🧬│ai-chat

[polar flax]

read the pinned guide in #🔍│help-w-okada

[mellow fog]

so okada is good

right

[dapper ginkgo]

@tepid basin here too

@wary silo go to #1159290752195633273

[covert lake]

What's ur PC GPU

[weak hare]

when when i talk it sounds distorted

esp when i breath

[covert lake]

Ask and elaborate in #🔍│help-w-okada

[prisma yacht]

do people here help with making ai image models orr is it just voice

[prisma yacht]

cause i tried training an ai imagae model using flux dev but it doesnt generate me the safetensor file

[polar flax]

I'm not sure which trainer used but there should be save every step option, just wait till it saves

[prisma yacht]

im using pinokio idk what do u mean by save every step

theres a sample every step but it doesnt work and the model says it finishes training in like 3 mins

[late quarry]

What model is good for deep voice naration?

[polar flax]

like this in flux trainer comfyui workflow

[prisma yacht]

Oh

I dont use comfyui i used pinokio

[prisma yacht]

idk how ot use comfui :/

is there like a tutorial how to use it

cause i have it installed

[solar torrent]

Blame people who keep telling me to do AI locally, not me.

[teal dove]

Ey

I have got a question

How do i make voice overs for free?

Like those in viral brainrot edits?

[solar torrent]

I've been in disbelief in people from previous AI Hub for a while. Is there anything I can help?

You mean AI cover? Well, there are options.

If you have GPU that's newer than GTX 10xx series in your PC, you can do RVC locally. If not, you might wanna look for cloud servive like Google Colab and Weights instead.

[polar flax]

huh? who told you to train flux locally? 💀

[solar torrent]

RVC.

[prisma yacht]

=

[covert lake]

mostly voice rn but u can #🔍│help-ai-art

[covert lake]

what's ur pc gpu and explain more detailedly what u want to do, Speech To Speech (STS) or Text To Speech (TTS) ?

[prisma yacht]

HELp

[prisma yacht]

oh im a bit slow i didnt see that channel

[fair snow]

yo wsp , im from sa . is available tiago pzk voice ?

[polar flax]

search in #🔍│find-models or weights.gg

[fair snow]

i find it, what is the best easy option to generate ?

[polar flax]

ilaria RVC https://huggingface.co/spaces/TheStinger/Ilaria_RVC

[fair snow]

thnx

[polar flax]

or if you have capable gpu, you can try local inference (read the docs below)

[rare sorrel]

Not available yet

[polar flax]

-rvc

[rare sorrel]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[covert lake]

what's your pc gpu

[fair snow]

NVIDIA GeForce RTX 3050 Laptop GPU

[covert lake]

How much vram?

[polar flax]

4 GB, surely

[fair snow]

yes, it is

[solar torrent]

Is Weights still giving me hidden ads when I got premium? Because I see numbers for blocked ads on an adblocker extension I installed.

[covert lake]

nvm then it's better to not do it locally, but you got Cloud (remote good pc, easier and faster than ur PC but it's limited):

• Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
• Ilaria RVC Zero: fastest and simplest that you can get for free
• Applio Colab: max 4 hours, not granted, of GPU

Easiest manual (you still have separate vocals and instrumentals): Ilaria RVC Zero

Easiest possible and automatic: Weights.gg

[polar flax]

it is still fine for inference imo, unless you care of the speed

[ebon nacelle]

Do you guys know if there's an open source f0 editor (like melodyne)? With note moving and modulation editing etc

[solar torrent]

I know not all people who have a very fast PC can spot on where's an ad hidden somewhere. However, my laptop is notably slow, so I launched up Chrome task manager to end task of huge load of ads.

[fair snow]

thnxs

[polar flax]

it could be potential for RVC devs to implement, except the polyphonic detection unless they implement a new f0 method which is probably SVS-based: https://arxiv.org/abs/2401.16837v1

[ebon nacelle]

So there's not something that exists already? I would have thought there would be because melodyne has been around 20 years. I don't mean polyphonic, I just mean for vocal editing

[rare sorrel]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[thin warren]

.

[gray rover]

I've been thinking on that too and could do some research later

[ebon nacelle]

🤘🏼Very interested. Btw did anyone get BigVgGan going on a fork with RVC? I'd be curious cause bigVgGan v2 is out now, supports up to 44k, and is apparently way faster.

[night lake]

Noobies did get it working but it was to memory intensive

https://discord.com/channels/1159260121998827560/1159290193619189821

[night lake]

You can look in there to find what they are working on

[ebon nacelle]

Says #no access

[night lake]

Try grabbing the AI testing role

[gray rover]

no, it's not faster

it's a actually much slower than hifigan

[ebon nacelle]

@gray rover I'm saying bigvgan v2 vs bigvgan

[gray rover]

I mean yeah but why comparing bigv 1 to 2

in any case, we're having mrf hifigan and refinegan being tested atm

so, bigvgan isn't a priority anymore

[orchid sparrow]

Hi what is more recommanded? W-okada or Voice.ai?

[gray rover]

w-okada

voice AI or kits are things we don't quite support

[orchid sparrow]

I had W-okada i think but i forgor which one i used in the past and when looked for W-okada mostly voice ai vids poped up

Thank you very much for letting me know

[covert lake]

https://cdn.discordapp.com/attachments/1159290193619189821/1283652791931506759/sponge-bob-night_.gif

[gray rover]

-rt

[rare sorrel]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[gray rover]

check the fork one

[covert lake]

ofc don't follow yt tuts

[gray rover]

^

Majority is wrong or outdated or simply following the trends / hype disregarding performance or cost

[orchid sparrow]

Yeah i see the one that showed W-okada to me first is gone now and now most vids about ai voice changers are promoting Voice.ai

By the way why Voice ai is considered the wrong one? Its a scam or smh?

[gray rover]

it's closed source

[covert lake]

paywalled + uses your voice for training

Wokada is literally free and open source

[glad nebula]

scam

voices are paid

[gray rover]

most of things like kits or voice AI are based upon open source stuff that's been pretty much taken, monetized and then claimed to be 100% own

[orchid sparrow]

Ah yeah i remember that when i looked up "If w-okada is safe" everyone were saying it is open source so its most likely safe

[gray rover]

• paywalls n scams

[covert lake]

fr and the only way to get them for free is by getting coints that is by letting it be a crypto miner basically

u have to keep always the app in background

[glad nebula]

scam +´cryptomining

[orchid sparrow]

Oh damn

Why people keep promoting that?

App that basically slowly destroy your GPU

[covert lake]

And I can confirm this bc I used it myself very much years ago, literally once, then deleted it bc i realized i dont even need a realtime voice changer

[covert lake]

The same people who would promote Opera GX after it's company has a record of spyware lol

ig money

[gray rover]

because most people are idiots, paid or simply don't know any better

👀

[covert lake]

for people who don't know about it: https://rentry.co/operagx

[gray rover]

wouldn't worry too much tho, you're in right hands here

[covert lake]

yea some people think paid = better ig but it doesn't go like that

[gray rover]

yuh, fully agreed

[covert lake]

not every single paid program is better than free ones

it's just 'easier' and paid

[gray rover]

only really good paid service of this sort I can trust is eleven labs

their style copying is just flawless imo

[covert lake]

I prefer Open Source tho

but true

11labs is good asf

[gray rover]

ye, hoping for open source to get closer to El levels ASAP

[orchid sparrow]

I remember using W-okada before for streaming (Got voice owner permission) and had to delate when i was cleaning discs

it was pretty fun and good quality

[covert lake]

yup, Wokada uses RVC (Retrieval-based-Voice-Conversion) models, which are the top open source Speech To Speech models

best quality since like a year

[orchid sparrow]

Can they also be used as Text to speech? I forgot if it was a option

[covert lake]

Well they are made for Speech To Speech natively,

Do you want me to send you a paragraph that explains you how to use them as TTS?

[orchid sparrow]

Sure please! Il copy and paste it into notebook and do it

[covert lake]

There are different Text To Speech (TTS) AIs:

GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/

Freemium 11labs: A easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS

FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site

With RVC Models:

RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)

If you wanna do tts locally with RVC Voice Models (if you got a good pc):

• You can get Applio in our docs
• While Ilaria RVC Mainline here (no guide as of right now)

If you don't got a good pc you can do tts with RVC Voice Models on cloud:

• Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide

• Use Applio UI Colab (with google colab T4 free daily limit gpu)

• if you don't wanna use edge tts, you could try another tts ai from our tts index and use the output as an input in rvc

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 56!

[gray rover]

applio has tts, or does it not work anymore / there's better solutions?

wondering cause I haven't really explored tts field lately

[covert lake]

I sent you different types of TTS

If you want to use RVC models, i explained that too

And in the 'tts index', you can find how to use TTS in realtime too for calls

i mean, 'realtime', you still have to type

[orchid sparrow]

Alright, thank you very much!

[covert lake]

It does work
Ofc RVC models aren't natively made for TTS tho

[gray rover]

True that

[covert lake]

An actual TTS program would be better than it technically

[covert lake]

Like GPT-SoVITS, F5 TTS or FishSpeech

[gray rover]

perhaps I could sometime integrate gpt sovits into my fork

[covert lake]

so RVC+GPT-SoVITS ?

[gray rover]

yeee, been thinking of it for a while

[covert lake]

You're welcome

[gray rover]

shouldn't be too hard

[covert lake]

I remember seeing ur pfp somewhere

is it stream overload or smt

[gray rover]

ah, you probs mean akiyama / yakuza dude

he used to have ame too

:' )

[orchid sparrow]

Yes, it is Needy streamer overload

pretty good psychological horror/sim

[covert lake]

I mean, you want to generate a speech with GPT-SoVITS then use RVC over it?

Would work i guess

[covert lake]

ye I heard of it sometimes even on TikTok but never really played it

[gray rover]

that's one way, I just thought more of simple integration of two into 1

[covert lake]

Maybe I should give it a try

[gray rover]

as in, unified interface

[covert lake]

mm interesting

[orchid sparrow]

Its a pretty specific game, if you like staring at screen clicking stuff and doing different strategy then it is good

[covert lake]

Wouldn't it make the project a bit too big tho 😭

[orchid sparrow]

But it is not fast or that interesting game if you already dont like slow and monotone games

[covert lake]

I mean I usually love story or fighting games

I played even visual novels, as long as the story is interesting

[orchid sparrow]

If you love story games then you should like Needy streamer overload, it is story based

[covert lake]

For example I played Bad end Theater, the story was good

[covert lake]

mm yea worth a try

my favorite game would be undertale

[orchid sparrow]

I played it too!!!!!!!!!!

[covert lake]

undertale has peak story

[orchid sparrow]

Really good game, i loved Ending song

[covert lake]

Oh lmfao never thought i'd find anyone else who played that

when i finished it a year ago, i tried searching for a fandom, but it was almost not existent

[covert lake]

agreed, the ending was also unexpected

I don't remember everything tho

[elder willow]

@covert lake can you help me

[gray rover]

what's the issue?

[covert lake]

what's the matter ?

wtf i got the notification 10 mins later 😭

discord moment

[elder willow]

i dont now how i can download it

[covert lake]

Could you clarify what you want to download?

[elder willow]

can you help me in a call

[covert lake]

And also what's your pc gpu?

[elder willow]

i would like to download okada

[covert lake]

Sorry but I can't really in a call, but I can text and show you guides tho

[elder willow]

i cant send Pictures in the chat

can i add you ?

[covert lake]

let's go over #🔍│help-w-okada

[sweet cargo]

hey i need rvc for pc

<@&1159293140440723499>

[wooden jolt]

Follow the guides

[river adder]

Why do they have mod ping perns

[wooden jolt]

i want to know everything

[stark scarab]

https://docs.ai-hub.wtf/essentials/whats-rvc/

[river adder]

No why can they point ppl

[marsh moss]

hlo

[dire plume]

@hidden grotto

[hidden grotto]

:wave: @dire plume, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[dire plume]

finally found

[gray rover]

Hmm... @minor blade I wonder, is blase in here under some different name or, totally ditched discord?

[minor blade]

I'm sincerely not sure.

[gray rover]

hmmm.. then who's the main maintainer after Blaise who's authorized to do any changes to mainline?

[stray violet]

Yo guys is there a website for making ai cover?

[minor blade]

Ask either Vidal or Deivih.

[gray rover]

A, ye, Vidal then

Thanks turt 🥬

[minor blade]

🐢: Lettuce!!! bites the lettuce

[minor blade]

You're welcome buddy..

[gray rover]

✨

[covert lake]

what's your pc gpu first

[random karma]

Hi,

I’m an Senior AI engineer, and I’m open to work. I’m excited about the opportunity to contribute to innovative projects.

If you have a great idea and need the expertise of a senior engineer, feel free to DM me.

Thanks!

[dapper ginkgo]

[ruby tapir]

hey if anyone here working w a reference model hmu

[ionic pumice]

#✦│chat message

[sage fulcrum]

Good day, can i ask if you guys have bot that humanize texts or paragraphs effectively?

[thin tiger]

What is the best AI image generator right now?

[gray rover]

@night lake hey man, as I'm on adding features to the ui, aside of configurable warmup and avg loss per gen / disc, do you see any other gimmick like that being added?

or perhaps @tepid basin or @glad nebula

Any propositions will do, as long it's within my capabilities

[tepid basin]

ill have a good idea at some point

[gray rover]

[tepid basin]

i am bad at thinking

[edgy bloom]

Congratulations UnitedShoes (by Weights)!

Your Ivysaur is now level 25!

New move!

Your Ivysaur can now learn Take Down!

‎

‎

‎

‎

[night lake]

all i can think of rn is being able to change the optim

[gray rover]

Oooooo, that sounds good actually

✨

'll see what I can do 'bout it
for now, it lands in the wip category tho. I believe I'd have to prepare some presets or more stuff required to have it configurable

[glad nebula]

changing the lr in the ui might be neat, in a small dataset a lower lr than 1e-4 helps a bit

[gray rover]

added to the list ✅

[night lake]

can maybe also add this: #🔊│ai-development message

[gray rover]

The avg loss ?

[night lake]

yeah

[gray rover]

Actually

that ( configurable ) + use of warmup ( custom duration ) + toggleable mute files
will be in initial fork's release

( soon, hopefully )

man, ngl, applio's handling it really well, the args n modularity 😌
Glad I made the switch

[elder willow]

😔 how do I make rvc not sound like siri when using discord

[gray rover]

Siri...

Can you elaborate?

oh yea raz, what would you say on ability to delay discriminator or gen? idk, sometimes can be useful
I suppose, I could also see custom grad norm setting too

[elder willow]

Its kinda hard to explain. Was gonna do voice trolling on a friend but the audio sounded super robotic when it came out.

[gray rover]

oh, well, then that's due to A) wrong pitch settings B) your inability to mimick the source speaker / aka adaptation or C) the model is just trash

[night lake]

custom grad norms sound interesting but idk about delaying gen maybe disc it just doesnt seem useful in many scenarios

[gray rover]

True, but might be there's a scenario one wants to experiment with giving a generator a lil headstart ( in case more than 1 disc is in use for example )
to not overburden a fresh gen

[elder willow]

yeah combo of b and c

[gray rover]

yet, that's most certainly C if you mention super robotic / rough / metalic and so on

[night lake]

yeah, you could add it and add text that says its only really useful in multi disc scenarios

[elder willow]

Are their any current voice models that have been trained in emotion or are all of them dead pan or monotone?

[gray rover]

I mean, ask people who make the models

If a model's trained on not diverse set or on a one that's lacking in pitch / expressiveness department..
there's nothing you can do

( man that f delay is triggering smh, completely not fitting my writing style oof )

[elder willow]

can I train one with an amd card or no?

[gray rover]

Tho ye Tachi, it's a matter of how the model was made and on what set

[gray rover]

as far as I know yes

provided it ain't 4 gigs

[elder willow]

I have a 7800xt

[gray rover]

Not sure on the exact workflow but, this should be somewhat useful?
https://docs.applio.org/applio/getting-started/installation

There's a section for AMD but again, I am not the one who worked it through, ideally you'd want to ask Noobies 5663

I think

[elder willow]

alright thanks for the help

[night lake]

Could also maybe add an option to change rmvpe's hop length iirc you said it was doable

[gray rover]

oh yea

tho I am still unsure if rmvpe even likes that

afteral, it's still not there ( as in, slider for that )

'll need to test it sometime n see how it behaves

[chilly lake]

you can definitely change rmvpe's hop len

[gray rover]

In that case, lands into wip

[night lake]

whats the wip list looking like?

[gray rover]

wip/todo list:
from the ui level;

• Different optimizers choice
• Custom independent learning rate for G and D
• Adjustable hop length for rmvpe
• Custom gradient norm value
• headstart of chosen network's element: G/D

from the training code level;

• more warmup options ( Cosine anneal and so on )
• more customization here n there, more automations that currently's done purely in config etc.

concept stage:

• Configurable discriminators pairing

Currently added:
from the ui level;

• warmup settings

from code level;

• mel similarity metric

wip rn:

• avg loss settings
• refactoring some descs etc.

@night lake pretty much

And beautiful theme ✨ afteral
lol

[glass junco]

my lil wayne ai working on with my beat me talking ahj

[night lake]

oo cool thx

[gray rover]

mh mh

off topic but, I think I'll ditch mechanic for translations

Am too lazy to dive in and re-type / re-translate all ngl

[glad nebula]

avg loss is so op

[gray rover]

😌

yes, wish people had realized it sooner :L

[analog current]

Hey everyone! Been working on the Alignment question for awhile and finally think I have a working framework!
Check it out and share! https://github.com/AlignAGI/Alignment/

[tacit needle]

guys how i get a relistic girl voice

i want troll my firends

[polar flax]

how about the log interval option in the ui? (having to edit the config file is quite a hassle)

[gray rover]

It's actually automated now in Applio so

No more point in doing it by hand

These however, esp the running loss, I'm pretty sure people will love

Will make picking the right epoch even easier

[gray rover]

Unless you mean a custom override?
but then I hardly see any point for that

[polar flax]

it's like smoothed loss value?

[gray rover]

Nope, something else

Currently the logging is done per last step in an epoch

it's naturally not the best indicator of model's performance

the avg running is averaging the loss per N steps / mini-batches

say, in ur case it's 48 steps per epoch, you'd set the avg for.. idk.. 6

you get the point

better actual epoch performance than just logging of the last which is just biased

Another thing that'd make it better ( and actually correct ) in future is proper evaluation phase with accuracy on unseen / validation set

[gray rover]

Cause If I'm wrong ( Noobies might correct me on this ) the smoothed loss is just smoothed running avg of standard logged loss values

whereas this one is per steps within an epoch specifically

If you need graphical representation
( standard behavior ):

[cedar mist]

i think im having a small issue all the voices and models sound pretty good to me when i hear them by ear but when i use thjem in discord or a game they're choppy

[gray rover]

Well ye, that's because of the gpu's overload it seems

either have to upgrade your gpu, tone down with voice changer settings ( sadly, get a higher latency ) or tone down with game's settings

[polar flax]

try this: https://rentry.co/VoiceChangerGuide#discord-crackle-fixes

[gray rover]

Or simply try the fork in case you haven't .. ah ye or that

[cedar mist]

with discord its not working well either

[polar flax]

yea the fork version is more recommended to try for better gpu utilization/less cpu bottleneck (regardless if you even have a 9800X3D)

[elder willow]

Can anyone suggest me a bot to do ai covers?

[polar flax]

or probably the candlestick graph model lol

https://tenor.com/view/pepe-stonks-pepe-peepee-pepe-peepee-up-only-gif-9717535959180192346

[gray rover]

Don't question my paint skills ™️

[polar flax]

I was thinking of each candle bar for each epoch

and then some adopted metrics like moving average, etc.

[gray rover]

I mean, there already is moving average

the pic depicted stock behavior of rvc / applio in logging

the running avg would work like so: ( depending on your N value )

It's the most straightforward approach without any gimmicks really

any overcomplication is just diminishing returns really, for next level we need evaluation phase 👀

[polar flax]

for example, maybe the fluctuation range, so we could spot some collapses inbetween epoch

[gray rover]

that's also no more the case, it's already figured out

Mute files + silences in the dataset

reason is, if some tiny slice ( 36 frames ) happens to be the spotlight one, that " collapse " happens, ye

Solution for that is simple, properly truncate the silence or resign from mutes ( which is yet to be fully tested )
Aside, if you need a direct indicator, it's already a thing actually, My mel similarity % loss

if you see it being any high value such as 70-90~ % range at such spike, then that's that ( but that's still just a diag metric really )

So, don't think there's any need for fluctuation frequency metric, provided users carefully take care of sets ( as it should be, the truncation of silence to reasonable levels )

[tacit needle]

i have no idea what that is

crypto

[gray rover]

Learn then

[ionic pumice]

you're one funny fella arent you.

[tiny blaze]

W forlorn

[solar torrent]

Skill issue.

[barren mauve]

hey

[solar torrent]

For Virtual Audio Cable, I let a web browser to output its audio to Line 1, from Line 1 to Reaper audio software as microphone to add some effects in real-time.

With paid version of VAC, I think you can use virtual Line In more than 2 at the same time. I have an idea about this: A program -> Line 1 -> an audio software -> Line 2 -> a voice call program like Discord

Or something like this: A program that outputs audio -> Line 1 -> W-Okada -> Line 2 -> Discord

[lament knoll]

best girl voice realistic

[covert lake]

That doesn't exist

There isn't a best one

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.gg/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[robust chasm]

hello ai hub by weights

[worthy coyote]

this won't stop

[dense heart]

[solar torrent]

[minor blade]

https://tenor.com/view/money-way-gif-9856378095025464049

🐢

[glossy fox]

https://huggingface.co/wok000/vcclient000/discussions its not virus?

[covert lake]

no its not

Wokada is the program to use RVC (Retrieval-based-Voice-Conversion, Speech To Speech Models) in realtime for calls

There's the fork (modified version), the deiteris fork which has better performance

that one you sent is the original

@glossy fox what's your pc gpu?

be sure to not follow yt tuts

[glossy fox]

rtx 4060

[covert lake]

that's good enough

-rt

[rare sorrel]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[covert lake]

1st link

its the wokada deiteris fork

meaning it's better

u gotta read it up ofc, there's no updated video

and follow the nividia version as u got an nvidia gpu

@glossy fox for any issues, ask in #🔍│help-w-okada

[glossy fox]

okay

[glossy fox]

So this link is safe?

sorry im have bad english I don't understand well

[covert lake]

yes but u should use the wokada fork not this one

that one is the original wokada

the wokada fork has better performance

both of those are safe

[covert lake]

if you got that link from a yt tut, you shouldn't follow it, bc its old even if it's safe

[glossy fox]

okay thx

[covert lake]

yw

[glossy fox]

yeah from yt

this video upload in 15 october

[polar flax]

must be 2022 or something lol

[untold arrow]

which one do i download

[gray rover]

covers or real-time?

Man, you gotta love how people ask a question but then have dd status ✨😂

[covert lake]

for doing what and what's ur pc gpu?

[covert lake]

nahu shouldn't follow that

[covert lake]

dd status?

[gray rover]

don't disturb

[covert lake]

and also playing a game 😭

while asking the most generic question

[gray rover]

yup, like, kinda wasting the time of someone who's willing to help

[covert lake]

fr

[covert lake]

!howtoask

[pine acorn]

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.

[gray rover]

lmao, good one. totally forgot this existed

[covert lake]

u gotta elaborate buddy

[gray rover]

Don't spam:

This channel isn't really for advertising ~ Perhaps, try #1159290752195633273 ?

[half charm]

hey guys do you know how to create ai art. using reference images? like i send ai these reference images and make it create art? in that style

[steady surge]

uvr is so fucked how did it do this

its perfect

[gray rover]

wtf is this

what kind of power is this..

[steady surge]

heres the timestamp of the original song https://youtu.be/qCXtErcqR4Q?si=0ugBz2Oeh0w88X17&t=246

idk how the fuck it knew those were vocals

the model is MB-Roformer-InstVoc-Duality-v2 with 256 segment size and 50 overlap

the 50 overlap made it take 20 mins to separate

it is a 6 min track though

[gray rover]

I mean, because it is vocal afteral

vocoded? yes, chorus, pitch-tune n shit ton of effects? yes, but nevertheless, vocals

AI too chad 😎

[steady surge]

its closer to static than vocals

[gray rover]

oh, you mean the background elements?

In any case, might still resemble somewhat vocal-type of data in spectrum so

[covert lake]

#1159290752195633273 else it will get deleted

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 58!

[covert lake]

@random karma

[vagrant plank]

I have a small question, is it legel to use voice changer reading a novel and post on tiktok, for example using gura's

[gray rover]

I suppose this topic's quite complicated, afteral, as far as I know there's currently no universally accepted / correct law in terms of voice clone models
But I suppose, if we speak of pure legality or ethics, it is somewhat " misappropriation of identity " ?

Models on their own are tools afterall, they fall under gray area category ( or so I see it that way at least.), just like cameras and stuff but the results / effects of usage matter the most

Now, hear me out. This is my own personal, thus subjective, opinion.. Using someone's voice for public work can be at times problematic especially if target person deems it as not right for them. if I was to do something similar to your example or really anything of that sort, I'd make sure to not cause any harms or misunderstandings.

( That means, including the label of the material being AI generated and /or a lil disclaimer you're willing to delete / remove it if requested to do so by appropriate authorities )

[elder willow]

aHR0cHM6Ly9kaXNjb3JkLmNvbS9iaWxsaW5nL3Byb21vdGlvbnMvZlBEV0VtbUQ2YjdSWHVEVnV1WmZGcEdx (Free nitro) if you can figure out how to get the url

[analog cosmos]

yoink

[covert lake]

already redeemed

[covert lake]

b64 smh

lmfao u actually got it

[analog cosmos]

[covert lake]

I thought it was fake

[analog cosmos]

same lol

[analog cosmos]

it was a epic games nitro gift

[covert lake]

those are still valid????

i thought they weren't after getting abused asf

[analog cosmos]

its a new one

[elder willow]

too slow

[covert lake]

oh lol

not like i have billing info

[elder willow]

actually if your not stupid make a bunch of discords and then put cc on it and just use it hella nitro easy for boosting and sh

[covert lake]

yea people abuse that alot

[dapper ginkgo]

bro thats just base64

[tepid basin]

@stark scarab your ffmpeg installer is based

im trying it on the next unsuspecting ai noob

[stark scarab]

it's just doucle click xd

[tepid basin]

doesnt get more simple lol

[astral ravine]

Ada yg punya model ai anime blue lock?

[covert lake]

speak english here

also

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.gg/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[tepid basin]

@covert lake best de reverb model for dataset and where to run it?

👀

i dont really care about sample rate cutoff cus i already hit it with melband

wait a sec... melband is 40k??

[ionic pumice]

theres no english to speak if they can't speak english to begin with xd

[polar flax]

it's fullband 44.1k unless your source is 48k

[tepid basin]

that is awesome sauce

@gray rover I have a suggestion for applio ui

who threatened blaise with a lawsuit

im gonna find them

[gray rover]

huh, that def sounds helpful

You got any reference or a nice tos to re-adapt?

ngl, I suck in such things ( if I am to make one on my own, that is lel

[night lake]

Ctrl c ctrl v applios

[gray rover]

oh

I suppose, you'd mean a thing where I nor people associated with the project hold any legal responsibilities for copyright infringement, damages blabla.
Thus, user is fully responsible for the use of the tool

Yea, sounds logical

[night lake]

Yup

[gray rover]

Yea, will do in a bit

we've got time, dw

[nimble iron]

Do you poop

[gray rover]

Don't ping me for no reason

Any trolling is not something I'll tolerate

[nimble iron]

It's a simple question

Yes or no

[gray rover]

There is no simple questions, do not ping me is a do not ping me

This is a warning

[nimble iron]

No

[gray rover]

Too many douches ended up getting banned or kicked lately, if that's what ya want or aim for, eventually, go ahead but immature or annoying people aren't well welcomed here

[nimble iron]

Mhm good day to you sir

[night lake]

Pov your mom didn't give you enough attention:

[gray rover]

lmao

[quartz thorn]

Real

[polar flax]

such a ragebait troll

[tepid basin]

Codename dodging the poop allegations

[gray rover]

ikr, smooth chacha

[glad nebula]

[analog current]

Hey everyone! Been working on the Alignment question for awhile and finally think I have a working framework!
Check it out and share! https://github.com/AlignAGI/Alignment/

[gray rover]

You could actually move it to #1159290752195633273

[analog current]

Thank you, will do!

[covert lake]

I think it's UVR-DeEcho-DeReverb

[rigid spire]

Greetings, can i show the video about my custom bio inspired model of artificial brain right here and ask for a few thoughts about it?

I didn't release it yet so it's nothing to send in promo i guess, i will make that it in a few episodes

[ionic pumice]

isnt it waves clarity vx dereverb xd

[edgy bloom]

Congratulations kar@those who are jolly : 🎄!

Your Dewott is now level 28!

[covert lake]

#1159290752195633273

[amber crag]

If anyone wants premium on Webtoon, here’s a code for one month of premium for free: ZvTWhMzKDaXaqPe (in case someone doesn’t know, Webtoon has comics, videos, etc.).

[meager remnant]

i need a little help with prompting with chatgpt, if you have any experience please join call

[ionic pumice]

i barely use webtoon so ion need that but u a goat fr for giving codes like that

[ebon coyote]

yo, we finally got rvc v3 (unofficial by codename0)
https://github.com/codename0og/codename-rvc-fork-3

[wanton cloak]

I am looking for X bot developer who also has exp in RAG system.
This is long term project, DM with your fiverr or upwork account and github.

[polar flax]

you mean twitter?

[solar torrent]

I don't have a Fiverr account. But like can you please put your at #1159290752195633273?

[wanton cloak]

can you do this?

[solar torrent]

No, I'm not a professional.

[wanton cloak]

do you know any professional?

[ionic pumice]

dude just say don't ping me lmfao it's not that hard

[solar torrent]

I'm not the type to say that.

[ionic pumice]

your choice ig

[solar torrent]

No nut November, but I nutted.

https://cdn.discordapp.com/emojis/906936891482439730.webp?size=48

[analog cosmos]

real

[ionic pumice]

uh huh

[covert lake]

use #1159290752195633273

To promote paid commissions

[stuck frigate]

Hi

[covert lake]

Hello

[river verge]

Wake up, everyone! A new RegalHyperus drum model just released!
Fall in Love Alone (Drum model no. 554)

[barren glen]

@hidden grotto

[hidden grotto]

:wave: @barren glen, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[fervent nova]

jak zrobić własny głos w AI?

[gray rover]

You should use English

but it's your lucky day.
I'm polish ( mogę pomóc jak chcesz ale to za 15 min gdzieś (( ~ mentioned I can help em in 15 mins ))

[minor blade]

My Canserbero AI álbum will release on my channel in 5 hours 😄

Nope, didn't use Suno nor Udio

[gray rover]

oooo, lemme know when that happens. send me a dm then 🔥

man I should f reset my phone's kb

SwiftKey been so trash lately

[minor blade]

For sure codyyy 🐢❤️

[gray rover]

🥬

[minor blade]

🐢: bites the lettuce

[gray rover]

Good turtle

[tepid basin]

Try out FUTO keyboard

Local autocorrect model that trains on your device. Very similar experience to gboard. Custom voice to text models too

[gray rover]

Yooo now that's more like it

Been waiting for proper solutions like that. Hopefully it's customizable similarily to swift ✨

[tepid basin]

I haven't used Swiftkey but switching from gboard was easy

I don't do multilingual either

[gray rover]

Oh yea, I do

soooo we'll see how it goes

[tepid basin]

You can export the model to a weight so you can tranfer across devices

[gray rover]

tbf, switfkey fucked up the moment microsoft took over

[tepid basin]

Oh wth

[gray rover]

prior to that it was so good damn

[gray rover]

DM

[stark wind]

I got a question

Anyone know how to make a text to speech bot for discord with voices that it can send through text???

[pearl condor]

hello I would like to cover my friend's sound for one song but how do I add it to the zip folder?

[echo kraken]

Hello

Do you think it's worth subscribing to Perplexity?

[tacit widget]

🤔 For those who know about LLMs, how fast should an RTX 4070 & 32gb ddr5 6400mhz typically take to generate long responses. On Llama 3.2-3B-Instruct

[lime patrol]

hello everyone

[gray rover]

Btw, Can you tell me more about it?

cause it feels like there's something I missed

[slow swan]

ai content is used in this

[tepid basin]

I was asking you to remove the EULA button cus I don't like buttons

But its understandable cus riaa will sue tf out of blaise

[gray rover]

it ain't eula tho

it's simple terms of use

[tepid basin]

Terms of service, ya know

I just like complaining hehe

[minor blade]

I'm not even sure if it was some sort of company (either related to VAs or a record label) that threatened Blaise.

[polar flax]

that non-functional option seems just like age check button

[gray rover]

man, it's just one checkbox, chille

it ain't some death sentence

[polar flax]

real

[simple cliff]

so im getting into this, i've found beatrice v2, but it only supports toml files, which i havent found any voices in #1175430844685484042 that have a toml file.. am i doing something wrong?

[severe yacht]

excuse me, where i can use the ai voice models?

😦

[gray rover]

#✨│ai-help ( if you wanna make covers / voice to voice ) #🔍│help-w-okada ( if you want to use real-time voice changer ) and @ me there

[solar torrent]

Your question didn't specify which AI voice conversation you wanted to use with. There are two RVC programs: Applio the RVC audio conversion, and W-Okada the real-time voice conversation that uses RVC voice model.

[glass junco]

CAN ANYONE TWLL ME HOW THEY MADE THESE AI!!!

https://www.facebook.com/reel/1117264942658581?fs=e&s=TIeQ9V&mibextid=0NULKw

[full bridge]

Astralabs

[stuck frigate]

I think about 1 to 3 seconds

Or if you want to make your own llm, you would need a big amount of gpus

[snow sedge]

How long is the prompt and which backend? Q8 gguf model should produce approx 90-100t/s

Which is, well, very fast

[solar torrent]

An LLM model can be trained as fast as Stable Diffusion image generation with 69 GPUs.

[hallow gorge]

i have realtime voice changer client and it takes 2 seconds for me

chat my rx 6600 is cooked

[covert lake]

Haven't even heard of that

just a gpt wrapper

promote in #1159290752195633273 if you're looking to get hired

#1159290752195633273

[meager remnant]

@chilly lake do you have any experience prompting

[chilly lake]

for stabiity AI there are special extensions, also you can ask chatgpt to make a prompt

[meager remnant]

i am a complete noob, i am just trying to prompt my chatgpt to make scripts in a certain way, and it is going south, if you could hop in the voice channel for 2 minutes i would be grateful, but understandable if you cant

[edgy bloom]

Congratulations MW!

Your Squirtle is now level 15!

New move!

Your Squirtle can now learn Water Pulse!

‎

‎

‎

‎

[ionic pumice]

bro has a squirtle

@edgy bloom p

[edgy bloom]

Your pokémon

1　<:_:721476118942580777> Dewott　•　Lvl. 29　•　52.69%
2　<:_:721476119374594108> Munna　•　Lvl. 30　•　21.51%
3　<:_:721474704036069397> Slowpoke　•　Lvl. 41　•　54.30%
4　<:_:721476309389148202> Axew　•　Lvl. 37　•　42.47%
5　<:_:721476203122524251> Sandile　•　Lvl. 20　•　39.25%
6　<:_:721475493525848113> Wailmer　•　Lvl. 1　•　66.13%
7　<:_:721474816607256627> Togetic　•　Lvl. 39　•　35.48%
8　<:_:721475493710266441> Lunatone　•　Lvl. 4　•　45.16%
9　<:_:721476020846329909> Weavile　•　Lvl. 9　•　44.62%
10　<:_:721476203592024095> Klink　•　Lvl. 23　•　56.45%
11　<:_:721475597095665694> Empoleon　•　Lvl. 20　•　56.45%
12　<:_:721476466759434302> Goomy　•　Lvl. 34　•　72.58%
13　<:_:721476020649066537> Tepig　•　Lvl. 21　•　41.94%
14　<:_:721476119039311884> Lillipup　•　Lvl. 9　•　47.85%
15　<:_:721474704220880906> Psyduck　•　Lvl. 36　•　46.77%
16　<:_:721474757941395567> Porygon　•　Lvl. 22　•　41.40%
17　<:_:721474704354836570> Krabby　•　Lvl. 31　•　21.51%
18　<:_:721474816699269181> Hoppip　•　Lvl. 31　•　46.24%
19　<:_:721476389156552824> Amaura　•　Lvl. 30　•　68.82%
20　<:_:721476203491491920> Gothita　•　Lvl. 26　•　47.31%

Showing entries 1–20 out of 67.

[ionic pumice]

my dewott could probably beat yours 😎

/lh

[meager remnant]

dont make me bring my lugia

[ionic pumice]

https://media.discordapp.net/stickers/1277491531195875421.png?size=160&name=thats+kinda+crazy+gang

[solar torrent]

Any voice model idea for draft training on Weights? I have too many premium voice model training items on Weights.

[chilly lake]

https://github.com/vladmandic/automatic/wiki/Prompting

just a guide what needs to be done for different model types

you can go to civitai, look for a model, look for generated examples at the bottom, see the prompts used

[covert lake]

Like who to make? Maybe Meeple from Brawl stars but idk, I don't really make ai covers anymore

[safe wigeon]

Yo guys

What program do i use for ai voices

[covert lake]

that's for image models tho not LLMs like chatgpt

[safe wigeon]

I wanna do something kanye related

[covert lake]

what's your pc gpu? and what are you looking to do (make ai covers or realtime voice changer for calls) ?

[covert lake]

For LLMs like chat gpt, you could consider a roleplay technique
Ex: You're a Movie Producer, you are an expert into making scripts for TV Movies. Make me an initial script and brainstorm me ideas for a movie about Batman fighting the Flash

You can also see many prompts only, such as https://github.com/f/awesome-chatgpt-prompts

[glass junco]

Link?

[dapper ginkgo]

astralabs.ai the website doesnt work but they have a working discord bot.

its in this server too

[glass junco]

Same as the IG video ?

[dapper ginkgo]

astralabs is AI covers only

they only do audio

[meager remnant]

yoo i appreciate it, i will look more into it!

[meager remnant]

thanks you!

[covert lake]

That's mostly for Image Generation Models like SD tho

[covert lake]

you're welcome

[meager remnant]

ohh fairs, yeah mostly looking for scripting

@covert lake How hard would you say it would be if i have 1000 shortform scripts and i want to "fine tune" or train chatgpt to make similar good scripts

[covert lake]

I never really did that tbh

[meager remnant]

ah fairs, appreciate you for the response

[covert lake]

maybe you can check https://platform.openai.com/docs/guides/fine-tuning tho

it's from their guides

[meager remnant]

thank you!🙏 i will check it out

[long ibex]

Modelleri hangi kanlada bulurum

[full bridge]

Yes

[covert lake]

Speak englishhere

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 60!

[covert lake]

yw

[mossy wyvern]

what does this mean:

[VCClient] wait web server...410 http://127.0.0.1:18888/
Traceback (most recent call last):
File "MMVCServerSIO.py", line 260, in <module>
File "subprocess.py", line 1209, in wait
File "subprocess.py", line 1506, in _wait
KeyboardInterrupt
[12972] Failed to execute script 'MMVCServerSIO' due to unhandled exception!

[chilly lake]

dont do this, obviously

[mossy wyvern]

i figured it out

i needed to delete this file:

stored_setting.json

and it loaded

[gray rover]

That was fucking stupid and you shouldn't be trolling people like that.
fyi, there's tons of people who don't even realize they have a gpu, having gotten their first pc so like, cmon man

I could see how at least 1 or 2 people in future would search up for similar error and follow it

it ain't funny, it's malicious

and potentially damaging. Refrain from doing such things

Alr, bet

🙂

@covert lake if you may

we've gotten enough of ragebait shitheads goofy ah kiddos

Tone down

[night lake]

it is fr

brochacho?

[gray rover]

dude's believing in friendship power

[night lake]

you have to be a llm for that

ikr

the temptation to say your mom

[covert lake]

[gray rover]

[covert lake]

nah

[gray rover]

tamed

[warped pagoda]

hi

[gray rover]

yo

sup

[warped pagoda]

bro i needed some help with ai image generation

[night lake]

panda pfp is so based

[gray rover]

#🔍│help-ai-art you'd want to go here

[warped pagoda]

based?

wdym by based?

[night lake]

A word used when you agree with something; or when you want to recognize someone for being themselves, i.e. courageous and unique or not caring what others think. Especially common in online political slang.

The opposite of cringe, some times the opposite of biased

[gray rover]

^

[warped pagoda]

having a panda pfp is courageous?

wow

[night lake]

in this context i mean your panda pfp is great and i love it, its not cringe and you are a good person for having it

[warped pagoda]

btw i cant upload images here so i might not be able to ask for help for a specific image

[gray rover]

You gotta level up first

[warped pagoda]

got it

[gray rover]

by talking orrrr, well.. if you need a speedrun I guess
#🤖│bots
And just keep on writing something random I suppose

[night lake]

@covert lake we need cleaning on isle ai-chat

[covert lake]

just cleaned it

and wtf was wrong with them

[night lake]

thanks 🙂

[covert lake]

thanks for @ me

[dapper ginkgo]

What do you need help with exactly?

[urban ether]

Hi

[red monolith]

Hey, can anyone help me with a video generator?

[urban ether]

No sorry

[red monolith]

I need to make a music video for an ai song i made

[red monolith]

😔

[urban ether]

What is the ai

[red monolith]

Sono

[urban ether]

Who's that

[red monolith]

Oh wait no

[urban ether]

All I know is perfection

[covert lake]

text/image to video AIs:

• Locally (runs on ur pc):
  • pyramid flow (Image/Text to Video)
  • cogvideox 1.5 5b: Image to Video, Text to Video
• Cloud (remote good pc, running on an online website for example, easier to setup):
  • Weights.gg (paid only)
  • pyramid flow (Image/Text to Video) (HuggingFace Space)
  • OpenAI Sora (paid only, in some countries)
  • lumalabs
  • Hailoua AI

[red monolith]

Suno, its not a voice, its an ai music maker

[urban ether]

Oh

[red monolith]

I make the lyrics and it generates a beat and the rest

[red monolith]

damn it sora isnt available in my country

welp

[covert lake]

Welp u can try with others

[torn peak]

anyone know any swamp izzo models

pls @ me in ur reply btw

[rustic sphinx]

short question. Is there anywhere a good tutorial or guide for how I can create my own voice model?

[torn peak]

need to know this too

[rustic sphinx]

allright. I will search here. If u see or know sth please let me know :D

[covert lake]

Replied in #✦│chat

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 61!

[covert lake]

What's ur PC GPU

[covert lake]

What's ur PC GPU

[rustic sphinx]

right now I am on my laptop with 32 gb ram and an ultra 7 and an intel arc. But if it would be better I can get to my pc with an 3060ti and a ryzen 9 with 24 gb ram

[covert lake]

I'm guessing with ultra you meant Intel ultra 7
And that laptop doesn't got a GPU, right?

[rustic sphinx]

yeah

its just for writing like now or lower stuff

I would do that voice model on my main pc with the rtx 3060ti and my ryzen 9 and 24 gb ram

Would that be enough?

[gray rover]

more than enough

tbf, any gpu with 6/8 gigs will do

[rustic sphinx]

ah nice thank you

[gray rover]

I work with 3060 and can assure you it's enough yuh

np man

[tepid basin]

6 gb 1060 can pull through too

[edgy bloom]

Congratulations UnitedShoes (by Weights)!

Your Ivysaur is now level 27!

[gray rover]

yes, using 4 gig is a misery however

[rustic sphinx]

Do you have a guide or know where it could be

[gray rover]

uhhh, ye, there's plenty around

[rustic sphinx]

on this discord

[gray rover]

https://docs.applio.org/applio

hmm wait

[rustic sphinx]

yeah

[rare sorrel]

📚 Audio Guides & Tools

• Creating Datasets for RVC using iZotope RX11, by Cauthess
• Gathering and Isolating Audio, by SCRFilms ❄
• Instrumental and vocal & stems separation & mastering guide, by deton24
• Vocal Mixing Tutorial, by Roomie
• https://mvsep.com/

More commands: -colab, -uvr, -karafan, -overtrain, /help

[gray rover]

@covert lake What's the current guides source that you typically recommend

just in case

[covert lake]

The ultra 7 offers the NPU which is good

But RVC doesn't support it, your desktop would be the best

[rustic sphinx]

allright thanks and what of those links should i use

[covert lake]

The ai hub docs are https://docs.ai-hub.wtf

[rustic sphinx]

the first one right

[covert lake]

Usually it's recommend to use either mainline or applio, mostly applio

[gray rover]

alr thx, gonna keep that in mind

I'd recommend my fork due to the easier training workflow

[rustic sphinx]

ok so should i use applio?

[covert lake]

I don't really know about your fork but maybe hard for a user without guides

Will ur fork also support NPU btw

[covert lake]

Yea

[gray rover]

no I meant like, it's still applio but incorporates the avg metrics which

I believe should be easier for newbies than understanding the last logging is the last step from the batch etc etc

[covert lake]

I'd honestly suggest applio

It got surely more community support

[rustic sphinx]

ok thank u all guys. Any good vids online btw? And do u know any application for anything for my npu?

[covert lake]

I mean if u wanna suggest ur fork u can do that

[gray rover]

In that case follow Nick
and once you wanna dive deeper, lemme know

[covert lake]

But I rather go with stability

[gray rover]

Actually, the fork's more stable as of now and has better readability
no bugs or issues at all

[covert lake]

Nope no yt tuts nor vids

Only written guides

All yt tuts are outdated

[rustic sphinx]

ok thanks

and what about anything with npu? like not online for voices but anything to use that npu

[covert lake]

I think https://github.com/rupeshs/fastsdcpu has NPU support

And well.. that's the only NPU supported program I know

I rarely seen projects support NPU

[rustic sphinx]

yeah same :D

allright now I only need 30 Minutes of my voice

And then I am ready to build my own dataset

[rustic sphinx]

is the fastsdcpu save to use?

[covert lake]

You mean safe to use? Yeah I used it on my laptop and phone, just CPU tho

It's open source, meaning you can even read the code of the project yourself

[rustic sphinx]

ah nice thank you

short question. I got the following error: ERROR: Could not find a version that satisfies the requirement mediapipe==0.10.9 (from versions: 0.10.13, 0.10.14, 0.10.18, 0.10.20)

[notice] A new release of pip is available: 24.2 -> 24.3.1
[notice] To update, run: python.exe -m pip install --upgrade pip
ERROR: No matching distribution found for mediapipe==0.10.9. But i updated pip and installed mediapipe 0.10.20

[hardy kestrel]

https://tenor.com/view/blade-runner-2049-guts-berserk-berserk-mhrcartoon-gif-1293165138765548862

[regal chasm]

hi guyz im new here. im lloking for a good and self-hostable model (or at least runnable with a python script) which i can use to clone italian voices. do you know any models with a good quality?

[glass junco]

Use applio bro so much better

[gray rover]

he's not talking about applio man

"mediapipe" isn't used in applio and from the context you can tell it's about 'fastdcpu'

[hollow tangle]

dipende cosa cerchi

[regal chasm]

Un modello text to voice con capacità di cloning su cui posso fare inferenza direttamente dal mio sistema, no API o altro. Ho le risorse necessarie per farlo

[hollow tangle]

text to voice ti posso consigliare gpt sovits

[regal chasm]

Ci do un occhiata

In italiano come se la cava?

[hollow tangle]

ah merda in italiano non funziona

mi ero scordata

li manca il pretrain italiano

guarda di open source praticamente non c’è quasi nulla per l’italiano

[strong ivy]

po p o

[regal chasm]

Peccato

Proprio zero?

Quindi usate solo servizi qui?

[hollow tangle]

io ho sviluppato la mia robetta ma è più voice to voice

text to voice pure ma usa edge tts alla base

[regal chasm]

Quindi per fare cover o simili?

[hollow tangle]

yup

[regal chasm]

Qualità bassa ?

O edge tts intendi un servizio tts esterno

[hollow tangle]

edge tts è il tts di microsoft, usa quello integrato

[regal chasm]

Temo non faccia al caso mio

Sarebbe carino allora fare un train su qualche modello.

Quali sono i modelli migliori in termini di qualità dai quali partire?

[hollow tangle]

modelli per rvc intendi? la tecnologia di cui ti parlavo

[regal chasm]

Piú text to voice ma anche in inglese

[hollow tangle]

controlla su #1175430844685484042

[regal chasm]

Altrimenti rvc non viene un po robotica la voce?

[hollow tangle]

dipende molto dal modello

[regal chasm]

Understandable

Ora ci do un occhiata e provo qualcosa, grazie mille 😉

[hollow tangle]

di nulla!

[night lake]

I actually have already written it all I just need to make a pr

[gray rover]

mh?

to the main applio?

or wut

[night lake]

I've written a guide for your fork and such on the AI hub docs I just need to make a pr and ask ray to merge it

[gray rover]

oh, in that sense

yea, sounds like a good idea

Thanks man

[night lake]

If you want to look at what I've written so far you can look here: https://github.com/Razer1724/docs_updated/blob/updated_docs/assets/RVC/Local/Codename_Fork.md

[gray rover]

lemme see

[grizzled sapphire]

anyone know what ai was used to make these? there's a bunch on the channel
https://www.youtube.com/watch?v=CTdMM2EtEOM

[gray rover]

A, actually @night lake

Better to mention the releases

those are always 100% guaranteed by me to be safer than repo

as I always do at least 2-3 checkups and files verif before pushing

[night lake]

Yeah, I haven't been able to update it

I've been busy

[gray rover]

yuh all good, just thought it's worth mentioning, just in case

[night lake]

Better be safe then sorry

[gray rover]

as for a compiled version.. welp. It's a long story but in a short, I've been with awfully shitty upload speed lately

so, that's a big rip

[tight canyon]

Ws chat

[gray rover]

ws good guy

[night lake]

Rippp

[gray rover]

ikr man, it sucks lol

glad I don't really upload anything recently

else I'd be f over

[night lake]

How many Mbps you getting?

[gray rover]

.... don't ask.. to put it in a perspective how really bad it is, 3 mb upload takes 20-30 secs

vs 20 dl per sec

[night lake]

[edgy bloom]

Congratulations Razer by Weights!

Your Grotle is now level 27!

New move!

Your Grotle can now learn Mega Drain!

‎

‎

‎

‎

[polar flax]

mobile data user moment

[gray rover]

Yea, long story and you're right

5g my ass, revolution they said 💀 we gonna get our brains melted they said

[night lake]

For the docs how would i explain how to use the avg loss? Like would i just say 'run a single epoch to find its step count then stop training and then set the avg loss freq to a number where it logs an epoch 3-5 times'?

[chilly lake]

running 1 epoch in the training loop to completion just to find the number of steps is so RVC... lame

there's len(train_loader), set it automatically

[gray rover]

well, so what, you recommend to do a synthetic / dummy steps checker for the training set or wut

[chilly lake]

if you say "i want to log 5 times every epoch", then you get len(train_loader) /5 and do that

[gray rover]

" So, I think it's most common to balance these factors by reporting an averaged loss over N mini-batches, where N is large enough to smooth out the noise of individual batches but not so large that the model performance is not comparable between the first and last batches. "

this should be some good info to start with

[chilly lake]

put if global step % value == 0, log whatevr

[gray rover]

We aren't doing logging X times per epoch

it should be based on known steps per epoch

this is a different methodology than what you have in mind

[chilly lake]

" 'run a single epoch to find its step count then stop training and then set the avg loss freq to a number where it logs an epoch 3-5 times'? this should not be a thing ever

[gray rover]

I never said " where it logs an epoch 3-5 times "

let's start with that

[chilly lake]

step count = len(tran_loader)

[night lake]

i was just throwing something at the wall to give an idea

[gray rover]

again

" well, so what, you recommend to do a synthetic / dummy steps checker for the training set or wut " within the ui, before training

I ain't sure if you even get the concept in the first place
You mention that the steps count can be checked by the loader and yeah, that's true
but the idea is for the user to get to know steps, then based on that, deduce and choose what they want

[chilly lake]

why do you need any step checker?

[gray rover]

Alr then big guy, then write a consise strategy for an average user / noob

[chilly lake]

on average, train len can be estimated beforehand

[glad nebula]

expecting average users to do this its crazy

[chilly lake]

user does not even need to decide, you decide how often you wan to calculate what you want to calculate

[gray rover]

Then write down the formula and explain it to Razer, then we good

cause if you like to step in my ideas, at least contribute big guy
( smh )

[gray rover]

No
Again, it is not the premise to decide how many times you want to log per epoch, epoch is not the main topic here
it is about what every N steps count you want the logging to take the place

it is per steps / per mini-batches logging re-interpreted to match " avg epoch performance " premise

[chilly lake]

but it depends on the size of the train?

[gray rover]

I just can't f get it, why can't you at least once not complain about meaningless bs man

[chilly lake]

train is 1000 steps, you want to log every 100 steps.. so that's 10/epoch

100 step mini-batch

or whatever

anyway, this is a crazy conversation, you do you

[gray rover]

You stepping in for no reason is crazy

talking about efficiency ey
fix that up first

[chilly lake]

ahead of you, my dude

[gray rover]

More like craving attention to prove you're the best engineer around despite not being asked to

Just please for the hell's sake, tone down your ego for f once

May I remind you I am a hobbyist that barely knows shit about coding, I do it for fun, because I want to, I add in what I deem good for me and I give an option to use it, not enforcing it. It doesn't have to be tip-top perfect like it is projected in your head

[polar flax]

some rural areas still have 4G LTE tho

[chilly lake]

dont put your words into my mouth and for fking sake, take a break

[gray rover]

You should take a fucking break, tf was you getting the convo in #✨│ai-help scolding down the dude for using pinokio? or whatever stuff, which I already did

If that ain't looking from above at someone goofy, then idk what it is

it was not needed. Just as your ego oversaturation man

also tf was that " applio 'lives' mostly thanks to me " kind of thing?

[glad nebula]

i think i've seen 4G LTE in my phone when the signal is ass lol😭

[gray rover]

Anyway.. you can put in something like this:
"If the total number of steps per epoch is unknown, the user may consider running a short 1-epoch training to determine the corresponding step count. While the choice of the averaging factor 𝑁 ultimately depends on the user's preference, Codename recommends experimenting with an averaging window that accounts for approximately 23% to 32% of the total steps in an epoch."

@night lake

or something, idk

[chilly lake]

.... ""If the total number of steps per epoch is unknown"... " running a short 1-epoch training "... again I'm looking at this in astonishment

[gray rover]

You were all against the averaging in the first place, so please, shut the hell up if you're gonna keep on bsitting around

See? u dunno when to stay quiet man

[chilly lake]

look, you dont need an exact number... number of samples / batch size gives you an average +/- 2 steps

[gray rover]

+/-, that's right

I don't need estimates

all that crappy estimating led to previous crappy auto log syncing, back in the day

[chilly lake]

len(train_loader) not enough for exact?

[gray rover]

Then I ask once again, do you recommend doing an ui based steps calculator / dummy checker prior to user's inputting the n value

either that or running a test run, I only want accurate numbers, not estimates

[chilly lake]

you're asking the user to do the dumbest thing since 'syng graph'

[gray rover]

it wasn't " dumbest " it was what it was at the given time

where were you back then hero

[chilly lake]

I did not know RVC even existed

[gray rover]

too bad then, you'd do way more good back then

¯_(ツ)_/¯

[polar flax]

ah, I see pinokio is a cloud service like lightning.ai

[chilly lake]

you dont even need to run 1 epoch to completion

[gray rover]

you didn't get the point of what I wrote I think, it was about that I already cleared up the man and noobies input wasn't needed

[chilly lake]

you just need to start it and stop right after

[gray rover]

Sure, but I don't want users to develop bad habits of failure-stopping

[polar flax]

ok chill

[gray rover]

that's just facts

was said once, noobies' input wasn't functionally required at all

[chilly lake]

some people just want to overcomplicate the simplest things

[gray rover]

some people have a habit of dismissing certain ideas without having their own input in it

[chilly lake]

make a decision that does not involve users who dont know and dont care about this thing

[night lake]

thoughts on :

"To use the avg loss you need to know the total number of steps per epochs, you can train one epoch to find the step count. Choosing an averaging factor depends on the user, however Codename recommends experimenting with a window that accounts for around 23% to 32% of total steps in an epoch. If you choose to not use 23-32% of total steps be sure that the logging frequency isn't to small because the losses can vary a ton and it can end up confusing you, and make sure for big loss frequency it isn't to big because it may smoothen the noise to much and not give you accurate results."

[gray rover]

Yes, seems good

[gray rover]

man, you're really surreal

Honestly, as I said idk, 2-3 weeks ago? I have no time for these nonsenses coming from ya

[night lake]

ight thanks 🙂

[chilly lake]

so how hard would be put ' size of the averaging window - [ ] 1/5, [ ] 1/4, [ ] 1/3 of the epoch size' instead? 🙂

[gray rover]

Np man ✨

[chilly lake]

23% to 32% seems to be pretty much that?

[gray rover]

Not everyone loves math or schematics or whatever

Some people can't even read right with proper attention span

we know you don't give a f about " kindergarten kids " but cmon, don't complicate what does not need to be complicated

[polar flax]

@gray rover #✨│ai-help message
I remember SCRFilms used crepe hop length 8 for some reasons

[gray rover]

uhhh.. well, I mean, you kinda can, ofc but.. at this 'reso' imagine all the potential error points

imo an overkill

Tbf I personally never go below 64

[polar flax]

btw can you share (again?) comparison between rmvpe and crepe hop length (feature extraction for training) in talking, ASMR, singing (from slow genre to metal screaming)?

[gray rover]

what comparison?

have I ever made such? ( genuinely

tho, as far as my intuition goes, main reason crepe could be smoother is because it uses gaussian smoothing during training, which can lead to softer pitch contours​, I suppose

• adjustable hop ofc plays a role here

I'd really have to ( some other time ) do adjusted rmvp vs crepe, both at 64 and 128 to say something more

but ye, probs the smoothing plays a role here

[polar flax]

(means crepe has less tendency of voice cracks than rmvpe?)

[gray rover]

one might like it, one might hate it. But that'd explain why rmvpe seems more rough and sharp and crepe is soft ( good for females

the voice cracks are really due to ' losing the track ' of the fundamental / f0

when there's harmonies for instance

rmvpe is without a doubt more robust in that area

you basically gain robustness and accuracy in ' noisy / imperfect ' scenarios over accuracy and smoothness in clean audio scenarios

[polar flax]

I saw some ppl have made a choir model

and how about in metal screams?

[gray rover]

well, screaming and growling is basically heavier tendency towards " noise " qualities of the audio than harmonic

same as breathing, sibilants etc

if the model's exposed to it, it can do fine

issue's is, vctk which is the base of pretrains naturally has absolutely no idea of what it is and how to interpret it

and when people typically make models targetted at screamos or whispering, they include too little of that data or even mix it in bad ratios

if ratios are more towards the data the gen / disc learned on the og dataset, naturally bias gonna occur ( or so I believe )

[gray rover]

so pretty much uhhh, if you get a decent base and / or focus mainly on screamos / provide enough data

it's possible for the model to do just fine
But that's really about anything, even gasping or spitting sounds

whatever you'd want ( but then there's hubert aspects which I won't hide, noobies knows more on it so, would be best if he revised what I said just in case. iirc, if hubert doesn't flag it as something meaningful, it doesn't get through or like, not in the exact same shape or form (( again, might be wrong

[night lake]

Ive found for whispering that you only really need around 10 min of data to get it decent

[polar flax]

I'm not sure if KLM includes metal singing

[night lake]

ofc more the better but its a good min

[gray rover]

well ye, but then an example

10 mins of whisper, 1.5 hours of normal talk ( a vtuber

model's gonna be biased towards / better in converging to speech

[gray rover]

then there's also exposition aspects. More data you feed it, higher the likelihood of the model figuring out the heck it works with

[night lake]

yea, you need a good balance

[gray rover]

that's what we mean by proper convergence pretty much

[gray rover]

yup

[night lake]

For anyone wondering this was done on a test of a 30 min set 20 of which was speaking

[gray rover]

that's kinda a decent balance, provided the 2 subsets of the data type aren't too diverging I suppose ( and the bs isn't too extremely low, then the small bs's induced noise + variability in data induced noise would be a mess )

I mean yea, back in the day I'd recommend a similar ratio, perhaps 10-15 mins of this and 4-6 mins of that

[polar flax]

both rmvpe and crepe can be confused on trying to infer harmonies and doubled vocals
I somewhat think there could be a new f0 method to develop that uses something like this: https://arxiv.org/abs/2401.16837v1
(or more precisely, like "singing voice separation")

[gray rover]

I think the best idea of mitigating the pitch confusion errors would be to incorporate some kind of context awareness

and past-now comparitive tracking + prediction, kinda lookahead concept

thanks for the paper. Will read it up

Seems like they do have the code but

Apparently there's no inference code anywhere

idk, maybe in some distant future when I feel like it or am done with all the current work, could dive deeper

Anyways, I head to bed and gonna leave training ( mrf test on 3 Okay no, seems to be hours 8 ( oof ) + ranger test and my tweaks eh ye )
in any case, potential updates gonna land in #🔊│ai-development
cheers all. Gnight ~

[pseudo pier]

Huh

[merry spruce]

I would like to ask how to cold-start an AI product. There has been no user using it.

Are you willing to use my AI product? https://photog.art/

[covert lake]

There are different Text To Speech (TTS) AIs:

GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/

Freemium 11labs: An easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS

FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site

With RVC Models:

RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)

If you wanna do tts locally with RVC Voice Models (if you got a good pc):

• You can get Applio in our docs
• While Ilaria RVC Mainline here (no guide as of right now)

If you don't got a good pc you can do tts with RVC Voice Models on cloud:

• Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide

• Use Applio UI Colab (with google colab T4 free daily limit gpu)

• if you don't wanna use edge tts, you could try another tts ai from our tts index and use the output as an input in rvc

Non c'è quasi nulla in italiano, tranne che Edge TTS, 11labs e XTTS2

Xtts2 é l'unico in locale, potresti provare quello, ma é 0shot e non più aggiornato cmq

Se no dovresti fare tts con I modelli RVC (speech to speech)

[covert lake]

Please tell me what's your python version in #🔍│help-ai-art

@night lake Btw why is styletts2, gtts, bark, tortoise and meta voice in the ai hub tts index ?

I removed:

• gtts: Google translate, shitty quality asf
• bark: after talking with @chilly lake (#🧬│ai-chat message) , realized it hallucinated
• tortoise: old, xtts2 is a fork of it and an improved version
• matavoice & styletts2: not that good and no longer updated

https://x.com/metavoiceio/status/1754983953193218193

Meta voice sounds kinda noisy and they literally don't say anything since March 😭

MetaVoice (@metavoiceio) on X

Presenting MetaVoice-1B, a 1.2B parameter base model for TTS (text-to-speech).
* Emotional speech in English
* Voice cloning with fine-tuning
* Zero-shot cloning for American & British voices
* Support for long-form synthesis

https://t.co/35dTjbUvU2
https://t.co/mKpUYtd5qo

Twitter · Feb 6, 2024, 9:42 PM

[covert lake]

use #1159290752195633273 else it gets deleted

[drowsy hawk]

Hey guys. Is there any tutorial on how to setup qwen 2.5 best or any other llm for coding?
Chatgpt is broken for a couple of days already, and I do not how to exactly set params to the LLM and context:(

[polar flax]

the 72b model requires A100 like in a huggingface zerogpu space, and I doubt if smaller models could still match GPT-4o

or there's also 32b coder variant

Claude is also another best API-based alternative

[drowsy hawk]

a this moment, everything will be better than 4o, as after update they made it literally STUPID asf

is it worth to try buy claude?

[covert lake]

What's your PC GPU?

[solar torrent]