but with extra American timbre
#✨│ai-help
copper marlin
the pronunciation.
has slight chinese overtones.
"vocal range that captures"
that part isn't English-sounding
accentwise
play the recording
half cove
its a spanish model....
copper marlin
ok lemme find another
(spanish isn't... chinese)
surely spanish folk have different intonations
Alice also sounds chinese
copper marlin
was the recording chinese, or was the actual model chinese-sounding
i'm playing the audio button on the Weights website
it has 'ur' sounds, and other sounds that might form part of chinese words, which don't sound like part of English speech even if it sounds similar and can plug the gaps + I can tell when people are using AIs if they are of this quality based on pronunciation errors
I don't want it to change accent. I want to sound exactly the same, but with a different voice (timbre/harmonics i think)
the original speaker for weights models is chinese.
"as a speaking sample"... etc
can one person in this entire server post an English accent example?
i know it's possible
yeah chinese speaker sorry nvm
i found one example of a chinese guy getting a brish accent with RVC
50% accent coverage
silver rock
whenever i try to open the go web file nothing happens
copper marlin
Queen Elizabeth has developed a slight Chinese accent.
in this video.
may be the original speaker, might not be
also, slightly American too
there's something off
ahahahah it was American. the speaker was american
lol
"something off"
like, my accent is 50% mixed with Chinese, and not much of the source material's accent
might be a data training issue
I'll stop being lazy and run it overnight
bleak nymph
RAAHHH I NEED DISCONNECT PREVENTION METHOD RAAAHHH
copper marlin
well, good luck idk
half cove
whenever i try to open the go web file nothing happens
elaborate
half cove
RAAHHH I NEED DISCONNECT PREVENTION METHOD RAAAHHH
i mean now they use captchas so you gotta just stay on the tab
half cove
bleak nymph
i mean now they use captchas so you gotta just stay on the tab
still
bleak nymph
https://tenor.com/view/epic-embed-fail-ryan-gosling-cereal-embed-failure-laugh-a...
breh
silver rock
elaborate
i downloaded nvidia version of rvc and extracted the files, i try to open go web and nothing happens
half cove
i downloaded nvidia version of rvc and extracted the files, i try to open go web...
like nothing does it open anything or just nothing at all
silver rock
like nothing does it open anything or just nothing at all
nothing at all
bleak nymph
guys how do i prevent rvc from disconnecting
going to another house i wanna leave it on
training
cant find ANY disconnect prevention method
azure marshBOT
RVC Guides (How to Make AI Cover)
static gyro
cant find ANY disconnect prevention method
steam link
you can watch your computer on your phone
silver rock
rare gobletBOT
Ayo? @silver rock level 3 !!! 
half cove
send image of ur cmd
silver rock
wait how do i do the path to feature index file
full marsh
is that the original RVC
spare shale
@half cove Do you know what uses https://vocalremover.org/
full marsh
operable program or batch file.
Press any key to continue . . .
how did you install it
full marsh
how did you install it
from github directly downloaded the precompiled version
rare gobletBOT
Ayo? @full marsh level 5 !!!
half cove
from github directly downloaded the precompiled version
what did you use to unzip
full marsh
what did you use to unzip
winRAR archiver
i mean the normal RVC official repo runs with no problem
full marsh
but i wanted to try the new applio
full marsh
oh wait a min...nvm im stupid i found the error
half cove
what was it
half cove
oh 😭
full marsh
thnx fr the help tho homie
west lark
i think google colab doesnt work anymore
it dont want to add audio and there is only somegirl and someguy
proper shale
i think google colab doesnt work anymore
this is an old Colab
west lark
there is a new one?
proper shale
yh
west lark
i was searching through the yt comments for like 20 minutes and couldnt find anything
proper shale
sly widget
west lark
<https://aihubdocs.github.io/en/rvc/cloud/inference/ilaria-rvc/>
really thank you for your time
copper marlin
hello
FileNotFoundError: [Errno 2] No such file or directory: 'C:\Users\cocac\Desktop\AI/preprocess.log'
tried mangio instead of the original RVC
got this issue
who knows the fix?
spaces in file path
-_-
original doesn't give this error
west lark
@half cove do you know what to do when audio file doesnt show up?
i refreshed many times
half cove
<@936673139419664414> do you know what to do when audio file doesnt show up?
just click convert
it should work
i think
half cove
leme try rq
copper marlin
My computer had an overflow error because of the dataset size, and the windows crashed and stopped displaying correctly, then the terminal had white squares appearing on it
Damn
west lark
maybe
copper marlin
How much free space do I actually need?
west lark
anyway thank you so much
west lark
My computer had an overflow error because of the dataset size, and the windows c...
maybe quit everything else youre doing rn like software
copper marlin
maybe quit everything else youre doing rn like software
I had to shut down the computer
It crashed because of the overflow error
west lark
works fine now?
copper marlin
My computer is sluggish which is weird
copper marlin
It says my apps are using 500GB but the apps are all like 1GB tops and I don’t have that many
west lark
copper marlin
Idk where the 500 is from
I actually haven’t experienced this problem yet
130 is usually fine enough to run with
I had it running well with like 5ish gb left over
Probs the start up time nvm…
I’ll send a pic of the crash if it happens again
I can run RVC with 130 free gb?
The original worked fine
Idk
You’re all doing Google collab I assume (or most of you)
west lark
I can run RVC with 130 free gb?
yeah
west lark
You’re all doing Google collab I assume (or most of you)
i am cuz i dont have nvidia
copper marlin
i am cuz i dont have nvidia
I used cuda for my stuff. You can do CPU only if you’re hellbent on offline
Takes time though
Like x10 longer
west lark
damn
rare gobletBOT
Ayo? @west lark level 4 !!!
west lark
maybe ill use it
copper marlin
So if you have a good CPU, Bark TTS takes like 12 mins
With GPU it’s like 20 seconds
Ok that’s a lot more than 10 times longer but you get the point
Tortoise idk
RVC depends
Can be a few seconds or a few hours depends on how you train it
Why does this happen with 130 free GB?
RuntimeError: Failed to load audio: [WinError 1450] Insufficient system resources exist to complete the requested service
never had this before lmao
there might have been a registry edit in a tutorial from ages ago
o_-
west lark
Ok that’s a lot more than 10 times longer but you get the point
yeah i do
west lark
Why does this happen with 130 free GB?
no idea, never had this
ashen notch
someone can send a site with voices?
rare gobletBOT
Ayo? @ashen notch level 2 !!!
proper shale
someone can send a site with voices?
median monolith
.
rare gobletBOT
Ayo? @median monolith level 3 !!!
spare depot
Should the names of the index files follow the convention: added_IVF276_Flat_nprobe_1_<ModelName>_Ov2_v2.index?
- added_IVF276_Flat_nprobe_1_Faputa_Ov2_v2.index
- Faputa.pth
If I get the file like above, is it okay if I change the name as below?
- Faputa.index
- Faputa.pth
bleak nymph
bruhh this thing is gonna take ages
def overnight
how is it gonna run overnight on COLAB
proper shale
Should the names of the index files follow the convention: `added_IVF276_Flat_np...
I mean, you can most likely change it, but I'm not entirely sure
proper shale
how is it gonna run overnight on COLAB
You shouldn't run overnight bcuz it will disconnect and if you use RVC Disconnected you will lose your training files. Therefore it is important for you to stop training and export the model after 1 or 2 hours of training
bleak nymph
yeah thats what i was confused abt
okay
so how do i continue training? there are 2 options but they dont make sense to my pea sized brain
fluid matrix
- added_IVF276_Flat_nprobe_1_Faputa_Ov2_v2.index
- Faputa.pth
If I get the file...
you can change it, but better leave it as it is. especially when submitting a model
proper shale
so how do i continue training? there are 2 options but they dont make sense to m...
once you're back tomorrow, set the same training variables (i recommend taking a screenshot of those), then skipping preprocess/feature extraction and train index steps, run the 2 for resuming steps
the first one, just set to 2333333, that's the number g and d use anyway
second one, just run it
then just run training step w same batch size and you should be good
spare depot
you can change it, but better leave it as it is. especially when submitting a mo...
great. Please make sure I understand. Is this not so that the RVC application has the ability to interpret the name of the index file, but rather so that users who download the model can know information about the index file?
rare gobletBOT
Ayo? @spare depot level 1 !!!
bleak nymph
i unchecked the save last checkpoint box
i have no clue what it does
bad idea i know
fluid matrix
great. Please make sure I understand. Is this not so that the RVC application ha...
yes, the name doesnt really matter. but its important to know what the file belongs to, so just dont change it
but you can
spare depot
good. Okay, okay.
proper shale
i unchecked the save last checkpoint box
oh man
it's the one that saves G and D every saving epoch
so you'll have to check drive and see what's the number
bleak nymph
so you'll have to check drive and see what's the number
ok
where will the number be?
bleak nymph
yeah where is that g_ file
proper shale
yeah where is that g_ file
rvcDisconnected > folder with your model's name
G and D will be there
proper shale
rvcDisconnected > *folder with your model's name*
this is inside your drive btw
proper shale
inside the zip folder?
no not the .zip, that's the dataset
proper shale
yeah im an idiot
nooooo don't say that, just gotta get used to stuff :)
bleak nymph
oh so just the one outside
bleak nymph
yeah ur a legend tyyy
bleak nymph
will do, thanks again
wait i forgot to ask (sorry if its a dumb question, im not 100% experienced with colab) but when im done and want to continue tomorrow, do i click the stop button and close browser? or disconnect runtime
rare gobletBOT
Ayo? @bleak nymph level 4 !!!
bleak nymph
because im pretty sure that would waste credits if i dont disconnect
but if i do disconnect it would delete the entire thing
proper shale
wait i forgot to ask (sorry if its a dumb question, im not 100% experienced with...
make sure you export to drive first, btw 🙏
but yeah, once you're done, click on runtime on the top > disconnect and delete runtime
so it doesn't waste credits
bleak nymph
ok thankss
proper shale
yw :)
stray onyx
how do i fix the "download failed" error when trying to download a voice model in huggingface
it just says this every single time
turbid pagoda
it just says this every single time
i dont think they resopnd as quickly on this one.
stray onyx
i dont think they resopnd as quickly on this one.
ok
wait
i got it to download
but its still not in the "choose model" dropdown
runic stump
Hi, been trying to get Mangio RVC running locally and I keep getting the same error message
File "S:\RVC-V2\Mangio-RVC-v23.7.0_INFER_TRAIN\Mangio-RVC-v23.7.0\infer-web.py", line 79, in <module>
conn = sqlite3.connect('TEMP/db:cachedb?mode=memory&cache=shared', check_same_thread=False)
sqlite3.OperationalError: unable to open database file
Tried multiple installs but no matter what I do I still get the same error. There are no spaces in my username or filepath if that helps.
runic stump
I don't know about huggingface, I could only help you with that one because it is a universal thing.
azure marshBOT
RVC Guides (How to Make AI Cover)
simple vault
reinstalling rvc locally, should i still use the mangio fork + fixes or is it unnecessary now?
proper shale
I'd recommend maybe using mainline RVC, it's updated and works similar to Mangio and is as stable as it too.
Check https://aihubdocs.github.io/en/rvc/local/mainline-rvc/ - has everything you need to know :)
simple vault
I'd recommend maybe using mainline RVC, it's updated and works similar to Mangio...
thankyouuu, ill trust you and go for mainline 😎
proper shale
thankyouuu, ill trust you and go for mainline 😎
Yw :)
Hope you enjoy it
median monolith
Is this normal?
proper shale
Is this normal?
yeah no but tbh idk
wispy gazelle
do you guys know by chance why some rvcs are not working for me?
there is no output sound after talking like
there are no errors in the cmd it just doesnt say anything
proper shale
there are no errors in the cmd it just doesnt say anything
ya sure? check for any Tracebacks
or anything that has multiple lines to it
wispy gazelle
maybe it has to do with this?
proper shale
maybe it has to do with this?
oh
in that case
follow this section
cuz you're on amd, installed models r laggy at first
this mitigates the issue
median monolith
yeah no but tbh idk
So i guess i will have to live with it (?).
proper shale
So i guess i will have to live with it (?).
Yeah... I'm sorry
median monolith
No problem
wispy gazelle
cuz you're on amd, installed models r laggy at first
so i overwrite the pth file with the new onnx one ?
proper shale
so i overwrite the pth file with the new onnx one ?
yup
proper shale
that worked thank you, so other than that i just have to get used to the lag?
yeah
it sadly has lag even on like, 4090
but that's like 40ms delay tho
wispy gazelle
so the better ur graphics card the lower the delay i assume
proven hill
not on pc, cant listen with good speakers
half cove
I'd use ila rvc
full imp
everytime I restart it now, it tells me it couldn't load logs
half cove
What's the problem ur having btw
half cove
everytime I restart it now, it tells me it couldn't load logs
I assume due to inactivity
full imp
"Error: Failed to load logs: Not Found. Logs are persisted for 30 days after the Space stops running."
yea
I mean it works
but idk if that message means something it's wrong
proven hill
yea use me
full imp
w-what
half cove
Also the colab version of ila rvc is faster
half cove
-colab
half cove
but , does that work with google colab
Yup
full imp
it's pinkkk
full imp
Yup
I meant, if I have to go on a google colab and check boxes, cuz I swear I'll cry if I have to do one of those
proven hill
its two single boxes
half cove
I meant, if I have to go on a google colab and check boxes, cuz I swear I'll cry...
You only have to click 2 then just use the gui
proven hill
and it gives you the link
full imp
I see
shut saddle
It’s been some time since I stopped my hype in Ai Covers, and now that I try to do one, google banned the colab
Now I don’t know what to do
stray onyx
does anyone know why many voice models simply fail to download?
half cove
Just use ila rvc
half cove
does anyone know why many voice models simply fail to download?
What are you using
stray onyx
What are you using
ila rvc
half cove
What link
covert turret
lol
stray onyx
anyway, about half the voice models i try to download give this error
half cove
Don't send the link publicly smh
half cove
anyway, about half the voice models i try to download give this error
Replace blob with resolve in the link
half cove
To make it a download link
half cove
Do it
half cove
Like your own Google drive link?
stray onyx
Like your own Google drive link?
no, but some of the voice models in #1175430844685484042 have a google drive link for the model
instead of a huggingface link
half cove
Most old Google drive links don't work but if it's your own just make it public copy link and boom
half cove
no, but some of the voice models in <#1175430844685484042> have a google drive ...
Prob by model reposter bot
Alot don't work
Like almost all
stray onyx
Prob by model reposter bot
so i just have to live with that, then?
half cove
so i just have to live with that, then?
You can check for more models on weights.gg
stray onyx
You can check for more models on weights.gg
so once i find one how do i get the link for it
nvm
found the button
half cove
so once i find one how do i get the link for it
For that you gotta download it and reupload it
Cuz it needs auth
stray onyx
For that you gotta download it and reupload it
reupload it to weights.gg?
half cove
No
covert turret
to drive
half cove
To either huggingface or drive
stray onyx
Make it public and copy link
i made it public but how do i get the link
half cove
Ima go now cuz it's 1am
half cove
i made it public but how do i get the link
Just click the 3 dots and find it
stray onyx
sorry for all the questions, i never use google drive for anything
half cove
When u make public on the bit
stray onyx
half cove
It shoes
stray onyx
it doesnt tho
half cove
On the bit where u make it public
timid remnant
How do I select a previous training step PTH file to upload to w-okada?
I tried going into \RVC1006Nvidia\logs\mymodel and picking a pth file from there, but it gives me an error in w-okada when i try to upload it.
hard chasm
Ask in #🔍│help-w-okada
storm garden
How do I use voice changers ?
median monolith
I plan on making a lot of models wich the datasets are considerably limited (the characters dont actually speak, there is not that much audio file of them making a sound, etc...), and i know that because of this, they generally arent going to sound as good as the ones with better datasets. But still, i want to make them sound the best that they can. And i wanted to know a place in this server wich could help me to do this.
simple vault
for the mangio fork is there a way to get rid of the horrible blue theme?
rare gobletBOT
Ayo? @simple vault level 1 !!!
fresh bluff
once I have my model, is there any good free way I can try it with text to speech? or is that a whole other thing
rare gobletBOT
Ayo? @fresh bluff level 2 !!!
forest sluice
once I have my model, is there any good free way I can try it with text to speec...
Use any text-to-speech app, download the output audio then put it into RVC
fresh bluff
but you have to go through that process first? no free app where you can upload your model and have it go straight text to speech
lilac pilot
so did i fuck up this model? should I keep training it?
grave brook
is there any notebooks i could use for batch infer? cuz my laptop from 2016 isnt powerful enough to
rare gobletBOT
Ayo? @grave brook level 1 !!!
static gyro
so did i fuck up this model? should I keep training it?
yes check mel and F
lilac pilot
yes check mel and F
what is mel and F?
frosty kayak
someone give me some good news...is it done? 801 epochs
molten pecan
someone give me some good news...is it done? 801 epochs
Prob. Show g/total
static gyro
what is mel and F?
look in scalaras, /loss/total g, D mel and F
im going to try the tiny model hack
1 minute of true dataset 100 epochs lets see what happens
frosty kayak
Prob. Show g/total
heres the update (w/ the g/total)...think i may have got excited too early but idk
molten pecan
heres the update (w/ the g/total)...think i may have got excited too early but i...
Oh yeah it started OT a while ago it seems
Try testing the model of the lowest point from g/total
@proper shale what do you say?
grave brook
is there any notebooks i could use for batch infer? cuz my laptop from 2016 isnt powerful enough to(x2)
frosty kayak
Oh yeah it started OT a while ago it seems
ok i always thought you watch to see when the d/total goes up... and the g/total wasnt that important....the lowest point on the g/total was around 200 if im not wrong but its at 815 rn
grave brook
is there any notebooks i could use for batch infer? cuz my laptop from 2016 isnt...
i saw that the RVC v2 Disconnected notebook can use CIL to infer though im not sure if it allows batch conversion via sepcifying audio folder
frosty kayak
also i want my graph back please
put it on scalars, and refresh
lilac pilot
@static gyro
static gyro
its the same screen
lilac pilot
oh wow looks like i started over training
@static gyro what do i want it to look like?
static gyro
is this a small dataset?
lilac pilot
its around 9-10 minutes
rare gobletBOT
Ayo? @lilac pilot level 3 !!!
lilac pilot
minimal silence, I didnt truncate the silence
static gyro
go through your 4 graphs
g total
D total
Mel
FM
static gyro
minimal silence, I didnt truncate the silence
in each one of your graphs where is the point where it goes up in 2 or more?
oof
lilac pilot
static gyro
oh thats right not fm kl sorry
hexed wind
lilac pilot
static gyro
ok next idetify the shift were all graphs change
in this example i see a divit in 8k through 12k
serene wigeon
so, I decided to make my first foree into AI music, following a tutorial that suggested RVC and how it is "super easy and works great with just using default settings", and the result seems to be a garbled inaudible mess.
The best I can guess is the problem that:
- the original audio uses a lot of talkbox effect, which seems to really throw the AI off. Idk how to get rid of that
- the original audio has multiple overlapping voices, even though it's only for a fraction of a second at a time, seems to mess the AI up. I'm not sure how to clean that up either
i'll try to show a sample of what it souds like in a bit
lilac pilot
what am i looking for?
static gyro
graph going up in a similar space the same as this divit in 2 or more graphs
if it is save the step count and make a snapshot
lilac pilot
when i determine over-training how to i identify what epoch it was at?
static gyro
when all graphs go up
flint geyser
-colab
flint geyser
try one of these
icy blaze
how?
static gyro
i need to go to bed now sleep land comes
lilac pilot
when all graphs go up
I get that part but how to I identify what epoch it started overtraining at?
icy blaze
night
flint geyser
for Ilaria RVC Huggingface you open the link then dublicate it
icy blaze
im just new to this whole thing tbh im confused
static gyro
training models late at night be like
azure marshBOT
RVC Guides (How to Make AI Cover)
icy blaze
thx
icy blaze
oof
icy blaze
well im still confused
proven hill
for Ilaria RVC Huggingface you open the link then dublicate it
please start suggesting the colab
hot lagoon
use the lowest loss ckpt
my guess around 6k?
flint geyser
my guess around 6k?
hover around it
icy blaze
can u explain to me why the software for the model is not showing up?
why is everything so hard
icy blaze
thx
rare gobletBOT
Ayo? @icy blaze level 2 !!!
hot lagoon
this is a okay one?
flint geyser
yeah
icy blaze
how i use this...
flint geyser
make sure to run install before run
icy blaze
flint geyser
click this one first
flint geyser
the play button on it
icy blaze
where?
flint geyser
icy blaze
is it run cell?
flint geyser
tis
icy blaze
ok
flint geyser
yeah
flint geyser
good
icy blaze
it says executing in 1 m and going down in s
flint geyser
it says executing in 1 m and going down in s
ok
icy blaze
do i just wait?
flint geyser
k
icy blaze
ahhh i see
flint geyser
its kinda my job here lol
icy blaze
haha
flint geyser
also @proven hill made this rvc fork
icy blaze
/anyway your a nice person
flint geyser
/anyway your a nice person
ty
icy blaze
see ya in a bit
flint geyser
ok click start
icy blaze
where?
rare gobletBOT
Ayo? @icy blaze level 3 !!!
icy blaze
the next one?
flint geyser
good
icy blaze
i have to go in a bit okay?
serene wigeon
since no one responded to me i want to double check, is it appropriate for me to post samples of audio here and ask where I went wrong?
i know i don't provide a lot of context but there's not much to give, I just used the rvc google sheet with all the default options
after isolating the vocals and removing revurb
dawn nebula
since no one responded to me i want to double check, is it appropriate for me to...
only direct ai outputs, no datasets
scenic gull
glad zealot
you can try that model or wait longer, sometimes it still goes down
glossy blade
I have an issue with public url, it's not working. Yesterday it was all good but today it tells me "Unable to register cuDNN factory"
the url just times out after a while
hot lagoon
Was a good fork for stop training ?
grave brook
is there any notebooks i could use for batch infer? cuz my laptop from 2016 isnt powerful enough to
i saw that the RVC v2 Disconnected's repo that the notebook uses can use CIL to infer though im not sure if it allows batch conversion via sepcifying audio folder
main tinsel
What's the difference between the rvc version? I got a hubert v3. Can I train this to the v2 model of the rvc-boss webui?
sand iris
What's the difference between the rvc version? I got a hubert v3. Can I train th...
wait what ?
I'm not following you there
main tinsel
I'm not following you there
umm.. Hubert v3 means our country's Hubert v3. It is korean project. it may not be real v3. So I'm wonder about the difference.
Is there a relationship between the pre-trained model and the server for training? Is it the pre-trained model that determines the model performance? or version of training tool.
rare gobletBOT
Ayo? @main tinsel level 1 !!!
sand iris
umm.. Hubert v3 means our country's Hubert v3. It is korean project. it may not ...
Would you have more info about that Hubert V3 ? Haven't heard of it yet
sand iris
Is there a relationship between the pre-trained model and the server for trainin...
the pretrained model is what you begin training from. It shouldn't change the result by a whole lot, but it can make a small difference in some cases
as far as I've seen, everything trains RVC the same way, so there shouldn't be a difference between the different software
and about hubert, it depends if v3 has the same output as the regular one. if it has, it'll train, else, depends
copper marlin
Is there a relationship between the pre-trained model and the server for trainin...
Does it have a different accent?
main tinsel
and about hubert, it depends if v3 has the same output as the regular one. if it...
Sorry I got it wrong. Sorry for the confusion. it is not v3. Then what's the difference between appgio v3 and mainline rvc v2?
azure marshBOT
RVC Guides (How to Make AI Cover)
brittle wing
howto fix high res
copper marlin
Can I use the weights of one model with another model’s voice, to transfer/preserve an accent without retraining?
maiden meadow
copper marlin
Ok
sand iris
Sorry I got it wrong. Sorry for the confusion. it is not v3. Then what's the dif...
Oh, they are both based on RVCv2
it's just the interface that is different
the technology is the same
delicate flame
how to make voice model ?
delicate flame
how do i make a model
copper marlin
Aightie. You load the voice files, put the folder path
Then click preprocess
Then do feature extraction
Leave it as V2, and follow the menu instructions for singing pitches
Keep the default selections for D and G
Then, you can either do one-click training (Which works from RVC Project) or you customise the training epochs etc
rare gobletBOT
Ayo? @copper marlin level 8 !!!
copper marlin
I’m currently trying to get Mangio RVC working to see if the accents extracted are any more accurate
dull orchid
I’m currently trying to get Mangio RVC working to see if the accents extracted a...
Would like to know this too
delicate flame
after lanching the thing i got an erro
and after i lanched the gui
i got the error "504 gateway timeout"
flint geyser
Read #📰│dev-updates
plush compass
Hey guys, is 8 the best batch size in Colab? Isn't 16 or 20 good? This is a dataset of about 30 minutes.
forest sluice
Hey guys, is 8 the best batch size in Colab? Isn't 16 or 20 good? This is a data...
Try it out
tired rapids
what's the recommended epoch number for around 4 minutes worth of voice recordings?
forest sluice
Try it out
Based on my experience, 8 batch size is the fastest
forest sluice
what's the recommended epoch number for around 4 minutes worth of voice recordin...
hmm, maybe 150-200
copper marlin
Would like to know this too
Despite having 130 free ROM GB, it fails due to being unable to allocate around 5GB
It’s not optimised for my machine
Perhaps there’s a problem with its compatibility with my GPU model
I have a laptop GPU not a desktop one
ah its a CPU error
huh
DefaultCPUAllocator: not enough memory: you tried to allocate 4718592 bytes
sand iris
Despite having 130 free ROM GB, it fails due to being unable to allocate around ...
wdym allocate ?
if you get anything about "failed to allocate", GPU VRAM is an issue
not storage
oh fuck I can't read
copper marlin
i need to change virtual memory allocation to system allocated (potentially)
hot lagoon
Pov batch size 4 :
My not sure why change batch size make it less overtrain as much like this here
it overtrain.
But just me change the batch size to 8
it looks less overtrian
brittle wing
Is there any way to use RVC with an AMD GPU or is Nvidia required
brittle wing
As in, run RVC on linux, or
sand iris
yeah
brittle wing
Is there a reason it only works on linux?
sand iris
AMD has something that 1:1 compatible with CUDA, but they haven't released it fully on windows
copper marlin
sand iris
so on the pytorch end (which is the library most use for deep-learning tasks), there's no release compatible with AMD cards on windows (yet)
brittle wing
Ah, I see. So my options are either run it on Linux, or use some online service like kits.ai
Sucks that those are the only options
brittle wing
For real? Could you give me a few recs? It'd be appreciated a lot
sand iris
-colab
sand iris
whole bunch of links lol. The programs aren't as pretty, but they work 
for model training, there's "RVC Disconnected" or "easyGUI"
and for inference, "AiCoverGen" or "Ilaria RVC" work
brittle wing
I appreciate the help, thank you.
rare gobletBOT
Ayo? @brittle wing level 1 !!!
copper marlin
https://www.youtube.com/watch?v=OUQPuZrzVo4
Doing this and selecting V1 on Mangio RVC (local installation) seems to work for feature extraction now
Before I had the cpu issue
Now I will start training the voice
Ok it seems to work
copper marlin
How we doing so far chat
Do you know why Mangio RVC doesn’t select the inferencing voice even if I have one?
I tried refresh. It works on the RVC Project version
tidal shadow
i cannot open the public link?
copper marlin
I have index files without pth files
CUDA error: CUBLAS_STATUS_NOT_INITIALIZED
RuntimeError: DataLoader worker (pid(s) 8220, 13888, 2856, 5284) exited unexpectedly
does anyone get the result that mangio doesn't work on their machine?
sand iris
what's your GPU
copper marlin
it didn't work until i reduced vm space and then it doesn't work now
copper marlin
what's your GPU
my GPU and CPU all work with RVC Project
sand iris
what are they
sand iris
do you tend to put your computer in sleep state instead of rebooting ?
hardy adder
boys
sand iris
Friend of mine has CUDA die when the PC resumed from sleep
copper marlin
do you tend to put your computer in sleep state instead of rebooting ?
it does stuff like that automatically
copper marlin
Friend of mine has CUDA die when the PC resumed from sleep
damn
my laptop hasn't slept during training
opal compass
hello, i was wondering what is the best link to use to make ai voices, like sing and stuff, i don't wanna train any, but the link i used last night no longer works, so i might aswell ask people who know what they are doing.
copper marlin
can i copy and past pretrained models from Mangio RVC into my working RVC Project directory...? lol
I want the accents to stop being foreign and be the same as mine and the training data's
I havent tried training for more than 2 hours yet
opal compass
oh
copper marlin
realistic but chinese sounding is what i had ended up with
opal compass
well i wasn't training, i just wanna use voices from wieght.gg, and make them do things
copper marlin
what advantages does Mangio have?
opal compass
but the colab link i have dones't work i keep getting a 505 error.
bleak nymph
lmk if anything goes wrong!
hi its me again, can you help me with this error?
copper marlin
I saw a youtube video and a quick train seemed to have a better accent
bleak nymph
hi its me again, can you help me with this error?
this is kit lemonfoots colab
not even mangio rvc
left oasis
what is it? I've never had that before.
copper marlin
this is kit lemonfoots colab
there seem to be many RVC's out there
opal compass
what is it? I've never had that before.
that is the same error i keep getting
rare gobletBOT
Ayo? @opal compass level 1 !!!
opal compass
i don't know what is wrong
copper marlin
internet connection error
opal compass
it loads fine and stuff but then i get that 504 error
copper marlin
probs?
I have only gotten that when trying to connect to some servers ( from memory)
idk why never bothered checking.
left oasis
that is the same error i keep getting
really?
opal compass
yeah
copper marlin
not your problem, it's the server's
copper marlin
copper marlin
so it will be working soon or something
this is why i always train locally
on my own pc
lol
opal compass
well im not training
im just trying to make ai music videos and stuff like i normaly do
i don't even know how to train
bleak nymph
hi its me again, can you help me with this error?
does anyone know how i can fix this error
copper marlin
i don't even know how to train
I have a question for you about the accents of your ai songs. Are they Chinese, or do they sound as accurate as the original source data?
this was an issue I'm trying to sort
opal compass
they sound fine.
copper marlin
slightly foreign is a no-no as well
copper marlin
they speak normal
are you willing to share a small sample or nah?
opal compass
i can
copper marlin
chat idk
opal compass
i don't wanna send it to the wrong room
opal compass
i will dm it to you
copper marlin
ok
opal compass
just to be safe
copper marlin
okedoke
bleak nymph
does anyone know how i can fix this error
anyone?
wish there were 2 separate rvc help channels, one for training one for inferencing
would make things so much easier to get the help you need
bleak nymph
ok
last time when the mod helped me it was late for me and probs normal time for them
so now im just alone
left oasis
has anyone solved the problem with 504? or do you just have to wait?
opal compass
not sure
rare gobletBOT
Ayo? @opal compass level 2 !!!
opal compass
the error is still here for me
opal compass
same
i made two videos last night
and now the server is doing the error thing
it's weird
left oasis
too
opal compass
yes two
both were mangle from fnaf
i was going to make more
but then the error happened
so i just gave up and stopped
lyric vine
tame python
Is there no colab in Gradio that has a download button? (Excluding hugging face)
bleak nymph
ughh screw this kit lemonfoot colab thing
cant resume training
tried everything
luckily i only trained it upto 20 epochs last night
-local
azure marshBOT
serene wigeon
only direct ai outputs, no datasets
I'm sorry I do not understand what you mean by that
rare gobletBOT
Ayo? @serene wigeon level 3 !!!
serene wigeon
The output of the Google shared code is audio, which is what I posted
copper marlin
Changing memory allocation lets me either use RVC Project at 49152 bytes or use more of Mangio RVC but still fail somehow with 2925 bytes
I don’t actually know if there’s a minimum requirement for either version. It would be cool if someone could tell me.
main tinsel
hello? I wanna download pre-trained hubert-base-korea model at https://huggingface.co/team-lucid/hubert-base-korean
but I don`t know huggingface.. Can I download hubert model here? "pth" file.
brittle wing
should i install tensorboard first before running rvc?
with the tensorboardvenv.bat script
lyric vine
tall wave
half cove
for anyone using colab please check #📰│dev-updates
bitter comet
does anyone know what to do after a mode collapse in Mangio? how will i revert back to the previous epoc save and then continue training? cause that's what i need to do when a epoc collapses right?
past robin
hiii question, the voice models that are posted here that also include the epochs etc in the title, does that mean that we have to train it to that many or that its already trained? is there anything we have to change to make them sound better? (except for the tune ofc)
rare gobletBOT
Ayo? @past robin level 1 !!!
glad zealot
hiii question, the voice models that are posted here that also include the epoch...
its already trained and ready to use
and no, you cant change anything about the model
mint anvil
copper marlin
hello? I wanna download pre-trained hubert-base-korea model at https://huggingfa...
Does that give you a Korean accent in your output voice?
brittle wing
how can i use with amd gpu?
thorny tinsel
Does anyone know what the best model for male-female conversion would be?
main tinsel
Does that give you a Korean accent in your output voice?
I have.. but I don`t know how to use this hubert model..
copper marlin
Damn
half cove
How can I get the public link on HinaMod Colab?
try the top link
u prob using cloudflare i think
young halo
u prob using cloudflare i think
Oooo I just notice, I guess I changed it by mistake
Thank you so much btw
half cove
Oooo I just notice, I guess I changed it by mistake
they just added it today
thats prob why
twilit shore
hitler yok mu
gusty kestrelBOT
I have found 130 results for the search adolf hitler...
Requested by aifeyz_26444
hazy sedge
Hi, can anyone help me to interpret this tensorboard, am I overtraining? I used 4 hours of dataset, currently it has 300 epochs and I think it sounds with more distortion than with 100 epochs. I was saving every 10 epochs, how many epochs would be the model that would give me better quality and less distortion? It is worth noting that you can see a strange line between 20 and 30k (For some reason, I retrained it later and it went from 18300 steps to 47340 steps all at once at the beginning of retraining, I don't know if I did something wrong). Could I retrain from 9000 steps (before the maximum drop and later rise near 10000 steps) to activate the automatic epoch to epoch saving and bring out the best model? I used the colab rvc disconnected v2, thanks in advance.
twilit shore
hitler varmı
turbid pagoda
Would anyone be able to tell me how to use a premade voice on hugging face. Im trying to make a cover and and I can't find the zip file link to copy and import the model into Google colab. I've tried downloading the entire model putting it in my Google drive and copying the Link that way but it still doesn't work.
This is what I'm trying to use
https://huggingface.co/sail-rvc/HankHillv2_e750
spare shale
@half cove amd Good for creating Voices?
half cove
<@936673139419664414> amd Good for creating Voices?
nuh uh
half cove
for training models?
spare shale
Yes, and the creation of covers
half cove
for training models?
you can only do that in linux
and they deleted inference on amd rvc
only realtime now
spare shale
This is not for me
This question was asked by a man
turbid pagoda
I'll settle for a video on YouTube or something on how to get the zip link for a hugging face model if someone has one.
rare gobletBOT
Ayo? @copper marlin level 9 !!!
copper marlin
I assume Rin is actually trained like that
turbid pagoda
Sorry I'm actually not looking to upload a model to hugging face I'm looking to get the shared zip file link of a premade model so I can import and use the model in Google colab.
copper marlin
Sorry I'm actually not looking to upload a model to hugging face I'm looking to ...
you might like this link
I'm currently processing with Rin
hoping for no errors
having a larger dataset definitely helps.
turbid pagoda
Thanks I really appreciate all the help! 😁
half cove
i dont think they mean pretrains but ok
copper marlin
could someone with a cheap computer have made RIN?#
half cove
could someone with a cheap computer have made RIN?#
they used a 4090 and 4080 i think
copper marlin
maybe with 3 months of continuous processing
copper marlin
maybe with 3 months of continuous processing
don't know for sure but it sounds about right considering that the best models are said to take a few weeks, and I assume that's for dedicated computers
loads of source material etc
ionic canopy
Does this mean the batch conversion failed?
Or is it still going? I havent used the inferring before.
copper marlin
Does this mean the batch conversion failed?
are you online?
offline, another app might be using that port
my RVC project install cannot run when another AI uses the GPU
left otter
guys whats the best settings for a RX 580 its keep make robotic voice
ionic canopy
I do local stuff
Oh. I thought you meant am i connected to wifi
Yes i am running locally.
copper marlin
huh?
ah good
well, make sure no other apps are up
aside from that i don't think i've seen that exact error
ionic canopy
There arent any. The web ui and the console are the only things up
The batch conversion was about 4,000 4 second files.
copper marlin
there's 2 things that screwed it up for me: 1) other apps 2) not enough allocated memory 3) the version of RVC
I couldn't get Mangio to work (yet)
RVC Project was quick and easy, one step download and then clicking the bat file
Mangio only half worked when I reallocated memory, but then didn't generate .pth files
ionic canopy
Guess I better go for the other half
copper marlin
vividd
ionic canopy
All those lines.
copper marlin
All those lines.
never heard them
good luck though
still waiting for my model to finish training
copper marlin
.
if you're training from 0, you need more source files, and to train it for more epochs.
ionic canopy
Wait.
copper marlin
more source material is the best thing
ionic canopy
Scratch that number. That was for a different voice.
copper marlin
I had 1) static and 2) a robot voice, after increasing the voice wav source files
and also epochs
ionic canopy
The narrator has 7,432 files
copper marlin
this is for having no pretrained model there
it gets more real with more data
less gaps in the audio
and no static in the recording
background static ruins it
and other noise too
you need clean files and many of them
bleak nymph
I do local stuff
can you help me with which local rvc v2 is best for training models?
azure marshBOT
RVC Guides (How to Make AI Cover)
proper shale
<@622856015444049937> what do you say?
probably
vital sphinx
why does rvc not want to use my gpu. I have a 6950xt and have the direct ml version installed
rare gobletBOT
Ayo? @vital sphinx level 1 !!!
proper shale
why does rvc not want to use my gpu. I have a 6950xt and have the direct ml vers...
if you have converted to onnx there ain't much you can do
W-Okada DirectML version is just like that ig
haughty dawn
How do I know how many epochs a model I made a long time ago has?
copper marlin
TIL I can train from a model that I trained instead of using the default D and G
leaden pond
so, I'm trying to use Applio for text to speech but I'm having a issue with TTS Voices feature In order for me to generate the Audio I need to select a voice from TTS Voices feature I want this to be accurate as possible but this feature doesn't let me, how exactly I'm supposed to get the best results ? also remember this is my first time using this
cloud jewel
Hi there, I need help with my ai cover for a song Invisible made by Duran Duran using Venom Snake voice model
rare gobletBOT
Ayo? @cloud jewel level 1 !!!
cloud jewel
In short the vocal covered sounds like Venom Snake is heavy smoker and it sounds very harshly
copper marlin
what is the difference between V1 and V2?
for me, i get a more chinese accent from V2 rather than V1
sand iris
what is the difference between V1 and V2?
okay, so RVC uses a model called "HuBERT" to convert raw audio to a vector of numbers. Those numbers represent the speech data in many dimensions, and the model tries to cluster the same sounds from different speakers at about the same place, so that if you or I just say "ah", it'll treat it as the same, and allow the model to generalize to all kinds of speakers
what v2 changed is that instead of taking a vector of length 256 from hubert, it now takes one that has 768 numbers
so more data to work with
copper marlin
ok. the audio quality seemed lower in V1 but the accent stronger in v2
is this correct?
more data from the original dataset? hence more of an accent?
dawn nebula
you can crank the index rate (the thing goes up and down) for the accent
copper marlin
you can crank the index rate (the thing goes up and down) for the accent
it's kinda faulty but i can try it
dawn nebula
it's kinda faulty but i can try it
works well for me
left oasis
ohhhh niceee worked
dawn nebula
it's kinda faulty but i can try it
which persons model are you training lol
copper marlin
it works well, but still slightly different.
it's the same effect but quieter or more subtle
dawn nebula
does anyone have an english hubert model?
theres no english hubert gg
copper marlin
I probably need 30 mins source files minimum for a voice like mine
it gets the tone correct before the accent
the accent goes away with more training
i was lazy and did like, 2-3
sand iris
theres no english hubert gg
it's trained on that apparently, idk the lang of the source
copper marlin
ok
thin stump
@ripe ivy I had to raise my headset volume to 40-60 to hear this
chrome tide
everytime I try opening "go-realtime-gui", it pops up with this
fuuuck
cant send images
C:\Retrieval-based-Voice-Conversion-WebUI-main>runtime\python.exe gui_v1.py>
The system cannot find the path specified.
C:\Retrieval-based-Voice-Conversion-WebUI-main>pause
Press any key to continue . . .
reef bone
Hey
Can someone explain this ? :
runtime\python.exe trainset_preprocess_pipeline_print.py C:\Users\italia\Documents\Voix 48000 14 C:\Users\italia\Documents\RVC\Mangio-RVC-v23.7.0/logs/mi-test False
Traceback (most recent call last):
File "C:\Users\italia\Documents\RVC\Mangio-RVC-v23.7.0\trainset_preprocess_pipeline_print.py", line 8, in <module>
sr = int(sys.argv[2])
ValueError: invalid literal for int() with base 10: 'italia\Documents\Voix'
sand iris
Can someone explain this ? :
runtime\python.exe trainset_preprocess_pipeline_pr...
space in file names
your "Voix[...]" folder has a space which makes it freak out
reef bone
there's no space
sand iris
kinda ? Been a long while since I messed with this
but if you extract somewhere outside of your user folder
and somehow don't have any permission-related issues, it'll work
reef bone
anf if I try to create another user ?
rare gobletBOT
Ayo? @reef bone level 1 !!!
sand iris
should also work but that means moving stuff over
chrome tide
space in file names
you know how to do this?
C:\Retrieval-based-Voice-Conversion-WebUI-main>runtime\python.exe gui_v1.py>
The system cannot find the path specified.
C:\Retrieval-based-Voice-Conversion-WebUI-main>pause
Press any key to continue . . .
rare gobletBOT
Ayo? @chrome tide level 1 !!!
sand iris
where and how did you download it
chrome tide
i downloaded these
visual studio 2022 community
python 3.10
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI
chrome tide
github, real time
sand iris
yeah that's the problem. It's unless you setup everything manually, and dl via github
or download a "fat archive" which doesn't require anything else to be installed
chrome tide
you know where i can download the 'fat archive'?
sand iris
fat archives are available on their huggingface : https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/RVC1006Nvidia.7z (latest nvidia release)
sand iris
it also does, but I wouldn't really recommend it
compared to the "actual" voice-changer
chrome tide
ah, alright
what's the difference between the real time one and actual
is the actual one just for sound files, or can u talk with it?
was planning to use shit like a grunt voice changer, for HL2RP servers and such since FL studios ones are dogshit
sand iris
that's what I would consider the "real" one : https://github.com/w-okada/voice-changer/blob/master/README_en.md
Uses the same tech, but only does realtime playback
RVC does file conversion and model training, and also has a realtime module
dawn nebula
1