-audio
#✨│ai-help
thorn abyss
thorn abyss
-colab
thorn abyss
-overtrain
azure marshBOT
-overtrain
Moved to /faq command.
quiet crystal
AI HUB Docs
proven hill
-rt
hallow thistle
proven hill
hallow thistle
That's crazy. You wanna run both Applio the RVC and fork W-Okada at the same time? 
heady gorge
where is chat
hallow thistle
where is chat
#✦│chat and #🧬│ai-chat are chat channels.
valid spruce
What sample rate should I use?
proven hill
What sample rate should I use?
32k
lean grail
I have been out of the loop for a while, whats the best way to use speech to speech ai now?
craggy brook
It's such a SHIT in audio recording applications or audio files that I'm embarrassed to listen to it and turn it off immediately. Even if I sound normal, my voice sounds incredibly deep on the recording. I can use it with my own voice and automatically detect it. I can make it better, more realistic, without distortion, or with what kind of application?????
lean grail
Ah thank you
proven hill
no probs
swift zephyr
anybody here got a kacy hill voice model?
proven hill
It's such a SHIT in audio recording applications or audio files that I'm embarra...
voice changer?
proven hill
anybody here got a kacy hill voice model?
check #1175430844685484042 or weights.gg
swift zephyr
check <#1175430844685484042> or weights.gg
it is not in there unfortunatly
gonna give up
craggy brook
voice changer?
Changing your own voice, whatever you call it
proven hill
it is not in there unfortunatly
then request it in #1159289738314919936 or commission a model master in #1191429836321849435
proven hill
Changing your own voice, whatever you call it
in real time?
craggy brook
yes
proven hill
yes
read this guide: https://rentry.co/ForkVoiceChangerGuide
craggy brook
Is there a shorter and faster training? as video. I don't speak English anyway
craggy brook
F***K
lean grail
https://docs.google.com/document/d/1YbXcLFPaGjhOdG5NFkK3QrucCEpHZBwFUxkeMO8aB18/...
This doesent seem to do realtime voice changing, whats the best way to do that?
lean grail
read this guide: https://rentry.co/ForkVoiceChangerGuide
is it this?
proven hill
is it this?
yes
lean grail
tytyty
craggy brook
F***K
And people say don't use it, they say it is very difficult to differentiate. I want to make another one with my own voice, nothing comes out other than sounding like a robot.
for example this
lean grail
Bruh, the guide tells me that a 3090 can do 20-40 ms chunk plus 2.7 but my 4090 needs to be above 100 ms chunk for it to not lag
Any place i can sort voice models by ranking or popularity?
golden rune
guys I'm using colab for real time voice changer
how can I bind my output voice to discord or any other software that uses my mic
glad mountain
for example this
Hey don't say that please
glad mountain
Bruh, the guide tells me that a 3090 can do 20-40 ms chunk plus 2.7 but my 4090 ...
i can do 80ms on my 6700 xt
so somethings wrong lmao
brittle wing
-colab
median finch
is there an easy way to upsample and downsample my audio stream on windows to pin it to the sample rate that different models use?
proven hill
is there an easy way to upsample and downsample my audio stream on windows to pi...
why would you do that
its useless
just check the sample rate with spek
fickle dust
Can u use okada live rvc on discord voice messages?
median finch
just check the sample rate with spek
so the input sample rate doesn't matter? let's say i'm trying to realtime with a voice that is trained on 32k, my microphone is 96k. there will be no issues or degradation?
sudden tree
anyone know when 3.2.9 or 3.3 comes out for applio
i failed at custom installing the damn package update
simple ore
just get a compiled build, sheesh
sudden tree
what? ppl yesterday were saying compiled build was outdated even though i installed 12/25
they said g/loss/total graphs were not correct
lone nest
Does RVC disconnected stem vocal from input file?
My data is already clear single person voice so if it does any step to separate vocal from background sounds or from another person voice I want to skip that step
proven hill
so the input sample rate doesn't matter? let's say i'm trying to realtime with a...
nope
tight frost
I managed to train my model, but it gets some robotic voice at the end of the sentence, tried many epochs (250-500). Do i need to train it more? How to set it correctly?
Should I train it more times? 23mins of dataset, totally clean with silence reduction in audacity
proven hill
I managed to train my model, but it gets some robotic voice at the end of the se...
did you check the tensorboard?
-docs
proven hill
first link
brittle wing
Will it be okay if I include laughs & speech in my dataset of the same person aside from vocals?
tight frost
Thank you, so lets summarize:
1.download applio and pretrain model
2.train it correctly
proven hill
Will it be okay if I include laughs & speech in my dataset of the same person as...
yes but not a lot or it might break it
brittle wing
yes but not a lot or it might break it
REALLY
proven hill
Thank you, so lets summarize:
1.download applio and pretrain model
2.train it co...
pretrain model is influded in applio
brittle wing
How many minutes of laugh and speech?
proven hill
use the og pretrain
proven hill
How many minutes of laugh and speech?
id say 3 laughs top
brittle wing
use the og pretrain
Nah I'll use KLM x4
proven hill
Nah I'll use KLM x4
why
brittle wing
id say 3 laughs top
How much speech
proven hill
How much speech
min 10 minutes
proven hill
Cause I want to
make sense but the normal pretrain is better
brittle wing
make sense but the normal pretrain is better
No
proven hill
wdym no
brittle wing
Pretrains exist to ease model training also to improve the model results
proven hill
Pretrains exist to ease model training also to improve the model results
do you think i dont know what im talking about 😭
brittle wing
do you think i dont know what im talking about 😭
Nah I just have personal preferences and that's all
proven hill
make sense
brittle wing
Nothing to do w you
brittle wing
make sense
Wdym og pretrain is better?
In what sense I think pretrains are a modern variant to ease model training and all
proven hill
Wdym og pretrain is better?
less noise
proven hill
In what sense I think pretrains are a modern variant to ease model training and ...
yea but they also add noise
brittle wing
less noise
Mhm okay but KLM doesn't add noise like titan
brittle wing
yea but they also add noise
And...? That's not a problem at all since I have a bandlab preset that has noise gate in it
Also a little noise doesn't matter that much
proven hill
low shard
make sense but the normal pretrain is better
it’s not like every single pretrain is bad
proven hill
it’s not like every single pretrain is bad
no, but the og is still better
low shard
It all depends on ur dataset language and lenght
low shard
no, but the og is still better
og is trained only on english
proven hill
ofc im assuming hes trainingnin english
low shard
imagine u want an anime image, would u use an anime or realistic image model 
proven hill
this doesnt make any sense 😭
low shard
this doesnt make any sense 😭
if u want to talk in italian, you would use an LLM that has been trained on it rather than one that didn’t
proven hill
if u want to talk in italian, you would use an LLM that has been trained on it r...
ye but he was training in english presumably
low shard
presumably
low shard
Mhm okay but KLM doesn't add noise like titan
what language is your dataset?
brittle wing
what language is your dataset?
Korean + few English
simple ore
what? ppl yesterday were saying compiled build was outdated even though i instal...
current repository has unreleased stuff, including better logging, but it is still a work in progress
tender basalt
What is the best version for mmvc
And where I can download it
low shard
What is the best version for mmvc
And where I can download it
I’m guessing you’re talking about Wokada, which is a realtime voice changer, different than rvc program
Use #🔍│help-w-okada
hallow thistle
MMVC is a name for W-Okada. This #✨│ai-help here is about RVC.
tender basalt
I’m guessing you’re talking about Wokada, which is a realtime voice changer, dif...
Sorry then
low shard
dw
sudden tree
current repository has unreleased stuff, including better logging, but it is sti...
Is the colab fully updated to repo
hallow thistle
Some files that have been uploaded to Ilaria RVC from Hugging Face are all saved somewhere in %temp%, I found out that today.
low shard
Some files that have been uploaded to Ilaria RVC from Hugging Face are all saved...
why would u use that locally
hallow thistle
why would u use that locally
Curious. 
brittle wing
ofc im assuming hes trainingnin english
Nope
brittle wing
<:Lace_shrug_FB:536284758326378506>
People have preferences
proven hill
brittle wing
Also I'm the one making the model so I decide for myself
proven hill
no one is arguing here
brittle wing
Okay does KLM x4 handle speech and laughs
brittle wing
no one is arguing here
I see
knotty moth
Okay does KLM x4 handle speech and laughs
x2 is better at handling high pitches
hallow thistle
Finally, it indeed went well. 
proven hill
Finally, it indeed went well. <:drayke:1159653657307000962>
insane
hallow thistle
It took an hour to finish because of how slow my laptop CPU is. It would be faster with a better PC and an NVIDIA GPU. 
proven hill
It took an hour to finish because of how slow my laptop CPU is. It would be fast...
it amazing it runs locally
lone nest
40 mins
proven hill
completely normal
lone nest
Meanwhile the 4 hours dataset gives 30 MB .index
proven hill
…how
knotty moth
Meanwhile the 4 hours dataset gives 30 MB .index
that's with minibatch k-means
with the normal faiss it would be ~2 GB
hallow thistle
I just made a 300 MB .index file
Average .index file size for RVC voice model weighs around thousand megabytes, so this is fine.
knotty moth
applio has the option whether to use faiss or k-means
lone nest
with the normal faiss it would be ~2 GB
Damn
hallow thistle
Imagine that Ilaria RVC from Hugging Face being developed to run locally. 
lone nest
with the normal faiss it would be ~2 GB
How does the speed compare
Thinking of changing the code to force it to use faiss but maybe Colab usage will run out first lol
knotty moth
Thinking of changing the code to force it to use faiss but maybe Colab usage wil...
u can do train index locally just using average laptop cpu
proven hill
Imagine that Ilaria RVC from Hugging Face being developed to run locally. <:nail...
why….
proven hill
lol
simple ore
applio has the option whether to use faiss or k-means
it onlt happend when you have 200k chunks
after ~4000 starts shuffling and showing a text output as it attempts to reshuffle pieces that are somewhat unique
with 200k chunks the extracted features get grouped into clusters and most of them get discarded
that again brings the index to somewhatmanageable size
verbal moon
Bro i have delay and my voice is very distorted can one fix this
hallow thistle
Bro i have delay and my voice is very distorted can one fix this
For W-Okada, go to #🔍│help-w-okada.
red violet
Could anyone help me creating ai cover? i have the model and isolated vocals ready ! 
azure marshBOT
Could anyone help me creating ai cover? i have the model and isolated vocals rea...
low shard
Could anyone help me creating ai cover? i have the model and isolated vocals rea...
What's ur pc gpu
red violet
RTX3050 laptop
past ibex
lary over
low shard
RTX3050 laptop
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- Cloud (remote good pc, easier and faster than ur PC but it's limited):
- Ilaria RVC Zero: fastest and simplest that you can get for free
- Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio Colab: max 4 hours daily, not granted, of GPU
Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio
red violet
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), ...
thank you so much
low shard
thank you so much
yw lmk
agile isle
hey i need a app to run my Models and i simply cant find any :c
trim sparrow
Is Applio broken ? I have a linear graph on a 20minutes audio training (with google collab), does anyone have a solution ? Here are my parameters Sampling rate : 40k; Pitch extraction : Crepe; Batch Size 15; and 500 epoch in total. I kept the batch size at 15 because it generally gives me better results. Hope you can help me. (The graph is loss/g/total)
analog obsidian
Is Applio broken ? I have a linear graph on a 20minutes audio training (with goo...
click the third blue box
trim sparrow
Ohh ok thanks a lot !
narrow nova
Is RVC v2 Disconnected down?
simple ore
What is it issue?
mismatch between sample rates of the model and the pretrain
513 - 32k model
mellow flax
sorry for tagging after so long, but where i can download these models? when i unfold vr or mdx these models are not there
viscid moss
sorry for tagging after so long, but where i can download these models? when i u...
u need to add that models manually
and u also need UVR5 beta for that
mellow flax
and u also need UVR5 beta for that
this one i got covered
quartz holly
Is RVC v2 Disconnected down?
having the same issue
mellow flax
Do you know where I can download them?
viscid moss
ur welcome
tight frost
is it ok if my dataset is 1hour and 32 mins? XD
mellow flax
is it ok if my dataset is 1hour and 32 mins? XD
as long as the dataset has clear audio, why not, but know that it will take a long time to train
tight frost
as long as the dataset has clear audio, why not, but know that it will take a lo...
1min to train 1epoch, and i set 800epochs sooo yep, pretty long
mellow flax
time will tell
narrow nova
valid spruce
Help
File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap
self.run()
File "/usr/lib/python3.10/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/content/Mangio-RVC-Fork/train_nsf_sim_cache_sid_load_pretrain.py", line 254, in run
train_and_evaluate(
File "/content/Mangio-RVC-Fork/train_nsf_sim_cache_sid_load_pretrain.py", line 438, in train_and_evaluate
"slice/mel_org": utils.plot_spectrogram_to_numpy(y_mel[0].data.cpu().numpy()),
File "/content/Mangio-RVC-Fork/train/utils.py", line 231, in plot_spectrogram_to_numpy
data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep="")
AttributeError: 'FigureCanvasAgg' object has no attribute 'tostring_rgb
pastel oak
Help
File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _boots...
mangio is hella outdated just get something more recent
valid spruce
What?, I used it earlier today and it was working perfectly!
I did the procedure I always do on all my models!
Could it be that I created a new account?
narrow nova
Help
File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _boots...
RVC v2 Disconnected?
I have the same error on mine
valid spruce
RVC v2 Disconnected?
I have the same error on mine
Yes
This just happened now @narrow nova
valid spruce
https://docs.ai-hub.wtf/rvc/local/applio/
I really hate Applio
It always disconnects by itself and I end up losing everything
pastel oak
It always disconnects by itself and I end up losing everything
isnt mangio rvc fork local?
narrow nova
yes colab
valid spruce
You are using colab?
Yes
pastel oak
Ok cant help wait for someone else
valid spruce
I create my models on my cell phone
pastel oak
But rvc disconnected & mangio rvc are really not recommended these days
i cant stop you from using it so goodluck
valid spruce
But rvc disconnected & mangio rvc are really not recommended these days
I know that
Unfortunately Applio is very unstable and always disconnects
@pastel oak Is there any version of Applio without UI?
pastel oak
why do i need discord to use applio
you dont
pastel oak
<@965636299509882980> Is there any version of Applio without UI?
not that i know of
simple ore
<@965636299509882980> Is there any version of Applio without UI?
noUI Applio Colab
sterile saddle
-colab
sterile saddle
-train
viscid moss
If you are facing RVC Disconnected errors try this fix
#📑│making-models message
or pip install matplotlib==3.7.0
simple ore
or pip install matplotlib==3.7.0
unique rock
or ``pip install matplotlib==3.7.0``
I don't really understand the process, I'm very confused
I train on a cell phone and it's a bit confusing. Can you explain the process to me in more detail?
viscid moss
I train on a cell phone and it's a bit confusing. Can you explain the process to...
u need to add that line like this:
unique rock
u need to add that line like this:
only that one?
viscid moss
only that one?
yep
mild sleet
Man when using weights the models are eh
sterile saddle
If you are facing RVC Disconnected errors try this fix
https://discord.com/chann...
thanks,it works
lone nest
`if big_npy.shape[0] > 2e5 or force_mbkm:` oh makes sense
Yeah it's 2.4 GB now
Is it used in the training step? i.e. will it affect the .pth file?
opal cobalt
hi did someone now a uvr model that remove the sound at the end of my audio pls ? i already test a lot of model with this audio but someone tell my to ask here where i can find custom model
mortal stirrup
can somebody help me i wanna train my voice in applio
opal kelp
Could anyone help me? training does not start
modest thunder
guys what is "epochs 300"?
azure marshBOT
guys what is "epochs 300"?
opal kelp
The same thing is happening to me too
upper furnace
upper furnace
The same thing is happening to me too
maybe I hope not but did they ban him?
opal kelp
-colab
flint solar
Why r u using mangio fork 😭
knotty moth
Why r u using mangio fork 😭
suddenly someone having had the matplotlib related issue, then fixed and more ppl using it
knotty moth
you two pls read the pinned message about matplotlib fix here, or you can try my tweaked version here: #1159290752195633273 message
but anyway I'd recommend the latest Applio 3.2.8 colab/kaggle with more features and improvements
hallow thistle
-gui
low shard
If you are facing RVC Disconnected errors try this fix
https://discord.com/chann...
I feel like this should be said in #📰│dev-updates
People never check pins
viscid moss
I feel like this should be said in <#1159380240271953940>
ye
can u do it? till Kit Lemonfoot fix it
low shard
can u do it? till Kit Lemonfoot fix it
Sure
low shard
Have you already told Kit about that?
I pinged him in #💬│staff-chat
viscid moss
I pinged him in <#1159289516268474570>
Oh, nice
low shard
Oh, nice
Maybe it's better I say to either use the modified link in #1159290752195633273 message or adjust the fix themselves
Bc I feel like telling them to upload the file will just make it harder for the average user
viscid moss
Bc I feel like telling them to upload the file will just make it harder for the ...
real
use this pic if u want
unique rock
If you are facing RVC Disconnected errors try this fix
https://discord.com/chann...
I got this
opal cobalt
hello, where can i find a custom model to remove this noise at the end of my audio with uvr5 pls ?
low shard
use this pic if u want
viscid moss
https://discord.com/channels/1159260121998827560/1159380240271953940/13269105913...
ty
viscid moss
hmmm lemme try rq
low shard
ty
Should I also specify that people can ofc use alternatives?
I genuinely thought fewer people would be using RVC Disconnected since there's Kaggle
viscid moss
Should I also specify that people can ofc use alternatives?
ye ye
unique rock
hmmm lemme try rq
Yeah
low shard
<@965636299509882980> Is there any version of Applio without UI?
I genuinely thought people preferred UIs more tbh
low shard
hmmm lemme try rq
valid spruce
I genuinely thought people preferred UIs more tbh
The UIs are great! The problem is that colab disconnects and gradio stops working😕
low shard
The UIs are great! The problem is that colab disconnects and gradio stops workin...
Btw since a while we encrypt the code which solves that issue
low shard
Did you have that issue recently or heard it from others?
Because that used to happen before, but shouldn't anymore
valid spruce
I use my cell phone so that must be why
valid spruce
Did you have that issue recently or heard it from others?
Yeah, it happened this week when I was creating a cover.
viscid moss
I use my cell phone so that must be why
maybe
cause on phone u need to switch the browser tabs on UI ones
valid spruce
I usually leave the tabs like this
low shard
Yeah, it happened this week when I was creating a cover.
What was the exact warning?
disallowed code execution? Inactivity?
valid spruce
disallowed code execution? Inactivity?
Reconnection
valid spruce
When the collab reconnected I had to reload my gradio
viscid moss
https://discord.com/channels/1159260121998827560/1159380240271953940/13269136759...
yooo that's was rq
nova rock
yo can someone help me rq?
azure marshBOT
yo can someone help me rq?
low shard
Reconnection
That happens for inactivity, so it’s because of the mobile tab, not because of the UI
if it was for the UI, it wouldn’t automatically reconnect
patent trellis
low shard
yo can someone help me rq?
be specific
viscid moss
Yeah
u face a warning message right? If so press cancel. Warning says "U need to restart or something"
And continue
nova rock
be specific
everything by my rvc seems to be correct like the setting but it just simpy isnt working
low shard
everything by my rvc seems to be correct like the setting but it just simpy isnt...
which
what error
You have to be specific
valid spruce
That happens for inactivity, so it’s because of the mobile tab, not because of t...
I see, I wish I could find a better mobile browser.
low shard
what pc gpu?
nova rock
like the voice changer isnt working, i can hear myself but the voice changer isnt applying
low shard
like the voice changer isnt working, i can hear myself but the voice changer isn...
wrong channel, rvc isn’t a realtime voice changer
share the tutorial link you followed in #🔍│help-w-okada
unique rock
u face a warning message right? If so press **cancel**. Warning says "U need to ...
oh, i hit restart haha, thanks
viscid moss
wrong channel, rvc isn’t a realtime voice changer
||R - Realtime
V - Voice
C - Changer
||
low shard
||R - Realtime
V - Voice
C - Changer
<:trolley:1159468147133395025> ||
hell no
viscid moss
xd
uneven relic
Hi, i have a question, The Hugging Face Pro plan is not useful for using Ilaria_RVC?
unique rock
Now it comes out like this, without needing to put: !pip install matplotlib==3.7.0
low shard
Now it comes out like this, without needing to put: !pip install matplotlib==3.7...
the creator updated it
low shard
Hi, i have a question, The Hugging Face Pro plan is not useful for using Ilaria_...
well, the huggingface pro plan will give you more usage time for zerogpu iirc
unique rock
the creator updated it
I see, because I restarted Google and everything came out.
uneven relic
well, the huggingface pro plan will give you more usage time for zerogpu iirc
I don't understand why I'm always out of credits even with the Pro version, I can only generate once every 15 minutes even with a Pro subscription?
viscid moss
I don't understand why I'm always out of credits even with the Pro version, I ca...
U should contact HF staff for that
I saw ur ticket and it's kinda weird how ur Zero GPU time is full
uneven relic
U should contact HF staff for that
Okay, its not normal so ?
viscid moss
and still can't use it
viscid moss
Okay, its not normal so ?
The Zero GPU usage updates every GPU request that u do, so ya it's weird
uneven relic
The Zero GPU usage updates every GPU request that u do, so ya it's weird
Okay
hallow thistle
Unless the Hugging Face system is just glitching out.
hallow thistle
You can somehow run Ilaria RVC (from Hugging Face) locally, but this way is kinda complicated to get it work, as you'll need to download the entire repository using Hugging Face Hub on Python. https://cdn.discordapp.com/attachments/1159289354439626772/1326182820430745682/image.png
I realized 2 FFmpeg .exe files were missing from this repository, so I added them using ones from Ilaria RVC from GitHub. The first time launching on my laptop, it just picked up FFmpeg from C:\FFmpeg rather than from inside the Ilaria RVC folder itself.
uneven relic
You can somehow run Ilaria RVC (from Hugging Face) locally, but this way is kind...
With an RTX 2060, it's complicated to run it locally?
reef tusk
does anyone have the links from cloud RVC?
hallow thistle
With an RTX 2060, it's complicated to run it locally?
I don't think the author made it to run on NVIDIA GeForce GPU. In the terminal, it says something like "T GPU", which I think it supposed to run on NVIDIA Tesla GPU like A100 or H100.
reef tusk
I don't think the author made it to run on NVIDIA GeForce GPU. In the terminal, ...
it works fine on my 4060 ti
hallow thistle
it works fine on my 4060 ti
Ilaria RVC from Hugging Face or GitHub?
viscid moss
does anyone have the links from cloud RVC?
-colab
viscid moss
-kaggle
reef tusk
oh
hallow thistle
The Ilaria RVC version I'm talking about is the one that used in its Hugging Face space, and I downloaded it in a dirty way. 
viscid moss
on colab, how much generations you can do ? Its like HF ?
per account if u re out of GPU just switch acc
uneven relic
per account if u re out of GPU just switch acc
All right so its better than HF ?
viscid moss
in terms of inference time ye
hallow thistle
low shard
in terms of inference time ye
google colab gpu is slower
if you meant about having more time to use it, or uploading longer audio files, then yes
but zerogpu huggingface time is faster
unless its a cpu one
viscid moss
if you meant about having more time to use it, or uploading longer audio files, ...
ye i mean this
uneven relic
google colab gpu is slower
It free on colab ? and do you have a guide ?
low shard
It free on colab ? and do you have a guide ?
yes, and guides are in https://docs.ai-hub.wtf
low shard
With an RTX 2060, it's complicated to run it locally?
wait you got a 2060 gpu
why don’t you do it locally on your pc instead?
huggingface space and google colab are meant only for people who got a bad pc, they are cloud computijg services
uneven relic
why don’t you do it locally on your pc instead?
Dont know how to install it xD
low shard
locally, it will be harder to setup and won’t be as fast as huggingface for example, but you will have unlimited time
low shard
Dont know how to install it xD
still in that docs i sent you, there is a guide for local too
i would suggest you applio
uneven relic
i would suggest you applio
I will check, thank you bro
low shard
I will check, thank you bro
yw
meager lava
Does anyone know what this means and how to fix it?
2025-01-09 [E:onnxruntime:Default, cuda_call.cc:116 onnxruntime::CudaCall] CUDA failure 35: CUDA driver version is insufficient for CUDA runtime version ; GPU=1040 ; hostname=DESKTOP; file=D:\a_work\1\s\onnxruntime\core\providers\cuda\cuda_execution_provider_info.cc ; line=125 ; expr=cudaGetDeviceCount(&num_devices);
simple torrent
Yes you can, you just need a pre-compiled fork and VxKex to run it
simple torrent
latent kettle
Do anyone know which TTS is best?
simple ore
Do anyone know which TTS is best?
depends on the language
low shard
**Does anyone know what this means and how to fix it?**
2025-01-09 [E:onnxrunt...
are you sure your drivers are updated
meager lava
are you sure your drivers are updated
yes, and i have amd
ancient current
what does "AttributeError: 'NoneType' object has no attribute 'tobytes'" mean
and ""HTTP/1.1 500 Internal Server Error""
meager lava
Are you on windows? Which GPU exactly
i have rx 6700 xt GPU
i tried another version (std instead of cuda) and it works now
low shard
what does "AttributeError: 'NoneType' object has no attribute 'tobytes'" mean
are u talking about RVC Disconnected? check #📰│dev-updates
low shard
i tried another version (std instead of cuda) and it works now
are you talking about Wokada (aka MMVCServerSIO or client)
this is the wrong channel, RVC is not realtime voice changer, wokada is
use #🔍│help-w-okada
wise cove
What's the difference between training model services and personal AI Voice model? ^^
pastel oak
What's the difference between training model services and personal AI Voice mode...
..nothing?
pastel oak
do i need mmvcserver?
i wonder whats going through your mind to write this question out of context out of the blue
like what do you need what do you want
wise cove
..nothing?
You sure~
Someone is offering both in the #1191429836321849435
worthy crest
i wonder whats going through your mind to write this question out of context out...
fixed lol
but ye i havvent used this for some time forgot what i needed to install
wise cove
Third one down I believe
worthy crest
i initially installed the other one yk
wise cove
i initially installed the other one yk
You need the client. It comes with everything you need
They changed it a but still technically the same thing assuming
worthy crest
ye i installed it
worthy crest
You need the client. It comes with everything you need
also
where do i post a photo#
can u tell me if my advanced settings are alr lol#
pastel oak
You sure~
Someone is offering both in the <#1191429836321849435>
Probably just a labelling thing everyone calls it differently
pastel oak
can u tell me if my advanced settings are alr lol#
youre talking in the wrong channel
worthy crest
<#1159290161683767298>
oh sorry
sand bison
Hello, is there anyone who can help me with the error: FileNotFoundError: [Errno 2] No such file or directory: '/content/Mangio-RVC-Fork/logs/
pastel oak
Hello, is there anyone who can help me with the error: FileNotFoundError: [Errno...
pinned msg
I didnt even read the error I just read mangio rvc fork and assumed its the same error, though you might have to do it either way
it cant find the folder, so the folder doesnt exist. check again
sand bison
The folder is created and the index is created, what isn't created is the pth.
hallow void
is there a way to use realtime rvc gui on roblox?
like so people on voice chat could hear?
low shard
is there a way to use realtime rvc gui on roblox?
Realtime rvc GUI isn't suggested
Old
Wokada deiteris fork is better
hallow void
is that realtime
hallow void
where can i download pls?
low shard
where can i download pls?
What's ur PC gpu
sand bison
my error is : FileNotFoundError Traceback (most recent call last)
<ipython-input-18-5d26394eab49> in <cell line: 86>()
86 set([name.split(".")[0] for name in os.listdir(gt_wavs_dir)])
87 & set([name.split(".")[0] for name in os.listdir(feature_dir)])
---> 88 & set([name.split(".")[0] for name in os.listdir(f0_dir)])
89 & set([name.split(".")[0] for name in os.listdir(f0nsf_dir)])
90 )
FileNotFoundError: [Errno 2] No such file or directory: '/content/Mangio-RVC-Fork/logs
hallow void
What's ur PC gpu
nvidia geforce rtx 3050 oem
hallow void
tysm
low shard
my error is : FileNotFoundError Traceback (most recent c...
What Google colab u using
sand bison
I don't know.
low shard
I don't know.
Send the link of what u using
sand bison
What is the latest version of rvc?
sudden tree
hey if i have raw studio session audio files but you can hear the backing instrumental through headphone leak what mvsep model would you use to clear that
also what settings in rx 11 to clear noise pre train
brazen sparrow
latent kettle
depends on the language
English and hindi.
maiden wadi
Hi, is it not possible to download rvc anymore?
hollow thunder
I keep getting this error, but everytime I restart the runtime it happens again, and it always disconnects before I get to do anything. I'm using Hina_Mod_AICoverGen_colab.
latent kettle
Hi, is it not possible to download rvc anymore?
Which rvc
jaunty mason
anyone have sunoAI ?
lone nest
Is it used in the training step? i.e. will it affect the .pth file?
So does it matter to the .pth file whether I used faiss or kmeans for the .index file?
glacial pollen
So does it matter to the .pth file whether I used faiss or kmeans for the .index...
no
honest canyon
how can i make my own voice a model that i can use
low shard
how can i make my own voice a model that i can use
what’s ur pc gpu
honest canyon
what’s ur pc gpu
its bad
brittle wing
What will happen if I include a few seconds of breathing in my dataset?
glacial pollen
What will happen if I include a few seconds of breathing in my dataset?
Nothing, it'll do fine. In fact, breathing samples in a dataset is a good choice
( provided they belong to the voice / whoever's the model's voice provider. If it's from someone else, can't promise anything )
knotty moth
the old guide that suggests removing breaths is kinda misleading
glacial pollen
^
brittle wing
Nothing, it'll do fine. In fact, breathing samples in a dataset is a good choice...
How many seconds of speech and laughs aside from vocals and rap?
glacial pollen
well, the deal is always balance but if we speak of misc such as breathing, there's no rule
same goes for laughs and speech tbh
brittle wing
Cause I want to include so many things also does KLM x4 support that?
glacial pollen
Key is balance so the model does not bias towards one and not the other
Not sure about KLM's behavior exactly as I do not maintain it ( aka, ain't sure what it does support and what not. I haven't tested it )
brittle wing
Key is balance so the model does not bias towards one and not the other
How many minutes or seconds
brittle wing
Not sure about KLM's behavior exactly as I do not maintain it ( aka, ain't sure ...
Uhm
glacial pollen
It depends man, in world's ideal scenario, you want to have equal balance for each subtype of content
we can't predict how things come out really
brittle wing
It depends man, in world's ideal scenario, you want to have equal balance for ea...
Wdym
glacial pollen
well, laughing, breathing, moaning, spitting whatever really, is a subtype
I can only give you an example of how I'd manage it
5 mins of diverse talking, 5 mins rapping, 5-7 mins of singing, 1 min of breathing ( good quality )
brittle wing
I can only give you an example of how I'd manage it
5 mins of diverse talking, 5...
Mhm I actually use the minutes randomly like
glacial pollen
was just an example
by no means was it in " sequential manner "
not that it'd matter anyway, all gets chopped inside and mixed / shuffled
brittle wing
In my case some rap verses 4 vocal songs
brittle wing
not that it'd matter anyway, all gets chopped inside and mixed / shuffled
Mhm yes but the person has so many voice variations.
Will the training support that
Deeper, higher
I don't mean pitch
glacial pollen
wdym deeper / higher if it's not pitch
brittle wing
5 minutes laughing, 10 speech?
glacial pollen
well no, that's a bad idea
brittle wing
wdym deeper / higher if it's not pitch
Well how do i explain
glacial pollen
Laughing should be treated as misc, as an extra
not the primary purpose of the model
brittle wing
A person can go from higher and louder tones
glacial pollen
well ye, that is a pitch
brittle wing
And lower deeper
glacial pollen
" higher, lower, tone "
that's a pitch
brittle wing
Hmm yes will the training support that stuff as well
glacial pollen
It supports anything you throw in
as long it's all balanced and enough of it ( not too little, not too much ) it'll do fine
glacial pollen
In ideal situation, yea
All comes down to how well you train it and how good is the set
brittle wing
In ideal situation, yea
How much speech And laugh
Cause I'm thinking of making ICP covers the laugh is required
glacial pollen
No idea man, as I said, we nor nobody else in here can give you any estimates
glacial pollen
machine learning like this is not deterministic where you can estimate how much of X is needed
better if you just trained a few times and see how it goes in various set proportions
brittle wing
Yes but I wanna include the person laughing
glacial pollen
It supports anything you throw in
as long it's all balanced and enough of it ( n...
^
This applies to anything
brittle wing
And talking
glacial pollen
seriously man, people used to make damn chainsaw or water drop models
brittle wing
You day not too little, not too much so 5 mins of laughing?
And how many of talking
glacial pollen
no..
too little can be 1 min of laughing vs 10-15 mins of speech
you get the deal
if you overdone one type of content, model will bias towards it
brittle wing
no..
too little can be 1 min of laughing vs 10-15 mins of speech
Will that be enough
glacial pollen
Yea so just try some proportions, do a few training runs and see how it goes
Doing a good model needs some testings naturally
brittle wing
5 mins of laughing 10 of speech?
glacial pollen
I suppose you can try ^
if that fails, try 2-3 of laughing and 8-10 of speech
and if that failed too, 5 of laughing, 6-7 of speech
brittle wing
Because when I convert into RVC the laughs sound so unnatural like aehehdjdo
brittle wing
I suppose you can try ^
if that fails, try 2-3 of laughing and 8-10 of speech
a...
I don't wanna waste time tbh
glacial pollen
Then you won't get your " perfect results "
brittle wing
Then you won't get your " perfect results "
Uh
glacial pollen
My best model was a matter of 2~ months worth of experiments
you can't toss all in, do a 1 run of training and expect miracles.. that's just the truth
brittle wing
My best model was a matter of 2~ months worth of experiments
You got si much time on your hands
glacial pollen
What's the problem of giving 40 mins to 1.5h of training
brittle wing
you can't toss all in, do a 1 run of training and expect miracles.. that's just ...
Uh really
It's not my first time training either
brittle wing
Can I send you the dataset one day?
glacial pollen
for?
as in, y
Cause like, I don't really do datasets inspection
( Neither It'd help the case in any way. )
glacial pollen
All you gotta know is that I haven't made nor maintain custom pretrains ( which supposedly do or do not support laughing or whatever )
the core concept of rvc is to handle speech or singing, that's the og premise. Whatever customs do, is out of my reach.
Now, if you intend to perfect out or nail experimental stuff or things that are known to rather fail most of the time, you gotta put some time and effort in
brittle wing
And tell me what's unnecessary or not
glacial pollen
Well, I don't really think it's necessary
You said it's not your first time training, right? so you should know the deal
and yeah, I can't tell you to what include and what not, without experiments you can't really estimate anything or guess how model'd perform
If you don't trust your skills or are short on time, just commission me or anyone else, see what'll come out of it
else give it some time and work, nail it and be happy, ye
no pain no gain as they say lol
glacial pollen
Cause like, I don't really do datasets inspection
( Neither It'd help the case i...
^
Sorry man, that's not something I am up to
brittle wing
Sorry man, that's not something I am up to
It's okay
uneven relic
Hi, on Applio, can we not record vocal longer than 15 seconds? Every time I try to make a vocal longer than 15 seconds, this error pops up
simple ore
Hi, on Applio, can we not record vocal longer than 15 seconds? Every time I try ...
why wont you actually use a software intended for recording?
at least use Audacity
latent kettle
@simple ore I want best tts for English and hindi language. I want to use it locally. I don't want to use any website or any software like that. I need a fully open source project. Do you know any tool ?
simple ore
for english the usual recommendation - f5 tts, fish speech, fpt-sovits (needs training)
for hindi it is tricky, the perhaps fairseq vits from facebook
or you need to train the models above
brittle wing
its not opening i got python and shit
latent kettle
https://github.com/idiap/coqui-ai-TTS?tab=readme-ov-file#tts-using-fairseq-model...
Do you know how to train custom models and how to install all these stuff
simple ore
not for fairseq
latent kettle
not for fairseq
How can I learn. Is there any youtube tutorial or any server?
low shard
its not opening i got python and shit
!howtoask
unique rock
Does RefineGan work on rvc disconnected?
tame mica
Does RefineGan work on rvc disconnected?
no
tame mica
what
tame mica
you mean colab?
unique rock
you mean colab?
refineGan, what does or what
tame mica
refineGan, what does or what
alternative vocoder for rvc
little pier
tutorial br para baixar o gravador?
rain solar
hey guys!
im very confused on how make an ai model, i got the vocals and all those things
but idk if i have to download a program that uses gpu or alot of cpu
to train it
jaunty anvil
honestly i was actually wondering where you even put in the models
is it elevenlabs?
sorry that i'm not helping i'm just hoping someone knows
bold yarrow
sorry?
jaunty anvil
the best plan for elevenlabs
bold yarrow
idk about elevenlabs
jaunty anvil
oh mb
bold yarrow
RVC
jaunty anvil
does that work with tts?
bold yarrow
yeah but the tts sounds kinda robotic
jaunty anvil
yeah i'm mostly looking for tts
bold yarrow
oh
custom tts, no i dont think so
many custom tts i've seen sound like throat cancer
like 15.ai
jaunty anvil
rip
bold yarrow
yeah, who knows what the guy behind 15.ai is doing
its been offline for like a year now
jaunty anvil
honestly i'm just trying to find a good tts
bold yarrow
there arent many in the field if im gonna be honest
the most you can do
is use elevenlabs to generate generic speech
and then use RVC like ilaria rvc
bold yarrow
to turn it into a character you wish it to be
low shard
yeah i'm mostly looking for tts
There are different Text To Speech (TTS) AIs:
GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/
Freemium 11labs: An easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS
FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site
With RVC Models:
RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)
If you wanna do tts locally with RVC Voice Models (if you got a good pc):
If you don't got a good pc you can do tts with RVC Voice Models on cloud:
-
Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide
-
Use Applio UI Colab (with google colab T4 free daily limit gpu)
-
if you don't wanna use edge tts, you could try another tts ai from our tts index and use the output as an input in rvc
jaunty anvil
thanks for helping
low shard
this might help
jaunty anvil
lmao
low shard
this guy destroyed every argument i had
I mean, you're right that 11labs is the best tts
except that it's paid
jaunty anvil
nah i've got the money
bold yarrow
it makes the voice sound a bit different
and you cant even control which accent it develops
low shard
nah i've got the money
Then you could just use 11labs, you can train models directly in that iirc,
Or try as he said to use it as an input for RVC
low shard
it's nice hearing all of the sides
Honestly I haven't used it since I like more Open Source tools
jaunty anvil
alright
so i'm guessing you can't tell me which plan is best either?
...sorry that came off rude
low shard
...sorry that came off rude
nah it didn't dw, I was actually tryna check their plans, but maybe it's better you ask someone who actually used it
I don't want give you bad info ofc
low shard
is rvc disconnected still available for training?
rvc disconnected has been fixed ye #📰│dev-updates
bold yarrow
alr cool
im trying to make a heavy 2007 ai model
but idk what to do with steps and epochs
epochs im guessing i can figure out using tensorboard, but ive never used it
and as for steps, idk what they are
low shard
epochs im guessing i can figure out using tensorboard, but ive never used it
epochs = unit of measurement of the training cycles, how many times ur dataset has been trained
there isn't a right amount
bold yarrow
i thought there was
im trying to make my model not undertrained but not overtrained either
low shard
and as for steps, idk what they are
steps = a single update to the models parameter
low shard
i thought there was
nope, there isn't, https://docs.ai-hub.wtf/rvc/resources/training/
btw don't share datasets here
bold yarrow
ok
low shard
copyright?
yes
brittle wing
Should I load dataset in rvc Disconnected when resuming?
austere stream
question, is there a guide out there that suggests a general total training time/epochs for a dataset length? like "if you have a 20 minute dataset, train for x minutes/epochs"
unique rock
Should I load dataset in rvc Disconnected when resuming?
No, only the folders with the information of the first training with the cell number "2333333" import model for resume
To retrain you skip all the cells inside the "preprocessing" cell and you skip to the train one.
worn breach
whats a website that has real time voice changers?
weights.gg isn't real time right?
simple ore
question, is there a guide out there that suggests a general total training time...
no, it is a random process and nobody can tell you a good estimate
turbid tiger
Is there any RVC Text to speech program out there that doesn't require a ton of github usage and python stuff?
I can't always afford to talk into it.
hallow thistle
Is there any RVC Text to speech program out there that doesn't require a ton of ...
Applio can do this.
And also Ilaria RVC on Hugging Face can do this.
turbid tiger
Thank you for being so quick, gah damn.
hallow thistle
whats a website that has real time voice changers?
W-Okada is a realtime voice changer program. For Weights, I've already told you in #🧬│ai-chat.
Voice.ai also has realtime voice changer service, but this is a real scam and never advised to use it.
bitter kite
Which voice model should I choose to insert the .json file?
hallow thistle
Which voice model should I choose to insert the .json file?
What is the point of that?
If it's some kind of RVC fork program that uses .json file for a voice model to display its information, I don't know. But Applio doesn't seem to do that.
bitter kite
I wish I could upload a photo
hallow thistle
If it's W-Okada, the only .json file that found inside a voice model folder is params.json.
hallow thistle
I wish I could upload a photo
knotty moth
question, is there a guide out there that suggests a general total training time...
just start training with given batch size and see how long the first few epochs take
turbid tiger
Applio can do this.
Is it normal for it to take forever to launch from the bat?
hallow thistle
Is it normal for it to take forever to launch from the bat?
If it takes too long to launch, even with faster CPU, it's not normal.
turbid tiger
It loaded eventually, cmds are weird sometimes.
hallow thistle
Can you screenshot that?
turbid tiger
There's nothing to screenshot, it loaded.
hallow thistle
Really? That's fine.
turbid tiger
Where do you put weights/logs for this btw
hallow thistle
Where do you put weights/logs for this btw
turbid tiger
ah, aight
radiant quartz
o/
i was on my way to train a model using google colab, but two combined datasets have 12 and 15 minutes. however, when i'm training using "rvc v2 disconnected", it keeps on writing
TF-TRT Warning: Could not find TensorRT
what is the recommended setting for training a good model?
(can't send the dataset in question because copyrekt moment. total of these voicelines when separated are 500 files)
carmine siren
-colab
analog onyx
rustic slate
how to configure work for a video card?
hallow thistle
how to configure work for a video card?
RVC or W-Okada?
rustic slate
1 min
hallow thistle
Sorry, but I don't a direct message from random user if nothing private. To talk about W-Okada, go to #🔍│help-w-okada.
rustic slate
ok ok
carmine siren
i am getting this error while making cover
carmine siren
i am getting this error while making cover
is there anyone who can guide me to fix this error
rough plinth
2025-01-11 10:58:57.4020271 [E:onnxruntime:, sequential executor.cc:514 onnxruntime:: ExecuteKernel] Non-zero status code returned while running Pad node. Name:'/rmvpe/mel extractor/Pad' Status Message: CUDA error cudaErrorNoKernel ImageForDevice:no kernel image is available for execution on the device
have solution?
quasi ginkgo
How can i use these models for text to speech locally?
hallow thistle
How can i use these models for text to speech locally?
Applio can do RVC TTS locally.
hallow thistle
can i get the github link
-rvc
quasi ginkgo
great thank you!
brittle wing
@tight ether I keep getting errors when trying to train with KLM x4
tight ether
<@245213101153189890> I keep getting errors when trying to train with KLM x4
errors?
tight ether
does the error occur during training?
brittle wing
does the error occur during training?
I can't train
tight ether
It seems like it might not recognize multiple speakers... I haven't tested it on disconnected, so I'm not sure.
brittle wing
Yesterday worked but I couldn't resume
tight ether
huh
brittle wing
It seems like it might not recognize multiple speakers... I haven't tested it on...
What
I should use Applio or what
hallow thistle
Where do i place the cpkt and pth files in applio
Ckpt and pth files? RVC voice model only use pth and index.
tight ether
I should use Applio or what
yeah, try it on applio or codename's Fork.
brittle wing
yeah, try it on applio or codename's Fork.
Uh seriously... RVC Disconnected is outdated then?
hallow thistle
I'm not sure about RVC Disconnected, but "RVC GUI" fork is indeed outdated.
hallow thistle
so where i do place the files
tight ether
not really, but we can't know all the updates the developers made, so it’s safer to use a fork that has already been tested.
quasi ginkgo
okay thank you!
tight ether
x4 was trained with lastest ver of applio so.. yeah
brittle wing
x4 was trained with lastest ver of applio so.. yeah
Oh okay
azure marshBOT
<@155030383648440320> sorry for ping bud, but can you help?
hallow thistle
!howtoask
brittle wing
<@155030383648440320> sorry for ping bud, but can you help?
Okay then!How do I resume w applio???
-guides
brittle wing
How do I give the dataset path?
brittle wing
not really, but we can't know all the updates the developers made, so it’s safer...
What's the right dataset path and how do I resume
tight ether
What's the right dataset path and how do I resume
are you going to train it locally?
brittle wing
are you going to train it locally?
On Applio Colab
tight ether
colab?
brittle wing
Yes
https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb?authuser=2#scrollTo=iwDUiyus9F3l there's a training tab but how do I set dataset path & how do I resume???
tight ether
I've never used colab before. 
brittle wing
Uhhh
knotty moth
you should train with either og pretrain or KLM using applio
brittle wing
you should train with either og pretrain or KLM using applio
How do I resume on Applio and how do I set the correct dataset path?
knotty moth
How do I resume on Applio and how do I set the correct dataset path?
resuming from the different application might break the model, better start over the model within applio
brittle wing
resuming from the different application might break the model, better start over...
How do I resume with applio I asked
I still haven't started training
Also I asked how ro I set the correct dataset path
knotty moth
How do I resume with applio I asked
the guide doesn't explain it?
brittle wing
the guide doesn't explain it?
No
brittle wing
you haven't read it further
Which guide is that???
Can you link the guide
Also how do I set the right dataset path
radiant quartz
o/
i was on my way to train a model using google colab, but two combined datase...
ditto
i have read the guide but rvc v2 disconnected does not work not because it exceeded time limit
brittle wing
-colab
radiant quartz
(well, dataset path was even right but seems like it doesn't work.)
i was lost about how many good epochs to train for that large dataset
12 and 15 minute dataset; then it's 28 minutes in total
simple ore
<@155030383648440320> sorry for ping bud, but can you help?
it is a skill problem
aka not running feature extraction
@brittle wing make sure you're not skippig steps
brittle wing
it is a skill problem
Hello how do I import the right dataset path on Applio colab also how do I resume training and ... Where do input the pretrain links on Applio Colab
brittle wing
<@456226577798135808> make sure you're not skippig steps
I ran everything necessary
Yesterday It worked idk what happened now
radiant quartz
🫠
tensorboard looks broken af
lilac raptor
can somebody help me out? Am trying to troll my friend that am a girl in an alt account , but he doesnt believe me until i say "hi im evelin and this is my voice" , can somebody help me out with an AI girl voice saying that 💀
simple ore
How do I resume with applio I asked
if you trained with applio, you just start training again with more epochs
if you trained with something else, you are breaking new horizons then
but dont worry, it is simple
brittle wing
if you trained with applio, you just start training again with more epochs
Mhm I'm used to RVC Disconnected but I only can inference w applio
simple ore
prepare dataset, extract features, then put d/g weights from rvc disconnected into the model's folder
brittle wing
prepare dataset, extract features, then put d/g weights from rvc disconnected in...
I'm asking how.do.i train w applio where do.i.put the.pretrain links I checked in the training tab...
simple ore
applio colab or what?
ui/no ui?
pretrain is a starting set of weights, if you are resuming training, it is no longer needed - resume starts with D_xxxx.pth/G_xxxx.pth
brittle wing
Is this already overtrained I can't believe it
I mean
12 minutes data w KLM 4.3 x4 Pretrain
The other graphs decreased idk
???
What's this
radiant quartz
12 and 15 minute dataset; then it's 28 minutes in total
all files are 44000khz but is 40k good enough?
brittle wing
all files are 44000khz but is 40k good enough?
Uh yes
radiant quartz
nice, thanks
i think i'll spend the whole night finding my way to train a model with that large set
radiant quartz
😭
but tensorboard says it's all blank
censored the name because it's for my oc/private purposes
hallow thistle
all files are 44000khz but is 40k good enough?
44000Khz? That's so many number. I've only heard of 40000Hz and 48000Hz being used to train a voice model.
radiant quartz
it's fine
tame mica
but tensorboard says it's all blank
refresh
radiant quartz
(i don't have time then i have 1 hr and 40 minutes)
i'll make a help forum post for this tomorrow
hallow thistle
radiant quartz
praying for my model tho
i have 500 files of dataset
then i separated it into 250
odd shale
i have 500 files of dataset
How much minutes on your dataset?
Have you cleaned it properly?
brittle wing
Have you cleaned it properly?
Hv I overtrained
radiant quartz
How much minutes on your dataset?
file 1: 12 minutes
file 2: 15 minutes
pinging me is fine so i need attention (busy btw)
brittle wing
radiant quartz
file 1: 12 minutes
file 2: 15 minutes
pinging me is fine so i need attention (b...
not all of them because some voices have muffled ones, during a call (no bass if i do equalizer), and the bleeps (yep, it's from a visual novel, if you ask) which i thoroughly curated it
radiant quartz
Have you cleaned it properly?
no
brittle wing
Hi I was just asking if my model is finished
radiant quartz
Welp.... rippy rippy.
i might clean those with a better start so i need to clear it
odd shale
With a bit of luck the model will may come out decent anyway if these noises and sfx aren't too present on the dataset
radiant quartz
how many epochs for 12 to 15 minutes?
i asked for a friend, then he told me like 450-500 epochs
brittle wing
knotty moth
i asked for a friend, then he told me like 450-500 epochs
see the tensorboard
brittle wing
Sir, I'm sorry to disturb but
brittle wing
see the tensorboard
Can you tell me
odd shale
i asked for a friend, then he told me like 450-500 epochs
The final epoch amount will depend on the tensorboard.
brittle wing
The final epoch amount will depend on the tensorboard.
Exactly.
knotty moth
Hello Makiiiiiii
radiant quartz
i did 128 instead of 64 in crepe hop length
knotty moth
Can you tell me
you should also test several interesting model checkpoints
odd shale
i did 128 instead of 64 in crepe hop length
Have you used rmvpe or mangio crepe on pitch extration?
radiant quartz
Have you used rmvpe or mangio crepe on pitch extration?
usually rmvpe and i don't touch others
brittle wing
you should also test several interesting model checkpoints
Mhm yes I'm going to rn but what can you tell me by looking at the graph
knotty moth
usually rmvpe and i don't touch others
mangio crepe uses the hop length, not rmvpe
radiant quartz
oh
analog obsidian
i did 128 instead of 64 in crepe hop length
in my testings crepe hop length 128 is very similar to rmvpe, almost exactly the same
tho if you want to use crepe use a hop length of 64 since thats when its better than rmvpe in terms of accuracy
tame mica
The final epoch amount will depend on the tensorboard.
they said their tensorboard wasnt working iinm :p
analog obsidian
but has 0 effects if you used rmvpe
odd shale
Mhm yes I'm going to rn but what can you tell me by looking at the graph
I can tell due to the graph that your dataset wasnt too clean.
odd shale
they said their tensorboard wasnt working iinm :p
What
brittle wing
I can tell due to the graph that your dataset wasnt too clean.
Uh but is the model overtrained or what
radiant quartz
mangio crepe uses the hop length, not rmvpe
ah, appreciate it
lyery's reply:
will take it lightly, so i'm taking notes for that
brittle wing
Nah I used a pretrain
tame mica
but tensorboard says it's all blank
this
brittle wing
I can tell due to the graph that your dataset wasnt too clean.
It's quite clean as can be.
odd shale
they said their tensorboard wasnt working iinm :p
Welp, i use kaggle Applio so tensorboard so that's not my problem.
odd shale
I'm asking of the model is finished or not
The model is overtraining if the g/total graph goes up
brittle wing
Also there's almost no noise
analog obsidian
ah, appreciate it
lyery's reply:
will take it lightly, so i'm taking notes for ...
keep in mind if you use crepe for training your dataset has to be very clean because crepe is not robust as rmvpe and handles noise horribly
brittle wing
The model is overtraining if the g/total graph goes up
Still needs training?
radiant quartz
in my testings crepe hop length 128 is very similar to rmvpe, almost exactly the...
i expected some high-quality output when i use my voice (i have my good mic, so dw) so here i am, training this
edit: the expected output has some accurate breathing since it has various evidences in a dataset (ye ifkyk if you're into visual novels 👀)
brittle wing
Wdym
odd shale
Also there's almost no noise
And have you labeled the dataset properly?
tame mica
Welp, i use kaggle Applio so tensorboard so that's not my problem.
i mean if they cant see the tensoreboard how are they gonna see if it's ovt xd
brittle wing
And have you labeled the dataset properly?
Wdym by labeled
odd shale
Still needs training?
I'm not sure, but you can test various checkpoints (pth's) from your model too.
brittle wing
I'm not sure, but you can test various checkpoints (pth's) from your model too.
Not sure
What-
analog obsidian
i expected some high-quality output when i use my voice (i have my good mic, so ...
Which batch size? Did you cleaned the dataset?
odd shale
Wdym by labeled
I mean, applied audio labeling on Audacity.
analog obsidian
if you dont want to clean the dataset use rmvpe for training
radiant quartz
Which batch size? Did you cleaned the dataset?
8 since i'm new to rvc disconnected
edit: total dataset duration when combining 2 files according to audacity is 28 mins 30 secs
analog obsidian
**8** since i'm new to rvc disconnected
edit: total dataset duration when combi...
thats fine
brittle wing
I mean, applied audio labeling on Audacity.
Huh what I only used Melroformer denoise also I don't have a computer
odd shale
Huh what I only used Melroformer denoise also I don't have a computer
Hm, it's fine then.
analog obsidian
**8** since i'm new to rvc disconnected
edit: total dataset duration when combi...
8 is the perfect balance between generalization (model capability of generating audio in this case) and stability
odd shale
8 is the perfect balance between generalization (model capability of generating ...
Jajaj el 8 en batch es mi favorito.
analog obsidian
tho one thing worth mentioning is that if you really want to the model to sound more like the dataset you can slightly increase it but this comes with the problem of the model having weird pronunciation
bc loses generalization
brittle wing
Hm, it's fine then.
Is my model overtrained or not...
radiant quartz
i need to finalize the pitch extraction tho
if i'm new to training models, without cleaning it (like only the dataset.zip placed aside experiment folder), can i use rmvpe?
analog obsidian
if i'm new to training models, without cleaning it (like only the dataset.zip pl...
Yup rmvpe batch size 8
Safest options
radiant quartz
if i'm new to training models, without cleaning it (like only the dataset.zip pl...
rvcDisconnected
ocname_bundled.zip
rmvpe, batch size 8
folder structure for tomorrow's training
how many recommended crepe hop length (scrolling back, is it 64?)
odd shale
Is my model overtrained or not...
Nope. ._.
analog obsidian
how many recommended crepe hop length (scrolling back, is it 64?)
for rmvpe it doesn’t do anything
only affects crepe
radiant quartz
i thought it's mangio crepe
analog obsidian
i thought it's mangio crepe
mangio crepe is the same as crepe
knotty moth
how many recommended crepe hop length (scrolling back, is it 64?)
again use mangio crepe if you want to use it
brittle wing
Then why is the line going like...
analog obsidian
If u want to use mangio set a hop length of 64
brittle wing
I can't see the screenshot properly but it doesn't seem like it's overtraining.
It's going down after 4.5 steps!
Look
Maybe after 150 epochs it will overtrain
radiant quartz
150 epochs in my experience: 20% sloppy
450 in my opinion: 95% accurate since it's completely becoming human because of pronunciation
odd shale
150 epochs in my experience: 20% sloppy
450 in my opinion: 95% accurate since it...
Not always.
radiant quartz
might train tomorrow, thank u all for the advice 
odd shale
150 epochs in my experience: 20% sloppy
450 in my opinion: 95% accurate since it...
Not always, because all will depend on the dataset, the voice you wanna train, and training settings.
So it will be always random.
rustic arch
analog obsidian
Started going up after 5k!!!
it has to rise forever
i usually wait 1 hour to confirm it stopped improving
overtraining is pretty obvious when you see it in a graph
brittle wing
overtraining is pretty obvious when you see it in a graph
Really are you sure
analog obsidian
Really are you sure
yuh, it just goes up forever
brittle wing
yuh, it just goes up forever
Wdym
I tested it at 150 and it already sounds like the person
brittle wing
graph gonna end up looking like this
It already looks like this...
After 170-180
I downloaded the 170e weight
analog obsidian
if ur happy with the result then stops the training
obsidian niche
yo guys, im new to this, whats the issue with this where it keeps glitchintg?
brittle wing
if ur happy with the result then stops the training
Mhm it sounds like the target person not that I'm happy w the results
analog obsidian
yo guys, im new to this, whats the issue with this where it keeps glitchintg?
#🔍│help-w-okada for realtime/w-okada help
obsidian niche
im using rvc
low shard
im using rvc
are you talking about realtime voice changer? if so, then that's not RVC, but wokada
analog obsidian
im using rvc
be sure to clean the inference audio, remove reverb and any other sound effects
don't inference audio with harmonies
obsidian niche
be sure to clean the inference audio, remove reverb and any other sound effects
is it possible to remove the inference and reverb with UVC?
i want to get minecraft steve to sing Die With A Smile
There's alot of harmonies (i think thats what it is) in that song
analog obsidian
is it possible to remove the inference and reverb with UVC?
Yup use mel karaoke to remove harmonies
obsidian niche
Yup use mel karaoke to remove harmonies
alr tysm
sand path
whats the best sample rate between 32,40 and 48k for the most accurate to use with voicechanger, with rvc2 and a dataset of about 35 mins (i just know some ai knowledge from about 1 year ago or more) and i use The Mangio-RVC-Fork so i dont know if its giving me disadvantages.
and should i turn pitch guidance of or on if im using a talking dataset
brittle wing
Yup use mel karaoke to remove harmonies
How do I filter my samples for inference?
craggy brook
How can I easily find the voice dialogues used by a game character? In tones and emotions
brittle wing
Overtraining?
crude flame
Overtraining?
It's going down so no
brittle wing
It's going down so no
Huh
Thank you for finally accepting me
It's a short dataset
crude flame
1:30 minutes w titan pretrain
Wow that's super short
brittle wing
Wow that's super short
And is it overtrained.
brittle wing
Wow that's super short
It starts to concern me when the like goes like this:
brittle wing
Do you have the avg graphs?
Avg graphs?
brittle wing
Is this applio or codenames fork?
RVC Disconnected