield

fuck

appreciate it

i aint got no room in my c drive tho 🔥

nope

same error

Traceback (most recent call last):
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\gradio\queueing.py", line 495, in call_prediction
    output = await route_utils.call_process_api(
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\gradio\route_utils.py", line 230, in call_process_api
    output = await app.get_blocks().process_api(
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\gradio\blocks.py", line 1590, in process_api
    result = await self.call_function(
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\gradio\blocks.py", line 1176, in call_function
    prediction = await anyio.to_thread.run_sync(
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\anyio\to_thread.py", line 56, in run_sync
    return await get_async_backend().run_sync_in_worker_thread(
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\anyio\_backends\_asyncio.py", line 2144, in run_sync_in_worker_thread
    return await future
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\anyio\_backends\_asyncio.py", line 851, in run
    result = context.run(func, *args)
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\gradio\utils.py", line 678, in wrapper
    response = f(*args, **kwargs)
  File "F:\applio\NEWEST\ApplioV3.1.1\core.py", line 58, in run_infer_script
    infer_pipeline(
  File "F:\applio\NEWEST\ApplioV3.1.1\rvc\infer\infer.py", line 278, in infer_pipeline
    get_vc(model_path, 0)
  File "F:\applio\NEWEST\ApplioV3.1.1\rvc\infer\infer.py", line 231, in get_vc
    cpt = torch.load(person, map_location="cpu")
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\torch\serialization.py", line 1028, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "F:\applio\NEWEST\ApplioV3.1.1\env\lib\site-packages\torch\serialization.py", line 1246, in _legacy_load
    magic_number = pickle_module.load(f, **pickle_load_args)
EOFError: Ran out of input

yep

didnt load properly

[golden ridge]

i need help. its picking up audio from the people im talking to. so there voice is going through the voice changer as well

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[prisma belfry]

does anyone know why my rvc wont work??

what do you mean??

it takes like 5 minutes than it says error and it just says that the audio was not properly imported or something like that

i dont have a screenshot

what do you mean

im using the website rvc

[rare goblet]

Ayo? @prisma belfry level 4 !!!

[prisma belfry]

i dont understand

chromebook......

i used a link

https://huggingface.co/spaces/r3gm/RVC_HFv2

huh?

model inference

idk what you mean

screenshot what

ok

btw im trying to make something to see if it works

i know

yeah i have a problem with that to

ok

um this happens to it just finished but its really fast and high pitched for some reason????

see listen this is a spanish song but idk why this happens

yes

[rare goblet]

Ayo? @brittle wing level 14 !!!

[prisma belfry]

how do i remove reverb?

can you send me the link for that

i cant cuz im on chromebook 😦

[wispy burrow]

There is a tab for UVR in that RVC.

[prisma belfry]

alright

now what?

[proud elbow]

The current space only uses CPU, so it's only for inference.

even it's not faster than average laptop cpus, so you'd better use the colab or local fork.

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[prisma belfry]

i dont understand your message

[rare goblet]

Ayo? @prisma belfry level 5 !!!

[prisma belfry]

bruh

[proud elbow]

wakarimasen

[rare goblet]

Ayo? @proud elbow level 36 !!!

[proud elbow]

either you're not tech savvy enough or haven't read the guide https://docs.aihub.wtf/

I thought it's just mine

[wispy burrow]

I don't understand...

[primal solstice]

why does rvc sound so different *much better when i switch the output to my headphones instead of the cable that i use in game as an output

referring to game chat

[prisma belfry]

whats the best and fastest website rvc?

[calm jackal]

-rvc

[azure marsh]

Documentation

📚 All-In-One English documentation

https://docs.aihub.wtf/

❔ English FAQ

https://docs.aihub.wtf/faq

Credits: AI HUB FRANCE

✨ More guides

Full AI Voice Model Training Guide (Local)

Link: YouTube
Credits: Christopher Villanueva

Model training with Mainline RVC

Link: Rentry
credits: Raven (ravencutie21)

AICoverGen Colab Guide

Link: Google Docs
Credits: Eddy (Spanish Helper)

Create a model with RVC disconnected (colab)

Link: Google Docs
Credits: Angetyde

How To Make an AI Cover With Ilaria RVC

Link: Rentry
Credits: 👽 Julia (ailen2091)

[pastel oak]

almost all are identical. the difference are mere seconds maybe minute

get mainline from the docs in pinned message

[pastel oak]

1. if you use CABLE then uninstall it and install VAC (Line 1) instead. its on step 6 https://rentry.co/VoiceChangerGuide

[marble forum]

Please help! my code has been stuck here. what does this mean?

its behaving much slower than the last time i trained a model.

@pastel oak

[stark sable]

what i need to do if i closed rvc v2 disconnected collab and nothing is saved

[rare goblet]

Ayo? @stark sable level 2 !!!

[stark sable]

• rate limit

i had save_frequency 10

[calm jackal]

@pastel oak can you please help me

[stark sable]

help immediately pls

[stark sable]

.

how to resume

[queen kraken]

if you didn't do "Export Model from Notebook to Drive" before closing the tab, everything is lost. saldy you have to start training again from the beginning

[stark sable]

oj

oh

thanks

[proud elbow]

make sure you have run the "Set variables" and "Preprocess" cells, also the model name containing only alphanumeric

[calm jackal]

Ok

[tough fiber]

bro im trying to fix this but no one can help

lol

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[brittle wing]

i dont need help as its a one time thing but its just an interesting bug lmfao

my output audio was sped up and pitched up

from 4:21 to 3:29

wait nvm fuck i do need help

it wont stop

every time it outputs its sped up

original vocals btw

wait what the fuck its only for that model???

i changed model and it was fine

[tough fiber]

it is exist i see lol

i deleted whole thing and i upload

somehow its work

even i see a error

weird

[topaz ridge]

--colab

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[frosty widget]

Hello, if I send you a sound file, can you convert it into a sound model?

pleas contact me in dm

[astral jungle]

Hey

[pastel oak]

Its reall not hard to do it yourself
https://docs.aihub.wtf/rvc/cloud/applio-colab/

[hybrid cove]

could anyone help me out please? I can't get the RVC to download the voice 😦

[slate halo]

i've trained a voicemodel and got the .pth and index files. what more do i need to do to upload it to hugginface and use it in google collabs via sharable link?

[rare goblet]

Ayo? @slate halo level 1 !!!

[slate halo]

i'm pretty new to this so i thought just putting them in a .zip together was enough but im seeing you need to make a config file and tokenizer and idk how to do that lol

is there a guide or tutorial somewhere, i cant find it in the documentaion of the colab im using or on youtube

[low shard]

config and tokenizer? no you need only the one pth and index file

[slate halo]

oh, what are the config and tokenizes for?

[dusty blade]

got an error message and it says "Illegal combination of I/O devices"

something wrong with the input output?

[pastel oak]

nput and output have to be the same type
(MME) at the end

[dusty blade]

how do i make it not laggy?

[naive elk]

trying to train feature index doesnt do anything

[low shard]

what error do u get and which rvc

[pastel oak]

what gpu do you have

[naive elk]

it just does nothing without any errors and mangio

[rare goblet]

Ayo? @naive elk level 1 !!!

[willow fable]

@pastel oak so rvc works but for like 2 seconds then laggs then voice play for 2 secs, etc.. i made sample legth 2.40 (the latest i can) but still, now i should get new gpu right?, nothing will help at this point? x D

[pastel oak]

You were the one with the 920mx right?

Yes please do yourself a favor and buy a new one lmao

whats your budget? maybe i can make a recommendation @willow fable

Do note that you might need to upgrade mainboard or anything else aswell to make it compatible

[willow fable]

what is the minimum though? to run it smoothly

[rare goblet]

Ayo? @willow fable level 2 !!!

[willow fable]

do you know? 👀

[willow fable]

thinking of buying a whole new computer package

since i'm using a laptop

i can't upgrage gpu 😦

[pastel oak]

RVC is much lighter than wokada so I am thinking the minimum is something with 4gb vram but I would still ecnourage you to get 8gb vram

if you can get an RTX XX60 (so like 2060, 3060, 4060, or 3070 etc. just not the xx50 ones) thatd be the best

else afaik GTX 1660 (super) runs decent aswell and is very budget

wouldnt recommend 1650 though

[willow fable]

okay thanks ♥

[scenic gull]

-overtrain

[azure marsh]

How do I know if my model is overtaining?

All-In-One Guide on how to make a good model

This guide explains how the D and G files works and much more: https://rentry.org/RVC_making-models

Credits: LUSBERT

Automated Overtraining Detection (AOD)

Will be available soon in #1159513888199540817

Credits: grvyscale

[scenic gull]

-pretrain

can someone tell me how to use a pretrain

[low shard]

check #1233407331405004954

[scenic gull]

i was just reading your guide lmao

im having abit of trouble understanding the concept of a pretrain

[low shard]

oh lol, hope the guide made you understand more, if it didn't you can ask me

[scenic gull]

so basically there's a pretrained model and my dataset fine tunes it?

oh your guide was great but im not good with tech so im struggling a bit

also if my dataset has a mixture of different languages should i just make the model from scratch?

[low shard]

well, basically, rvc already offers an original pretrain, which is like a mix of different voices, when you train using a pretrain, it basically finetunes with your guide, so everytime you train anything, like a model, you will also get the D & G files which are the finetuned pretrain files

The concept of pretrain is a mix of various voices trained to make so that training a model will be easier to do

like if i make a pretrain with different spanish voices, it will help you when you use it for making spanish models

[low shard]

do you wanna make a model or pretrain?

[low shard]

dw all good

[scenic gull]

im making a model

[scenic gull]

so does pre training help with both the model quality and the time it takes to make a model?

[mystic marsh]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab
• 🆕 UVR5 NO UI for Google Colab, by Eddy Google Colab

How to use?

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[low shard]

its suggested that you dont make it from scratch as it would take like, TONS and TONS of hours, its better if you use a pretrain which is phonetically close to the languages that your character speaks

btw could you explain me wdym with a mixture of languages? like your making a model of the character being dubbed in like spanish, english, etc all in one?

[low shard]

yep, helps alot, training without one would make things alotttt harder

[scenic gull]

its the model of a singer and shes done songs in both her native language and english

[low shard]

okay, if you want you can look up #1235952130855010365 or the list in my guide to see if there are any pretrain with the language phonetically close to both her native language and english

[scenic gull]

if i use titan (which seems to just be english) would that still work? (although im aware that it would be much more ideal to find a pretrain with multiple languages)

[low shard]

for english absolutely, but whats the singer native language?

[scenic gull]

korean

sorry should have just made it clearer from the start lmao

also im using mangio rn, is that ok? should i get applio?

[low shard]

nahh dww