Ayo? @scenic gull level 5 !!! 
#✨│ai-help
rare gobletBOT
scenic gull
should i download these files and switch them with the original pretrain in my mangio folder?
low shard
mm korean is not close to english tbh, as of now there is no pretrain with both of them
i think you should try to separate the dataset in 2 parts, one english trained using titan and one korean trained using KLMV7
Or if you want, before doing that you could test to train the whole dataset using either titan or KLMV7 but i that may interfere on one of the language of the model
mmm mangio isnt really updated from like summer, applio is better
low shard
should i download these files and switch them with the original pretrain in my m...
in the pretrained folder of that hugging face repository of titan, you should download the g & d files and put them into the pretrained_v2 folder in mangio
As i said mangio isnt very stable anymore tho, if u want applio made using custom pretrains even easier
scenic gull
wispy burrow
mmm mangio isnt really updated from like summer, applio is better
Eh...I wouldn't it say it's better...
scenic gull
in the pretrained folder of that hugging face repository of titan, you should do...
is applio v3.1.1.exe the one I should be getting?
low shard
Eh...I wouldn't it say it's better...
the thing im saying is that mangio rvc isn't updated anymore
brittle wing
-colab
low shard
is applio v3.1.1.exe the one I should be getting?
yea 3.1.1 is the latest version
scenic gull
which pretrain did you use? looks like youre korean too lol
low shard
how did the model come out?
scenic gull
yea 3.1.1 is the latest version
thanks for all your help, keep up the good work!
low shard
thanks for all your help, keep up the good work!
yw
mmm, honestly never trained a model that speaks multiple languages in one, will search abt that to be sure abt it
scenic gull
yo checked out your youtube and your stuff is good, any tips?
wispy burrow
But still didn't pass QC to get Model Maker?
scenic gull
-guides
clear dune
yea 3.1.1 is the latest version
I'm having this very strange bug regarding Applio
it's this
"Your local 3.1.0 version is older than 3.1.1 the latest version"
I wonder what's causing it because I downloaded the latest update
azure marshBOT
RVC Guides (How to Make AI Cover)
cursive scaffold
marble forum
Is It okay to train a model with an MP3 file? Does that affect anything in the course of training or what?
violet heron
Is It okay to train a model with an MP3 file? Does that affect anything in the ...
It’s better to do it in .wav since it’s loseless
But you can do it with mp3
low shard
"Your local 3.1.0 version is older than 3.1.1 the latest version"
I honestly don't know sorry
halcyon oriole
I don't know how to do it, the site is very complicated, can you help me?
violet heron
I don't know how to do it, the site is very complicated, can you help me?
What site?
!howtoask
patent trellis
halcyon oriole
What site?
rare gobletBOT
Ayo? @halcyon oriole level 1 !!!
halcyon oriole
I saw this site from a video. Does this belong to you?
violet heron
https://colab.research.google.com/drive/1Gj6UTf2gicndUW_tVheVhTXIIYpFTYc7#scroll...
Outdated
-colab
low shard
https://colab.research.google.com/drive/1Gj6UTf2gicndUW_tVheVhTXIIYpFTYc7#scroll...
outdated, check our docs https://docs.aihub.wtf
wide parcel
im on applio and when i finish traning it doesnt save a .pth file
any reason for this?
now errors thrown or anything
empty
tawdry shell
My speaker audio keeps bleeding into the RVC AI, dunno why and I can't stop it from glitching as well- any tips/advice?
wide parcel
no
wide parcel
My speaker audio keeps bleeding into the RVC AI, dunno why and I can't stop it f...
noise suppression
i used save only latest instead
tawdry shell
noise suppression
The problem with that is it cuts off my laughs or non-verbal sounds
wide parcel
The problem with that is it cuts off my laughs or non-verbal sounds
if youre running your mic through rvc generally you want that
tawdry shell
Also what do you mean "noise suppression" I don't know if that appears in the RVC UI?
wide parcel
no dice
wide parcel
Also what do you mean "noise suppression" I don't know if that appears in the RV...
are u using webui?
or a live translation
like okada or sum
wide parcel
yeah you should still have it
somewhere around the f0 settings iirc
its been a while tho
tawdry shell
What do I press to turn it on??
wide parcel
there should be a checkbox
tawdry shell
somewhere around the f0 settings iirc
What the heck is an f0?
wide parcel
What the heck is an f0?
zawg
tawdry shell
????
jovial crystal
hi ican only hear the sound in the beatric model i cant hear it in other models
how can i fix it ?
wide parcel
hi ican only hear the sound in the beatric model i cant hear it in other models
what is "the sound"
no pero tengo un traductor automático
jovial crystal
beatric model
jovial crystal
what is "the sound"
only the “beatrice” model recognizes my voice
i cant hear other voices
wide parcel
no hay errores, no tengo el modo de depuración activado ni nada por lo que no muestra ningún resultado
wide parcel
i cant hear other voices
not being to hear any output at all is a more fundamental problem
i can only really help if the audio sounds bad in the first place
^
jovial crystal
if i uploading a new sound doesnt help i can hear it again only beatrice model working
wide parcel
No, lo dejo abierto, sin embargo, no veo errores porque no he habilitado el registro de depuración o similar.
tawdry shell
Is there a way for me to run RVC on the cloud? I only realize now that I've been completely overworking my laptop running it simultaneously with a game so I'm wondering if there is any other way I can execute it without sending my computer into a peaking meltdown
wide parcel
Is there a way for me to run RVC on the cloud? I only realize now that I've been...
yep
rtfm 🙃
azure marshBOT
Suggestions for @tawdry shell
tawdry shell
read the fucking manual
... geez. I was just starting out so I decided to ask. Nevermind.
tawdry shell
also go to <#1159290161683767298> w this
Again, I don't use okada.
wide parcel
... geez. I was just starting out so I decided to ask. Nevermind.
its what rtfm stands for mb
wide parcel
Again, I don't use okada.
you use something that fundamentally works like okada
i did
violet cape
wide parcel
creo que lo resolví
Por alguna razón, cuando lo activas para detectar sobreentrenamiento, no guarda la versión final.
no tengo idea de por qué
ohhh hfuck
bright moat
how do i use the voice model after downloading
rare gobletBOT
Ayo? @bright moat level 1 !!!
low shard
rvc gui is outdated, its better u check our docs https://docs.aihub.wtf
violet heron
RVC-GUI is very outdated
violet heron
-local
azure marshBOT
Suggestions for @brittle wing
violet heron
Original RVC or Applio
thin stump
What am I doing wrong here? I get a runtime error on tensorboard and list index out of range error
azure marshBOT
RVC Guides (How to Make AI Cover)
graceful obsidian
Alright, finished:
https://github.com/codename0og/rvc-realtime-voice-changer/tree/main
That's for anyone wanting rvc realtime. ( light-weight, standalone and simplified installation / running. enjoy ~ ✨
@red kayak made it easier for every1 to get it up n running, check it out if you want.
I gotta go to sleep so ~ til then
azure marshBOT
🎙️ Realtime Voice Changer (Online)
ruby zinc
paperspace, colab or GTX 1660 super (overclocked)? training
mystic anvil
keep getting
ModuleNotFoundError
on step 2 of the rvc v2 website when trying to download an ai voice
can someone help me
ruby zinc
What colab/hf are you using?
rare gobletBOT
Ayo? @mystic anvil level 1 !!!
azure marshBOT
Suggestions for @mystic anvil
ruby zinc
-colab
mystic anvil
ok
ruby zinc
ilaria rvc is the easiest one for beginners iirc
rare gobletBOT
Ayo? @ruby zinc level 2 !!!
ruby zinc
yes
ruby zinc
do you have the model?
mystic anvil
idk how to use htis
ruby zinc
ok
go to the download section
this one
you can either put the pth file in there if you alr downloaded it
here
or you could search it in the bottom thing if you don't have it
or if you have the link to it, you can use the one at the btop
mystic anvil
ruby zinc
here
this one
one at a time
light
you good with training?
if so, should i go local (gtx 1660 super OC) or colab?
mystic anvil
its taking forever to start tho
ruby zinc
its taking forever to start tho
wdym
mystic anvil
ruby zinc
go to download
mystic anvil
at download
mystic anvil
yes
ruby zinc
but then i forgot the subscription on after i stopped using it (I never put money on my card)
so now they hate me
for giving them a bunch of failed charges
mystic anvil
dragging the pth and index file
rare gobletBOT
Ayo? @mystic anvil level 2 !!!
mystic anvil
goes in there
ruby zinc
what
violet heron
@mystic anvil what are you trying to do?
ruby zinc
that "no webui" part kinda concerns me
ruby zinc
your wifi
ruby zinc
oh
rare gobletBOT
Ayo? @ruby zinc level 3 !!!
violet heron
400 download and 400 upload
If the model has a HuggingFace download link you should be able to use that instead of drag and drop
violet heron
bro ive been here for less than an hour and im about to unlock <#115926530091568...
I’m level 97
Started being active since like June
ruby zinc
these roles are pretty accurate
mystic anvil
If the model has a HuggingFace download link you should be able to use that inst...
with appolio?
violet heron
with appolio?
Think so
mystic anvil
can i use the rvc models archive sheet
violet heron
can i use the rvc models archive sheet
yeah but #1175430844685484042 has more models
ruby zinc
i used to be in here but im switching accounts for personal reasons
mystic anvil
do i need any output information
ruby zinc
that's why i know so much already
ruby zinc
i'd prefer not say who my old acc is but it was barely level 5
mystic anvil
why cant i use gradio
ruby zinc
the last colab
ruby zinc
imma go sleep now
ruby zinc
have fun with whatever you're trying to do
mystic anvil
@violet heron
ruby zinc
mystic anvil
mystic anvil
where is model list
violet heron
mystic anvil
this
is the ultimate loop around
i thought i got a huggingface link to avoid putting in
pth and index files
since they are so slow to download
yeah its insanely slow
oh wait
im a bot
the voice model and index file are aleady selected
rare gobletBOT
Ayo? @brittle wing level 2 !!!
mystic anvil
@brittle wing?
violet heron
On?
marble forum
Day 2 of asking for a cowboy western accent model recommendation. Help please!
livid gazelle
how do I download the final product? (I used the bot)
placid holly
what does the exe option do? is it run and go, or it install itself (extract)??
I'm usually use the zip one, it just extract run and go
ah i see, simplify zip version
thank you
violet heron
Download section
Put in the model files
Or if you have a download link put that in
rugged breach
I like to merge models to get a unique voice, but the models have to be the same rvc version and sample rate, etc. Is there any way to convert a voice model, or are we stuck with whatever the person who made it happened to use?
scenic gull
are you using a huggingface cell
this is 0.987 smoothing, when should I stop training?
proud elbow
beware that it might be mode collapses. besides that, try smoothing 0 and ignore outliers off then see.
scenic gull
wym
steel forge
Looking to make a really good model using Codename's merge guide
i assume i should probably retrain this
all other metrics looked great
Now that I pay attention, all my previous logs are having rising FM values
crimson basalt
Anyone know how to fix these errors? Tried everything and still getting them.
steel forge
Anyone know how to fix these errors? Tried everything and still getting them.
sounds simple but have you tried downloading/unzipping the mmvc folder again? Basically reinstall
I searched the search bar and that's a mysterious error
I see. I'm on applio though
crimson basalt
sounds simple but have you tried downloading/unzipping the mmvc folder again? Ba...
Yep I've reinstalled it loads of times.
steel forge
I'll take their advice with this and not worry too much
tried to link message but copied ID
its in staff chat anyway but ill repost here since it's relevant
gusty ginkgo
what does this error mean (Hina_Mod_AICoverGen_colab.ipynb)
gradio.exceptions.Error: 'index -1 is out of bounds for axis 0 with size 0'
generating an ai cover
I made the model myself
autumn jacinth
Can anyone provide me the updated RVC disconnected Google colab notebook in which Titan pretrain and Ov2.1 pretrain is included 🥺
azure marshBOT
RVC Guides (How to Make AI Cover)
gusty ginkgo
I closed the web ui tab and shut down the runtime already but all I had changed was the pitch in the web ui
and yes while infering
wispy burrow
Now that I pay attention, all my previous logs are having rising FM values
Same here.
fm is rising with all my logs too.
inner cloak
It's been a while for me since I have tried making a model. The colab for Applio doesn't work correctly when downloading custom pretrained models. Gives an error in colab. Anyone have a fix/better colab? I am uploading the titan model over manually.
Alright. I'll try that one instead
rare gobletBOT
Ayo? @north imp level 1 !!!
inner cloak
Does Applio need the dataset in a zip or can you just upload them to the drive root? I have them in a zip rn in rvcDisconnected folder.
gusty ginkgo
The vocal pitch
azure marshBOT
RVC Guides (How to Make AI Cover)
mystic marsh
-colab
gusty ginkgo
-6
tight kindle
Looking to make a really good model using Codename's merge guide
what guide is this, and where can I find it?
proper shale
i think it's on pins?
verbal kestrel
Do anyone know how to fix “index file is not found”
pastel oak
Do anyone know how to fix “index file is not found”
are you using rvc voice changer? thats not an error dw
or u using something else
verbal kestrel
are you using rvc voice changer? thats not an error dw
Yeah I use voice changer
verbal kestrel
Ok
pastel oak
Idk if that is an error message though, just means you didnt upload an index
verbal kestrel
Idk if that is an error message though, just means you didnt upload an index
Then how to upload it
violet heron
Then how to upload it
Do you have a .index file?
A lot of people don’t use it for the voice changer tho
verbal kestrel
Do you have a .index file?
No I don’t have it
verbal kestrel
A lot of people don’t use it for the voice changer tho
How to get it?
violet heron
How to get it?
It should of came with the model
verbal kestrel
Wait let me see
verbal kestrel
It should of came with the model
I can’t find it
verbal kestrel
It should of came with the model
Are they anyway to fix it?
pastel oak
Are they anyway to fix it?
if you dont have it you cant do anything, its not the end of the world
people dont use it for realtime anyway
very specific uses
steel forge
@graceful obsidian i know theres been a lot of discussion about this lately, tried making sense of this using the search bar here. Do we know if Applio also potentially reverses FM value? No matter the dataset, it always goes up for me
graceful obsidian
<@1073140391035011183> i know theres been a lot of discussion about this lately,...
if it goes up as in, spikes are with upward tendency and not downward ( or generally up, no matter which model you train )
then yes
That'd mean it's reversed
Tho, imma check the code n do some comparison
pastel oak
my fm always has an upwards trajectory also on mainline so i assumed its normal anyway
graceful obsidian
btw
you guys want my mel similarity thingy?
basically, it calculates the similarity of model's generated Specs and ground truth specs
graceful obsidian
my fm always has an upwards trajectory also on mainline so i assumed its normal ...
well, it shouldn't be
the loss is always meant to go lower
on the landscape
so it's def abnormal in terms of presentation ( and can be confusing for some )
ah, I think I found the issue
hmm, or maybe not
that's hella weird
losses implementation seems aligned with mine
same for the training loop, loss calc and tensorboard logging
🤔
This shouldn't be happening really
wispy burrow
so it's def abnormal in terms of presentation ( and can be confusing for some )
Is it only visual bug or does it affect our models too?
graceful obsidian
Visual.
Trust me, if it was the literal bug, the models would sound truly trash
as it'd be an exact opposite of a 'positive' type of loss for fm
but anyway, if the code more or less aligns, maybe it's tensorboard related bug?
@wispy burrow can you send me any of tfevents to test? ( where reversed fm occurs )
Gotta see if it's the tensorboard or the loss logging / calculating itself
@steel forge so overal, I did some comparison - everything seems fine
can you send me any abnormal tfevents?
Cause while I'm at it, can test stuff
steel forge
I can share some, what do you mean by abnormal?
graceful obsidian
those in which you observe the 'reversal' behavior
steel forge
gotcha. DMs in a few moments
graceful obsidian
alr
alright, got the results
so, basically @steel forge @wispy burrow @pastel oak
It's just visuals. If it's not for the bad / not the best convergence that is
if you observe rest of the stats doing okay or fine, more or less downward with proper convergence
yet fm goes up ( but like, legit goes up and not fluctuates ), that's just some visual bug
crude flame
my fm always has an upwards trajectory also on mainline so i assumed its normal ...
It does always go up on mainline
graceful obsidian
cause like, I the tfevents provided also took a closer look at some of my older weaker models and ye, that's normal
if the convergence sucks, it'll be fluctuating with some down-tendency points
feature matching and kl are more complex elements for the model to tame, compared to just 'clean ish' spectrogram replication
there's way more to it, in fm, way way more to it. So ye, it's normal and fine
here's an example of my Kurisu training ( long time ago )
and that was the tendency
steel forge
So, the highest point on the fm chart is the best step (ignoring collapses and such)
just making sure im understanding, because i want to try your merge guide out
graceful obsidian
nop, you still follow the loss values
unless your convergence goes really well or your training " quality " in general, it's kinda harder to follow the tendency
alternatively, you could try to enable vertical scaling ( middle icon-button under the graphs ) enable 'ignore outliers' and set it to hmmm
sacred tendon
p
graceful obsidian
stuff like 0.990, 0.985 or perhaps 0.885
( and resize ofc ~ last icon-button
grave void
stuff like 0.990, 0.985 or perhaps 0.885
( and resize ofc ~ last icon-butt...
If i wanted to get into model making and AI stuff what would be a good place to start?
graceful obsidian
lots of info lmao
in a short, if you can't observe a clear tendency due to fluctuations or poor convergence, focus on the loss values
fm in most "worse training" cases, will fluctuate and eventually go up anyways
graceful obsidian
If i wanted to get into model making and AI stuff what would be a good place to ...
As in?
If it's about voice cloning, you have RVC
graceful obsidian
yea, that'd be RVC
grave void
Alright good to know
graceful obsidian
you have few forks of it ( a modified repository we call fork )
along with original rvc's repository ( so, original rvc )
azure marshBOT
RVC Guides (How to Make AI Cover)
graceful obsidian
and applio ( perhaps most famous fork aside an already outdated mangio-fork of rvc )
https://github.com/IAHispano/Applio
Instructions for installation are on repo but you can also find a lot of valuable stuff in here, on the server
low bolt
how can i make the voice changer work on my mic bc it only works on the app and in discord my friends can hear only my normal voice while in the app my voice is modified
violet heron
how can i make the voice changer work on my mic bc it only works on the app and ...
Did you download the audio cable
formal stream
hey,could anyone explain to me how to make models for the RVC GUI? I just downloaded it and thought it would also be able to create models but i guess I was wrong, can anyone help me please?
violet heron
hey,could anyone explain to me how to make models for the RVC GUI? I just downlo...
Don’t use RVC-GUI it’s very outdated
-local
azure marshBOT
violet heron
Use original RVC or Applio instead (applio has a installer)
low bolt
Did you download the audio cable
how can i do that
violet heron
how can i do that
Download and run setup64
After setting up set the output on the voice changer to line 1
And on discord set the input to it
formal stream
Use original RVC or Applio instead (applio has a installer)
well...can any of those create models?
formal stream
good question, shame i dont know
rare gobletBOT
Ayo? @formal stream level 1 !!!
violet heron
good question, shame i dont know
Task manager - performance - GPU 1
If GPU 1 isn’t there then GPU 0
formal stream
Radeon RX550
violet heron
Radeon RX550
AMD GPU, can’t train locally
You would have to use a google colab
-colab
violet heron
RVC Disconnected (it has no ui), Applio, or Ilaria RVC for training
Make sure your clicking the colab version
formal stream
you commented on Applio before, is it good?
violet heron
you commented on Applio before, is it good?
It had its issues in the past but it has gotten better
It’s good for beginners if you can get it up and running
For the google colab just run each cell one by one
And open a link if it asks you to
Don’t need to install anything
formal stream
this google colab thing is like,its actually running in another computer correct?
violet heron
this google colab thing is like,its actually running in another computer correct...
Yes
formal stream
well...thanks,now remaking the first question
how do i make de voice models in that Applio from Google Colab?
wide parcel
ive probably asked this a few times in the past
whats the rough max of sample time before the quality plateaus?
median monolith
i see, thanks for the clarifications. i also will suppose there is no way to directly download a youtube video audio as a wav.
tacit jetty
How can i use the rvc models that i downloaded
rare gobletBOT
Ayo? @tacit jetty level 5 !!!
balmy prawn
Hi, is there a forum/topic with settings recommendations ? I've been mix n' matching a bit, but sometimes the outcome is so different from what I was expecting it almost feels random (Or as if I wasn't in-control) :
graceful obsidian
@balmy prawn
protect voiceless - best to be kept at 0.3 or 0.33 ( default ) - too lengthy for me to explain, kinda lazy
resample the output - don't touch it. You want it to be always set to '0'
transpose is just by how much you want to pitch either up or down the final ( model generated ) output ( measured in semitones);
12 ( positive ) and -12 ( negative ) = 1 octave up and down respectively (( ps, for positive do not input '+' just an integer value ))
- feature index is, in a short:
How much of your model's features ( accent, pronunciation, style-ish, phonemes handling etc. ) you wanna use for the generation.
The deal with that is, if the model was trained poorly ( didn't adapt well to handle own features ) or on a small dataset ( audio samples used to make the model ), it might not find anything 'fitting' for the input ( acapella/vocals/speech ) audio you'd infer ( use ) your model on, which could result in some audio-artifacts, model ' struggling ' and doing scratchy sounds / noisy sounds or such.
A good example for you to visualize it is:
Imagine some part of your input vocals or speech has a word "world", now, if you were to set the index feature ratio at 0
( meaning, no model's own features are used )
it will take the reference of how to pronounce or articulate phonemes contained within the word "world".
However, if you set the index feature ratio to 100, it'd take the phoneme handling
( for example; way of articulating / pronunciation of: 'l' or 'r' or 'd' ) from your model's features
a good values set I always recommend to try ( and experiment! cause why not? ) are: 0.3, 0.5, 0.75 and 0.85 ( remember. Better the model, higher you can go with index ratio )
- Now, the "volume envelope" is kind of rms-normalization / rms-scaling.
in other and simple words; if you set it to 0 = model will reference dynamic range of the input audio ( f.e.: vocals )
if you set it to the max = model will ( more or less ) output quite dynamic-range-castrated output, kinda steady in volume, ish
my recommendation is to use it at '0' or if you really have to? 0.1 or 0.20? maybe 0.25
sterile tiger
how do i change language on gpt so vits
formal stream
how do i make de voice models in that Applio from Google Colab?
hey can anyone help me with this? or say on wich chat I should be asking this?
violet heron
How can i use the rvc models that i downloaded
RVC
violet heron
hey can anyone help me with this? or say on wich chat I should be asking this?
Train
rare gobletBOT
Ayo? @formal stream level 2 !!!
formal stream
sorry
violet heron
there is no train tab or button dude
Your using the HuggingFace one not colab
indigo garden
is there a way to make ai covers with an rcv model, but changing the lyrics and keeping the same flow of the original vocals?
or are there any other alternatives?
haughty pulsar
how do i change language on gpt so vits
for right now there are only english japanese and chinese
indigo garden
is there a way to make ai covers with an rcv model, but changing the lyrics and ...
?
hallow kindle
how can i make the rvc less stuttery
violet heron
how can i make the rvc less stuttery
Use RVMPE
hallow kindle
ive only got rvmpe_onnx
rare gobletBOT
Ayo? @mystic anvil level 3 !!!
mystic anvil
its literally at 510 seconds now
mystic anvil
can i get somebodys help
am trying to convert a hugging face model to a vocals audio ive added myself
but its literally been over 10 mins and this hasnt finished downloading
low shard
are you using the huggingface space? it uses cpu so its way slower
wide parcel
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1248 | step=26208 | time=18:06:08 | training_speed=0:00:13 | lowest_value=2.8077754974365234 (epoch 568 and step 11909)
50
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1249 | step=26229 | time=18:06:21 | training_speed=0:00:13 | lowest_value=2.8077754974365234 (epoch 568 and step 11909)
Saved model 'F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new\G_26250.pth' (epoch 1250)
Saved model 'F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new\D_26250.pth' (epoch 1250)
Saved model 'F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new_1250e_26250s.pth' (epoch 1250 and step 26250)
50
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1250 | step=26250 | time=18:06:44 | training_speed=0:00:22 | lowest_value=2.8077754974365234 (epoch 568 and step 11909)
50
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1251 | step=26271 | time=18:06:57 | training_speed=0:00:12 | lowest_value=2.8077754974365234 (epoch 568 and step 11909)
50
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1252 | step=26292 | time=18:07:10 | training_speed=0:00:12 | lowest_value=2.8077754974365234 (epoch 568 and step 11909)
50
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1253 | step=26313 | time=18:07:22 | training_speed=0:00:12 | lowest_value=2.8077754974365234 (epoch 568 and step 11909)
50
F:\applio\NEWEST\ApplioV3.1.1\logs\kevy_ellevy_new | epoch=1254 | step=26334 | time=
seems to be freezing or something similar?
inner cloak
how do I upload the dataset to colab applio? Does it need to be in a zip or do I just make a folder and tell it where the wav files are? Edit: Guess I'll just be ignored and figured it out 👍
also the extra pretrains section errors out.
mystic anvil
@low shardso use gpu?
low shard
<@911742715019001897>so use gpu?
Yes, you can use the applio google colab that uses free daily T4 Gpu
Its way faster
You can find it in our docs
azure marshBOT
RVC Guides (How to Make AI Cover)
inner cloak
Thank you! Now who do I straggle for messing this up? 🤔 (JK)
yeah, titan is what I am trying to use. I've tried manually uploading it to colab but it kicks me out because it takes too long. Thanks again tho.
heady grove
night moss
I'm having trouble training models, any models I create sound kinda wobbly I guess.
I've attached one of my models with said effect.
inner cloak
Anyone else experiencing storage issues on collab? I've got like 8gb left and I have only been training for an hour in a half so far. Anything I can do to prevent it from running out of space?
nevermind realized it was the g and d files.
oblique venture
are there any alternative for mvsep to clean audio? both colab or another website is ok. The mvsep website for some reason has a crappy download speed (every other web download speed is normal)
oblique venture
Anyone else experiencing storage issues on collab? I've got like 8gb left and I ...
you can mount your drive and change your model saving path to there, so when the colab crash for whatever reason, you can still continue later since your progress saved in drive
mighty owl
night moss
I'm having trouble training models, any models I create sound kinda wobbly I guess.
I've attached one of my models with said effect.
inner cloak
you can mount your drive and change your model saving path to there, so when the...
I'll check it out then.
rapid patrol
Does anyone know why my input is in output and my output is in input ?
I had to reinstall the Real time Voice changer, and then out of a sudden it didn't work anymore and now my output and input are swapped around
proud elbow
nevermind realized it was the g and d files.
keep "save only latest" ticked
mortal bolt
Does anybody know any tips on how to make a good voice model?
rare gobletBOT
Ayo? @mortal bolt level 1 !!!
stray adder
Somebody can tell me what's wrong?
wispy burrow
Somebody can tell me what's wrong?
That last line usually means it's an outdated Colab.
And I think your Model link is wrong.
Model links from Huggingface normally ends with ?download=true
stray adder
That last line usually means it's an outdated Colab.
And I think your Model link...
"And I think your Model link is wrong.
Model links from Huggingface normally ends with ?download=true"
Yes, no, I mean that link is by default, I tried with mine couple of times and nothing so I thought to use that one.
"That last line usually means it's an outdated Colab."
How can I fix it?
wispy burrow
"And I think your Model link is wrong.
Model links from Huggingface normally end...
I'm not too familiar with Colab (never used it).
But if it's outdated you need another one.
wispy burrow
"And I think your Model link is wrong.
Model links from Huggingface normally end...
-colab
azure marshBOT
RVC Guides (How to Make AI Cover)
thin stump
@frozen trout Try to keep it like this . And at least make the pth file name make sense so people know what's the model name
So instead of "Twirus" put "TwilightRus"
rare gobletBOT
Ayo? @thin stump level 25 !!!
frozen trout
So instead of "Twirus" put "TwilightRus"
okay, thx for help
i'll try
thin stump
okay, thx for help
i'll try
Yw, here's the full file name for reference
I'm unsure if he'll make you redo it but it should be solved
frozen trout
Yw, here's the full file name for reference
added_IVF299_Flat_nprobe_1_TwirusV3_v2.index
TwirusV3_e300_s8100.pth
like this?
rare gobletBOT
Ayo? @frozen trout level 1 !!!
frozen trout
not renamed files
thin stump
```added_IVF299_Flat_nprobe_1_TwirusV3_v2.index```
```TwirusV3_e300_s8100.pth```...
Oh, just slightly modify it Twilightrus. He wouldn't notice 
yes
frozen trout
Oh, just slightly modify it Twilightrus. He wouldn't notice <:trolley:1159468147...
hehehe, got it)
thx
thin stump
hehehe, got it)
thx
Oh, i didn't read his feedback completely but you can keep "Twirus" the same. Sorry for overcomplicating it
you just have to resubmit it now
frozen trout
Oh, i didn't read his feedback completely but you can keep "Twirus" the same. So...
hmm.. keep twirus, but full name of index and pth files?
thin stump
hmm.. keep twirus, but full name of index and pth files?
Yes, so when you do submit it on #1175430844685484042, that information gets auto-filled on weights.gg. It's best to keep the full name of the files
winged sundial
yo, RVC software do yall recommend?
silver sky
yall probably get like tons of messages asking this. its my first time so I apologize. What kind of algorithms should I be using, both for mangio and the RVC GUI.
thin stump
yall probably get like tons of messages asking this. its my first time so I apol...
Do you mean pretrains for model training?
thin stump
yo, RVC software do yall recommend?
For AI covers, I recommend https://docs.aihub.wtf/rvc/cloud/ilaria-rvc/
thin stump
Do you mean pretrains for model training?
and we use RMVPE. Crepe is only if the dataset is extremely clean
silver sky
Do you mean pretrains for model training?
No im talking about the ones like crepe, where you can control how the output will sound
winged sundial
For AI covers, I recommend https://docs.aihub.wtf/rvc/cloud/ilaria-rvc/
not exactly Ai covers, im looking to fuck around with the beats I produce, tryna rap in them, then convert it to the artist who's beat it sounds like type shit
rare gobletBOT
Ayo? @winged sundial level 1 !!!
thin stump
No im talking about the ones like crepe, where you can control how the output wi...
Yeah, results are better if you do inference with RMPVE than Mangio Crepe 80% of time. But also depends on how you vocal isolated the song and the quality of the AI model
thin stump
not exactly Ai covers, im looking to fuck around with the beats I produce, tryna...
It's the same thing. You would need to rap then do inference with the model you're referring to https://docs.aihub.wtf/essentials/how-to-make-ai-cover/ . But if you want to do it in realtime voice changer for some reason, then you can refer to #🔍│help-w-okada
winged sundial
nah realtime is gonna be ass imo
winged sundial
It's the same thing. You would need to rap then do inference with the model you'...
bet, thanks
i used it like months ago, but got a new PC, so i completely forgot what to do
thanks haha
thin stump
Np, I personally use tryreplay.io for local use since i'm too lazy to setup Ilaria rvc every time
winged sundial
how is it?
thin stump
I think there's better options out there like Applio if you have a GPU, but I like it for what it is
ruby zinc
Np, I personally use tryreplay.io for local use since i'm too lazy to setup Ilar...
Is rvcguipkg still good? I used to use that
rare gobletBOT
Ayo? @ruby zinc level 4 !!!
thin stump
I'm being very vague on this since I don't want to disappoint anyone
thin stump
Is rvcguipkg still good? I used to use that
never heard of it
ruby zinc
Oh
winged sundial
I think there's better options out there like Applio if you have a GPU, but I li...
yea I've got a GPU
ruby zinc
https://github.com/Tiger14n/RVC-GUI
This thing
winged sundial
https://github.com/Tiger14n/RVC-GUI
This thing
ahh
yeah
that the old one I used
ruby zinc
It got archived since I last used it
winged sundial
I made this with it, so you can decide how the AI sounds
thin stump
-colab
thin stump
Use easyGUI, that's the updated link
ruby zinc
What should I use if I want to do local inference since the one I used to use is ded?
The normal gradio WebUI?
thin stump
https://docs.aihub.wtf/rvc/local/mainline/ for local use (inferencing), or alternatively you can use tryreplay.io
Applio has training built in specifically for GPUs and good for inferencing https://docs.aihub.wtf/rvc/local/applio/#applio
https://docs.aihub.wtf/rvc/local/mangio/ also this is fine too
rare gobletBOT
Ayo? @brittle wing level 15 !!!
gray crow
Hey, maybe someone knows.
If I create a data set with different audio files of the same singer.
First, what is the best tool to use to determine the sample rate?
Secondly, if I train an initial 48k dataset to 32k, how will that affect the model?
So it doesn't matter what the initial sample rate is as long as I match it to the one I'm using for training?
Or is it better to check over mel spectogram to see what comes closest and change it to the one?
Also thanks for your reply man appreciate it! ❤️
rare gobletBOT
Ayo? @gray crow level 4 !!!
tardy bone
@red kayak OK, I got the histogram to show up, now what
Which graph should I look for?
red kayak
<@460577350900514837> OK, I got the histogram to show up, now what
its a graph..
not a histogram
anyhow
rare gobletBOT
Ayo? @tardy bone level 6 !!!
proud elbow
Which graph should I look for?
g/total and g/mel
proud elbow
I'm not getting any graph data wtf
refresh the whole page and wait the training log more than 1 epoch?
(I mean the tensorboard iframe)
tardy bone
yea
proud elbow
though I have only been using RVC disconnected
red kayak
thats gradient g
tardy bone
I honestly don't understand
I recall there used to be a plot line corresponding to values which I can easily check against
Plus there's an "INACTIVE" thing on the orange bar
tardy bone
g/total and g/mel
loss/g/total?
@red kayak The graph plummeted
should I let it run for a while until it goes back up?
tardy bone
Like that?
Just did what you told me lel
If the graph goes back up, that's where I should stop, right?
@red kayak
What if the graph goes stale tho
red kayak
just let it train
tardy bone
damn train 😛
tardy bone
@red kayak Once the graph shoots up, that's where I hit the brakes?
red kayak
mhm
red kayak
just let it train
u need to be patient
models fluctuate
they go up and down all the time just cuz they feel like it
red kayak
just a random increase dw about it
tardy bone
OK
safe walrus
what's the main reason for a really low loss amount?
some datasets reach a lower loss compared to others, and the ones that get the lowest usually sound the best quality
with similar quality of datasets, which made me think what the reason would be (mayb size?)
red kayak
what's the main reason for a really low loss amount?
every run differs
and every parameter will have different loss amounts
its how ML is really
safe walrus
what parameters u meaning?
red kayak
what parameters u meaning?
batch size, learning rate, F0 and sample rates all play a role
safe walrus
every run differs
i did couple runs on one, but it never seem to get better no matter how long i let it train for
red kayak
- dataset size etc etc
safe walrus
yeah okay that makes sense
red kayak
i did couple runs on one, but it never seem to get better no matter how long i l...
yes if u use the same ones then u'll get similar results
safe walrus
it feels so random
red kayak
it feels so random
cuz it is
its all random
the values given for processing arent exact
instead of like the number 15
the computer has to use 15.2355 smthing smthing in the formulas given to calculate loss and generate stuff
safe walrus
ah okay
rare gobletBOT
Ayo? @safe walrus level 1 !!!
red kayak
unless u somehow managed to nail the exact lenght and everything
safe walrus
yes if u use the same ones then u'll get similar results
u mean i should switch up the dataset or? i mostly always use 16bit 48k wav as datasets so hard to switch that up. well i could try 40k i guess
azure marshBOT
RVC Guides (How to Make AI Cover)
red kayak
u mean i should switch up the dataset or? i mostly always use 16bit 48k wav as d...
depends i need to see if ur audio is actually 48k
u can do so by using free software like spek
safe walrus
yeah i use that applio tool
red kayak
not that reliable compared to the actual thing like spek
safe walrus
shows u the spectogram thingy and then i double it... usuallly its around 22k
tardy bone
not that reliable compared to the actual thing like spek
I accidentally set the max epoch to 300
tardy bone
and the graph dipped at that threshold
safe walrus
u can do so by using free software like spek
ill give that a try too
safe walrus
and the graph dipped at that threshold
dips are a good thing?
tardy bone
dips are a good thing?
dunno
safe walrus
sure, its les loss
tardy bone
safe walrus
u usually wanna go for the lowest loss step
yeah that looks fine
its suppose to dip
tardy bone
at 300 epoch
red kayak
u usually wanna go for the lowest loss step
well it depends yeah
u dont wanna get stuck in local minima tho
safe walrus
true
red kayak
such as mode collapses and such
safe walrus
but u only discover that after viewing it for a while
and his is going straight down still
red kayak
thats why u need to look at raw data and smoothed data to look at the tendencies of all the graphs
but thats a little advanced
check pins
tardy bone
300 epochs, 3000 steps
safe walrus
yeah also the overtraining isnt that worrying in my experience
tardy bone
took me an hour and 39 minutes tops
safe walrus
red kayak
yeah also the overtraining isnt that worrying in my experience
well doesnt kill quality but it takes away generalazation
safe walrus
this is what mine usually look like
tardy hedge
Hi guys, can i find help here ?
red kayak
u ccant really see it with the smoothing u got on
rare gobletBOT
Ayo? @safe walrus level 2 !!!
safe walrus
Hi guys, can i find help here ?
sure
tardy hedge
My rvc doesn't change anything to my voice, it's the same after trying all models
tardy bone
u ccant really see it with the smoothing u got on
So I should be fine with my model then
mossy bison
hello guys for me rvc doesnt work it doesnt detect my voice at all
tardy hedge
hello guys for me rvc doesnt work it doesnt detect my voice at all
check ur in put device
safe walrus
So I should be fine with my model then
yeah, and if u got weights saved per epoch u can just get earlier versions and hear the difference
tardy hedge
select the good mic
rare gobletBOT
Ayo? @tardy hedge level 1 !!!
tardy bone
yeah, and if u got weights saved per epoch u can just get earlier versions and h...
Makes sense
safe walrus
plus u got 10 million graphs
thats true, i like my graphs smooth tho 🤣
mossy bison
check ur in put device
i put my microphone and output as my headphone speaker doesnt work any fix?
red kayak
thats true, i like my graphs smooth tho 🤣
it dont work like that
red kayak
this thing told me
nah nah old
safe walrus
im not that experienced yet to understand the unsmoothed graph i think
tardy hedge
does rvc works on amd ?
red kayak
does rvc works on amd ?
mhm
proper shale
does rvc works on amd ?
meh
tardy hedge
is it a yes ?
red kayak
im not that experienced yet to understand the unsmoothed graph i think
go ahead tell me then
mossy bison
i put my microphone and output as my headphone speaker doesnt work any fix?
so it basicaly for me RVC works only on CPU
and doesnt work on gpu
rare gobletBOT
Ayo? @mossy bison level 1 !!!
proper shale
see, it can't train
tardy bone
@red kayak Check this shit out mang
red kayak
is it a yes ?
applio has support for amd
red kayak
<@460577350900514837> Check this shit out mang
i told u, see how much better it sounds
when u put just a little more work into it
red kayak
just telling u stuff, but u can consider it one sure
rare gobletBOT
Ayo? @tardy bone level 7 !!!
tardy bone
just the 2 of us
hmm
red kayak
would fit her voice
tardy bone
keep training though
well the graph dipped at that point
would this model get rejected if I submit it as is tho
red kayak
dont get lazy put in the work
red kayak
it probably aint done training
red kayak
don't make me get incensed 😢
do that and we'll see what we'll do with ya
tardy bone
do that and we'll see what we'll do with ya
well not really, I'm just a little tired
not that I'd go berserk, mind you
red kayak
grumpy yes ik
tardy bone
yea sorry about that
red kayak
just train more, it has potential so dont just throw the towel in already
tardy bone
If you want I'll do a retrain first thing in the morning
red kayak
If you want I'll do a retrain first thing in the morning
u can continue training later
tardy bone
but let me send in some samples for evaluation
tardy bone
u can continue training later
Can I retrain with just the .pth and .index I got from the session?
red kayak
no u need the whole model folder for u to retrain
tardy bone
shit
red kayak
the pth and index dont contribute to training at all
next pumice
hello im having some issues with rvc not detecting my lappy gtx 1650
tardy bone
hello im having some issues with rvc not detecting my lappy gtx 1650
Does it have two GPUs installed?
next pumice
Does it have two GPUs installed?
if an igpu counts
tardy bone
you'd have to assign the 1650 as the default GPU to python.exe
tardy bone
the pth and index dont contribute to training at all
So it's a complete retrain then
safe walrus
so that it doesnt sound robotic?
robotic output is usually the problem of the model, not the settings
next pumice
do i also have to install python standalone or is it already installed with rvc?
tardy bone
do i also have to install python standalone or is it already installed with rvc?
depends on the RVC you're using
winged sundial
fairs, im tryna find a good eminem one
red kayak
So it's a complete retrain then
i guess so but u can drag the model files into ur gdrive
next pumice
depends on the RVC you're using
which one would you reccomend for gtx 1650
safe walrus
which one would you reccomend for gtx 1650
try applio, it's the latest best one i believe
tardy bone
i guess so but u can drag the model files into ur gdrive
Guess I'll retrain just to be doubly sure
the whole sesh I did lately took about an hour and a half
winged sundial
minimum req for applio is 2060 ti no?
safe walrus
i never knew there was minimum req, it just usually req cuda (rtx) and minimum 6gb vram but i mean u can try it out right
next pumice
MMVCServerSIO_win_onnxgpu-cuda_v.1.5.3.15.zip i will try this one and if it doesnt then i will try apollo
rare gobletBOT
Ayo? @next pumice level 1 !!!
proper shale
minimum req for applio is 2060 ti no?
if you wanna train yeah
minimum would probably be like... 1060
winged sundial
how do i train them btw?
rare gobletBOT
Ayo? @winged sundial level 2 !!!
winged sundial
since i cant find a good eminem one, i'll make one myself
safe walrus
i wish there was a good live infer app, i tried that chinese one but it sucked 😦
shy karma
Hello guys. When i try to convert some screams to another person's screams, it does not detect the original screams after converting. Any solutions?
rare gobletBOT
Ayo? @shy karma level 2 !!!
odd shale
Hello guys. When i try to convert some screams to another person's screams, it d...
Umm.. I'm not that sure if RVC is able to handle screams.
Not even if the model you wanna use doesn't contain scream data.
That's how it works.
modern knoll
hello guys , how to create own voice?
proper shale
hello guys , how to create own voice?
just record yourself speaking/singing
uncut sequoia
I can't find turkısh voice models
proper shale
I can't find turkısh voice models
just search for whoever you want
modern knoll
thx
rare gobletBOT
Ayo? @modern knoll level 1 !!!
midnight fern
How do i make a easy ai model on ios
strange surge
"ModuleNotFoundError " when i try to do the second step with the link , pls help ?
proper shale
"ModuleNotFoundError " when i try to do the second step with the link , pls help...
give us a full error pls
strange surge
give us a full error pls
i can't send image
ModuleNotFoundError Traceback (most recent call last)
<ipython-input-5-5eb462b363b1> in <cell line: 4>()
2 #@markdown Link the URL path to the model (Mega, Drive, etc.) and start the code
3
----> 4 from mega import Mega
5 import os
6 import shutil
ModuleNotFoundError: No module named 'mega'
proper shale
How do i make a easy ai model on ios
proper shale
ModuleNotFoundError Traceback (most recent call last)
<ipy...
what colab is that 
easy gui v2?
strange surge
what colab is that <:nails:1159569314848972891>
the link of the zip ?
proper shale
the link of the zip ?
no, what colab are you using
strange surge
easy gui v2?
yes
proper shale
yes
yeah thats
outdated
https://docs.aihub.wtf/ has newer Colabs and guides on em
proper echo
https://i.ibb.co/gVR268h/image.png
hi may i ask why my rvc doesn't have that 1st box (i can't do the wasapi thingy)
mine only has the input and output
pastel oak
https://i.ibb.co/gVR268h/image.png
hi may i ask why my rvc doesn't have that 1s...
do you have Nvidia or AMD gpu?
proper echo
nvidia
pastel oak
Download this, this is the most up to date version of rvc:
https://huggingface.co/Shadicti/RVCLatest/blob/main/RVCNvidiaLatest.zip
it has the box by default
pastel oak
❤️
rare gobletBOT
Ayo? @low berry level 1 !!!
low berry
Meow
uhm
i have question
the virtual cable appears everywhere besides the input device
on my discord
worldly field
Hi everyone
Question, is there any tool to separate the two voices in a podcast?
to have two files, one with the audio of one person and another with the second speaker
the closest I found only gives the timestamps
spaces/Delik/pyannote-speaker-diarization-3.1
wild vapor
Hi everyone
Question, is there any tool to separate the two voices in a podcast?...
are the voices overlapping
azure marshBOT
RVC Guides (How to Make AI Cover)
worldly field
are the voices overlapping
lets assume yes
rare gobletBOT
Ayo? @worldly field level 1 !!!
worldly field
but in a real case, a podcast/interview, we can discard the overlap as the reminding time of the audio will suffice
generating datasets requires too much time editing files x.x
lusty spruce
pastel oak
what gpu
i have the nvidia version repackaged available from the latest version
not amd yet
gusty kestrelBOT
@graceful quail, I have found 1 results that match your search!
Requested by karii8352
gusty kestrelBOT
@graceful quail, I have found 1 results that match your search!
Requested by karii8352
viral swallow
I haven't been able to log in for a long time. Where can I find the google collab link for rvc v2?
pastel oak
I haven't been able to log in for a long time. Where can I find the google colla...
outdated proabbly check the newer colabs
-colab
pseudo saffron
how would i uninstall RVC?
pastel oak
how would i uninstall RVC?
delete the folders thats all
pseudo saffron
its not allowing me
pastel oak
can you be more specific
pseudo saffron
whenever i try to uninstall a folder, it just tries to detect files infinitely
the only one that does this is runtime
and i cant get rid of it
nvm
rare gobletBOT
Ayo? @pseudo saffron level 1 !!!
pseudo saffron
it just randomly stopped being corrupted
fluid matrix
literally just delete the folder you have the files in, thats it
pseudo saffron
that wasnt working until now
fluid matrix
if you have any models you want to save, backup the files you need from the weights or logs folder
pseudo saffron
i dont know why it was corrupted but the problem i had just fixed itself
pastel oak
I have a problem and I can't send pictures why?
You need level 1 on the server for pics
explain the problem for now
later screenshot
exotic cedar
can models be optimised for voice conversion so that they work better with certain voices?
or like can they be tuned for a specific voice so that they work ideally with that voice?
rain thunder
steel kite
I'm trying to use the AICoverGen and despite me using collab, they're claiming I need an NVidia Driver how do I fix this?
azure marshBOT
RVC Guides (How to Make AI Cover)
thin stump
@ruby zinc You only need to look at g/total to detect overtraining since you haven't trained in a while https://docs.aihub.wtf/rvc/resources/epochs--tensorboard/#--monitoring
For advanced tips:
#1213509354343637065 message
then read this:
#✨│ai-help message
ruby zinc
thx a lot
thin stump
yw, it also doesn't hurt to try looking at kl and mel since g/total is only the average of that
ruby zinc
surprised it's not overtraining yet, i only gave it 6 mins of audio
doing 200 epochs bc it's a small sample size
proud elbow
I'm trying to use the AICoverGen and despite me using collab, they're claiming I...
make sure the runtime is running T4 GPU (in the top right status)
ruby zinc
overtraining?
covert turret
looks like it
covert turret
lol
covert turret
no afaik
rare gobletBOT
Ayo? @eternal dawn level 3 !!!
eternal dawn
nvm'
steel forge
overtraining?
Two things; show us the rest of the graphs, and also mention how many steps your model is currently at
From looking at that graph you sent, it's definitely not overtraining yet. But we need more context with the rest of the graphs
Steps are a better measurement for training progress, because epoch is effected by your batch size and dataset size
cobalt pilot
How do I remove backing vocals from a song?
azure marshBOT
azure marshBOT
RVC Guides (How to Make AI Cover)
tardy bone
Train more?
tardy bone
Hello?
tardy bone
incase you use rvc pretrain at least
I use the TITAN pretrain yea
gray crow
I use the TITAN pretrain yea
For Titan i heard: Batch size 8, 5 min dataset, 270 epochs
At least the said thats the sweet spot. Not sure about it. Im also wondering if i should train more than the given epochs
tardy bone
For Titan i heard: Batch size 8, 5 min dataset, 270 epochs
I did a 300 epoch run yesterday and had Litsa evaluate for me
they said it's fine but I need to train it more wtf
it took me well over two hours to pull that off
gray crow
interesting. I trained one yesterday too on titan. I will try to continue training there too.
tardy bone
My dataset clocks at over 6 minutes
gray crow
My dataset clocks at over 6 minutes
can you send a sample?
gray crow
You think these results sound scuffed to you or is it worth posting?
tbh i think its pretty good.
Except for the pitch being a little high i guess. But same problem for me with my trained one
So did they say what exactly will improve when further training? @tardy bone
tardy bone
Six minutes 55 seconds
tardy bone
Except for the pitch being a little high i guess. But same problem for me with m...
Not really I guess? I gave it a bit of a listen and it sounds fine to me
tardy bone
tbh i think its pretty good.
I mean, should I post the 300-epoch model now as it is? I'm worried it might get rejected again
I currently have a 430-epoch model tho
gray crow
I mean, should I post the 300-epoch model now as it is? I'm worried it might get...
tbh i dont even know why it got rejected in first instance. As i said IMO it sound clean. Dunno what needs to get better in order to be accepted.
tardy bone
tbh i dont even know why it got rejected in first instance. As i said IMO it sou...
Well the first iteration was done on a hastily-made dataset which is why it sounded so unclean
the second one is what the current samples are made out of
should I train even more then?
I see
Any suitable samples I could test against?
tardy bone
Also, since I am resuming from a backup, should I just click Start Training?
without having to do the preprocessing and feature extraction again?
tardy bone
@golden creek Train moar?
tardy bone
@red kayak Should I just stop it already? 550 epochs and not much progress has been made
rare gobletBOT
Ayo? @tardy bone level 8 !!!
golden creek
<@760477201921212466> Train moar?
yes just try to losten to different checkpoints and what sounds best, either 300 or 500
low quail
was the server wiped or something? or models of inactive people got nuked?
pastel oak
was the server wiped or something? or models of inactive people got nuked?
Server was terminated sometime end of 2023
low quail
oof
proud elbow
was the server wiped or something? or models of inactive people got nuked?
it was in last october
tardy bone
yes just try to losten to different checkpoints and what sounds best, either 300...
hmm let me send you guys some samples at 400, 450, 500 and 550 epochs to see if it's alright
narrow sorrel
what program do you use to cover ai songs? applio? and can it also just do text to speech (just text)?
i want to use clyde from sout parks voice as a youtube narrator, but don't know if it can do just text to speech/which program i should use
pastel oak
what program do you use to cover ai songs? applio? and can it also just do text...
theres a section on ai covers and tts on our docs guide. check pinned message
if you dont have a strong gpu, then use applio colab
it also has tts i think but dont know off the top of my head, check the tts category on the guide
red kayak
<@460577350900514837> Should I just stop it already? 550 epochs and not much pro...
-upload
red kayak
@tardy bone
narrow sorrel
theres a section on ai covers and tts on our docs guide. check pinned message
i...
thx, but what does it mean there is a usage limit for colab? it says here 4 hours
rare gobletBOT
Ayo? @narrow sorrel level 2 !!!
tardy bone
thx, but what does it mean there is a usage limit for colab? it says here 4 hour...
You can only do so much within an hour or two
that or Google would arbitrarily cut you off from the GPU
low shard
thx, but what does it mean there is a usage limit for colab? it says here 4 hour...
google colab gives around 4 hours for free of daily gpu
red kayak
3 hours and 10 mins :p
low shard
You can only do so much within an hour or two
the time depends user to user and also by the usage btw
narrow sorrel
google colab gives around 4 hours for free of daily gpu
ok nice, i thought you only get 4 hours and that's it for free use
low shard
nah its daily hours lol
narrow sorrel
the time depends user to user and also by the usage btw
yeah i am going to use only tts for now so won't be probably even an hour
red kayak
ok nice, i thought you only get 4 hours and that's it for free use
u can check ur stat console up top by clicking on it and it'll show u ur remaining time too
charred marsh
I trained 2 separate models, one was with a 6 ish minute sample, and 300 epochs, and the other one was a 45 ish minute sample, and nearly to 150 epochs. They both sound nearly identical, and not at all like the character (Vantage, from Apex Legends), although the 300 epoch one is smoother. What do I need to change? It sounds mostly like a female voice, and thats all. And thanks.
azure marshBOT
RVC Guides (How to Make AI Cover)
charred marsh
Thanks, you were really fast.
pastel oak
I trained 2 separate models, one was with a 6 ish minute sample, and 300 epochs,...
Every model has a different point of reaching its best version. We use tensorboard to track the graphs
On the docs guide it is explained
charred marsh
I will look at that, thanks. I'm very grateful for your guys's help.
rare gobletBOT
Ayo? @charred marsh level 1 !!!
rare gobletBOT
Ayo? @molten oar level 1 !!!
odd shale
Umm.. We strongly suggest to use huggingface links instead of Mega and Drive ones.
Have you tried downloading the model from that mega link and uploading it to your Huggingface account?
Have you checked if that model link isn't down?
Try putting it on your Huggingface account.
Also, you can use Applio colab.
All you gotta do is execute those Install and Start cells and click on the public link
Nope, first execute both cells and them click on the public link
rare gobletBOT
Ayo? @odd shale level 68 !!!
odd shale
The applio GUI got a download tab where you can put your huggingface link
Let me give you a guide about how to upload models to huggingface
Scroll down till "Upload to Huggingface" part
You too.
safe walrus
ehw