Realtime AI Voice Changer without a Graphics Card - Google Collab & RVC
HTML-код
- Опубликовано: 23 авг 2024
- Links referenced in the video:
W-Okada GitHub - github.com/w-o...
RVC Training Playlist - • RVC (Retrieval-based V...
AI Hub for AI Voice Models - • How to Get AI Voice Mo...
Optimizing settings in the Voice Changer - • Optimizing Settings in...
Hardware for my PC:
Graphics Card - amzn.to/3pcREux
CPU - amzn.to/43O66Ir
Cooler - amzn.to/3p98TwX
RAM - amzn.to/3NBAsIq
SSD Storage - amzn.to/42NgMFR
Power Supply (PSU) - amzn.to/430bIhy
PC Case - amzn.to/447499T
Mother Board - amzn.to/3CziMXI
Alternative prebuilds to my PC:
Corsair Vengeance i7400 - amzn.to/3p64r22
MSI MPG Velox - amzn.to/42MnJHl
Cheapest and PC recommended:
Cyberpower 3060 - amzn.to/3XjtZoP
Come join The Learning Journey!
Discord - / discord
Github - github.com/Jar...
TikTok - / jarodsjourney
If you found anything helpful, please consider supporting me and the content I am trying to produce!
www.buymeacoff...
There is a currently an issue with pyworld not building and won't allow for the voice changer to work. This was brought to my attention and is an issue as of 9/11/2023.
*The fix, found from AI hub:*
After this line in the code:
print("\033[92mSuccessfully cloned the repository")
Copy and paste this line:
!sed -i 's/pyworld==0.3.3/pyworld==0.3.4/' requirements.txt
Together, it should look like this:
print("\033[92mSuccessfully cloned the repository")
!sed -i 's/pyworld==0.3.3/pyworld==0.3.4/' requirements.txt
I saw some one link it in the ai hub discord server that has a fix in it
@@RuizzziuR do you know whereby in the server this link is? Is it applio?
@@RuizzziuR Thanks mate 👍
For me it works fine but then nukes itself after a while. Maybe a connection issue?
Can you help please with error ERR_NGROK_3200 after some time of using collab?
OMG, thanks for this one and sharing information on how to make the technology accesible for those who can't afford expensive gear.
you can do 112 chunk + 16384 extra, its bc colab has 2cpu cores that running high extra ruins performance a bit, if you change advanced settings putting it on rest, 0.2 start 0.8 end, and turning silencefront off also improves it a lot more too
My collaboraty always off after 1 hours, do uou knownwhat problem?
@@habibahmad654 google Colab free will always kick you out after s certain time I think. Well because it's free
@@habibahmad654You have to pay Colab Pro
@@habibahmad654 because its the free version its normal lol
Better settings, for FAST and AMAZING results:
If you're using a index: f0: RMVPE_ONNX | Chunk: 112 or higher | Extra: 8192
If you're not using a index: f0: RMVPE_ONNX | Chunk: 96 or higher | Extra: 16384
Powerful and cheap, thank you very much😊
It works, but after like 5 min the ngrok server gets crashed by disconnecting the runtime all because it says that it was running a not allowed code, the problem is that to make some video about, you'll have to repeat the same process like 100 times.
rvc code is banned on google collab
@@dartsgame_ Wait, don't tell me... are you kidding?!
nope
Unfortunately@@WindowsFan2006
i got that error when I edit pitch or extra options
@@dartsgame_wait so this whole video doesn’t work?
I was already asking you for help in another video you posted about this. And the video is here, thank you so much for everything😂🎉
Thank you for giving us this fantastic gogle
collab, I ask for that. I'm using it and loving it, I hope it never stops working, as it is very useful
after 10 mins of use its turns off and says its against their free use policy or something and to use this i need to use paid version of colab
Yeah, googles recently been banning a lot of these open source projects because of all the compute getting used up. No way around it :(
rvc code is banned on google collab
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
chex 0.1.86 requires numpy>=1.24.1, but you have numpy 1.23.5 which is incompatible.
pandas-stubs 2.0.3.230814 requires numpy>=1.25.0; python_version >= "3.9", but you have numpy 1.23.5 which is incompatible.
torchtext 0.18.0 requires torch>=2.3.0, but you have torch 2.0.1 which is incompatible.
torchvision 0.18.0+cu121 requires torch==2.3.0, but you have torch 2.0.1 which is incompatible.
how to fix that?
Hi. Can someone help me? After 5-10 min of using I see this error ERR_NGROK_3200, and if I reload page I see same. I did all by your guide(
thanks, finally my gpu can breath when combining with other ai.
Hey I had a couple questions I wanted to ask to you personally since your videos were very helpful, clear and understandable.
First of all I saw that some voices were more comfortable in a language, often english in the discord hub I am taking them. I wanted to ask if there’s a way to use an English voice in my native language (french) or if it’s going to sound awkward.
And second of all I have a bit of trouble understanding what « Epoch » is, Is it the more the better?
Thanks a lot for your help and time answering this comment
Eh, so the closest thing I can think of is it's learning the phonemes (sounds) of a language. The more overlap of phonemes there are between languages, the more natural it kinda is. But there may be an accent to it. When you use the voice, index is what changes the "accent". This is just my experience with Japanese.
Finishing one epoch is equivalent to the training having gone through all of the audio files. There's a lot more that goes into it, but you might wanna check out my video on training better models for further explanations.
@@Jarods_Journey Thanks a lot for your time and answers I appreciate 😄
Also from what I saw does the microphone plays a lot (people recommended me a dynamic cardioid mic instead of my crappy headset mic)
Will it work locally on gtx 1650
Jarod, you’re the best. Thank yuuuuuu!
finally thank you very much for this
Hey! I have tried this but it just doesn't work for me. I speak but I cannot hear anything, most I can hear is some random sounds but after that it doesn't work at all. What can I do?
I feel like the entire page already changed since this upload.
Hi end up getting it to work?
@@xnemessis ya, but i didnt think it was that great, thing is , if you re streaming , you want an app to seamlessly run in the background, i think the setup wasnt so great, and you needed to connect to a website, and it was very unhandy in usage, so i ditched it, i need an easy speech to speech voice changer to really sound like a real anime girl.
@@darky4555 I really appreciate your reply, I had the okada and tried running it on my pc without the server but it was just too choppy my gpu is old
so I looked into this method using google gpu, unfortunate you couldn't get it to run seamlessly, I think the technology is very new still and over time will be better,
I was still willing to give it a try using this new method, but I saw a few comments saying it's no longer free, which deters me even more
I'm rly glad that make a google collab version, my gpu can't handle running it for games, only work on diacord XD
its not worth it tho, i tried it and had a 20 second delay
@@drezzington rly? Have u check it through? I'll try it myself later
@@Bird..... yeah i already tried it a week ago or so, with the recommended settings too. it works without problems but the delay is just insanely long
@@drezzington thx for the info
@@drezzington I know what u mean now, the delay is unstable and the voiec is cracking
Frequent errors occur. Please check if the model of the framework being targeted is loaded.
I keep getting this pop up
Getting error: pipeline is not initialized, how to fix it?
is not an accepted origin. (further occurrences of this error will be logged with level INFO)
Hello, when I use collab, the most time-consuming connection is suddenly disconnected, and when I try to connect again, it does not work at all. This issue has been happening to me for several days, please give me a solution so that I can use it again.
Me too i think its no stable
bro. Its dont work here. and my interface is diferent
Hi , Jarod, from one mechatronics engineer to another, nice work appreciate it a lot , I wanted to ask is there any open source TTS model , I can use which gives near real time or quick inference , I need it for a project.
You might want to look into ViTs models as their inference time is much faster than transformer and diffusion based TTS's. Doesn't sound as good though.
Thanks , sound quality is not a problem for me right now @@Jarods_Journey
oh sweet! now I don't have to worry about the fact that I can't run vr chat and rvc at the same time without crashes
Jarod would you be willing to add the github or a link to the colabs that you talk about. That is always helpful. I didn't see the link so if it is there and I just had an oversight I apologize.
Ah, mb. I forgot to link them
hey Jarod , when i try download the respository it gives an error saying
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
chex 0.1.85 requires numpy>=1.24.1, but you have numpy 1.23.5 which is incompatible.
torchdata 0.7.0 requires torch==2.1.0, but you have torch 2.0.1 which is incompatible.
torchtext 0.16.0 requires torch==2.1.0, but you have torch 2.0.1 which is incompatible.
torchvision 0.16.0+cu121 requires torch==2.1.0, but you have torch 2.0.1 which is incompatible,
what can i do ?
yeah mine too
I have a issue that makes the program crash for some reason and it says *runtime disconeccted*. Any help?
Is there a way to use some other public web hosting other that ngrok? I think gradio does have some way to set up a temporary public address, right?
The local tunnel one is bugged😥
So every time I turn it on I have to do those steps again?
hi jarods, im just stumbled upon some trouble. i'm following your previous tutorial about train your own AI voice model. the problem arise when i'm already training it, when use slider to change the index to 0.6+ in RVC Client okada it seems the voice sound like two person talking in the same time.. but when im not using the index option or go 0.4 - lower it seems normal but without much difference.
do you perhaps have another update tutorial or some suggestion?
I'm not too sure on what's happening here, this seems more like a glitch or maybe a model issue. You might wanna open a github issue or check there to see if this has occured elsewhere
3:13 I love how you blurred out the authtoken, but only once (it says the same token on the bottom of the page 💀)
TYVM!!!!!❤❤❤
Thanks for sharing the info. Do you share your Marine's voice (10k)?
Np, unfortunately, I don't share my trained models.
I'm from Brazil, I have a lot of difficulties understanding English, I'm copying the images to try to do it, thank you
Can I use this for training a high-quality English voice model from my own English spoken records (recorded with my voice)?
I'm not a native English person, but speak it on a nice level. Thanks!
Bro i got dead serious question. Can you train a toddler voice? I mean 2 yo baby with only couple of word is it possible to train them?
If I can train a guitar, I'm sure you could if you wanted to.
@@Jarods_Journey no way.. You have the video on this channel?
Thanks for free knowledge
Thank you for this!
I don't get the option to connect to drive
Please I need help, I'm trying to find more natural deep voices on the discord models but I can only find a few like Corpse and Markiplier but I want more, just a natrual sounding one. Can someone help?
How would you get the voice changer from the browser into your Discord?
Output to a virtual audio cable (e.g. voicemeter) and input from there into discord as a mic
How much mbps where require to use this voice changer smoothly?
some of the cells are missing for me on colab
the "[Optional] Setup/Start Google Drive" is missing, is there any ways to make it available everytime without redownloading it again ? thanks in advance !
i can't see too
Hello I'm a Mandarin speaker and I wanna use Japanese model in AI voice changer, how can I train those Finished-Japanese Models with my own voice in order to pronounce more clearly.
Hi jarod, I'm from indonesia... I'm really thankful to you, but I found a little problem... when I choose RVC model suddenly the page becomes white with no buttons anymore, and can't even change the sound either, do you have a solution?
Does this cost money??
How many compute units does this use per hour? on colab
I need some help here, idk if this is the okada's issue or my mic issue but the voice cuts in the end of my phrases, i can't sing when all my stuff gets cut
Hi just wondering how much time does a vits training actually take . . . For ljspeech kind of dataset, with a single v100 card
It seems that they recently made this method paid. Are there any similar free alternatives to this one?
Unfortunately, a lot of companies are paywalling their compute as projects like these get more popular. Will update with alternatives if something occurs
my rvc is picking up desktop sound . I use 3.5mm 1 hole jack in laptop pls make a video about it.Stereo mix and playback is off
Hi I installed this and did same setup as you did. But I am trying to use it on Google voice app. But it doesn’t work. Please hell
Is it possible for my NEC laptop? I rather want to install it offline there while I download it from my phone to transfer it there
for some reason there is showing only cpu in the voice client, earlier it was showing tesla t4.
wht should i do to fix it, with cpu i cant get properly voice
on 3:15 you showed the token, fix this
Oops lol, eyes, what even are those for...
Yo. I just tested this and it works pretty well.
can you please tutorial video on how to separate AI Voice & Real Voice with Voice Meter Pro Software?!
Hi, where do I find male voices? this aihub server is full! There is no way to enter and download the templates
Do you have a guide for training with collab? I just don't quite get it
So It needs a great internet speed right? Im assuming this will affect my ping while gaming
Apparently it needs great "wifi" 😂. It's just streaming audio, not that different from Spotify etc. The difference is it's bidirectional - your voice is uploading whilst the generated voice is downloading. If you can stream RUclips videos in 720p without buffering, you'll be fine. Most people won't notice it at all - latency ("ping") or otherwise.
@@whitey4986 well i tested it , and it wasn't really good thb , and yes i can even stream RUclips in 4k without having problem , im not in 90s bro , so it either it needs an even better wifi or it is what it is ^~^
For some reason I can't upload models to the colab version. I select the RVC option, the .pth and .index files, but clicking upload doesn't do anything. I've tried waiting for a large amount of time (though the .pth file is only 52mb), refreshing the page and restarting the runtime. The upload button doesn't even change to show the upload progress like it did in the video. Also tried uploading only the .pth and no .index. Does anyone have a solution for this?
Same here i cant upload
are the graphics card powererd ones better than the ones without? or does it not matter? I have a rtx 4090 so if graphics card makes it better I will try that 😂
Graphics power is always better in this instance. I don't think the collab works anymore unless you have pro, but since you have a 4090, it'd be better for you to just run locally.
@@Jarods_Journey ayy thank you so much bro
Is it necessary to have an external GPU for swapping the voice?
yo how do i fix error: subprocess-exited-with-error on the cloning of repository, its been occuring after ive used it once
why i cant log in in ngrok, the page of sign in and log in are down, ERR_CONNECTION_TIMED_OUT
hello the current website doesn't have the first have the always use colab gpu part of the video
just asking i have good internet but it takes 5 hours to download it on hugging face is there anyway i can fix it
Just an update. As of now you have to have a paid account to use the voice changer for more than a few minutes. So free tier is sol atm.
hey i have a question, iknow that using rvc voicechangers is bannd in colab, but if I get a paid version, could i use it without being disconnected?
Yes, it should work, as so I've heard.
I have a question, I saw this in another channel and I have been looking for a solution to a problem in several places, it happens that when you are using the voices of nothing google colab stops working and I have to recharge everything, but the bad thing is that it secede every 5 minutes! Please I need a solution 😓😭
did you ever find a solution to this?
What is the minimum nVidia GPU recommended to make AI-trained voice models at home on the desktop?
(AMD is out for me, because lack of development, there is much wider support for CUDA & Tensor cores)
On average, what time does it take to train a model with a recording of 10-15 minutes long?
Minimum I recommend is 3060 12gb. It's the cheapest used option and will allow you to run most AI tools as long as your fine with it just processing slower
so do we have to re run everything on collab everytime we want to use it...
Hello everyone, I'm having a problem. W-Okada doesn't recognize my AMD GPU, does anyone know how to fix it?
tysm :D
How well do you think a 1050 ti would handle the voice changer?
pretty well if you optimize it enough :)
Is the cuality of the model worst with this way?
Not by much, but you may get choppiness or lag due to it being online.
it dossnt work outside of google as ling as im in google it works is it possible to fix
Hey bro, I own an AMD GPU. How to install it locally? I checked the other video, but that’s for NVIDIA GPUs
I also have an AMD GPU, but the problem i get is the "code not allowed" error.
@@WindowsFan2006 i managed to get mine up and running.
Unfortunately, I can't find the guide I used anymore. Seem to come from Github
@@WindowsFan2006 managed to find an alternative(?) Or maybe it's the same, but search how to install stable Diffusion on AMD and click on the GITHUB search result
I cant change the gpu setting to a use a gpu? I will send ss if needed.
which headphones or mic you are using?
Sony xm4 headphones and behringer xm8500 mic with a xenyx q802U sound board
Where is the model for marine rvc?
Haha! Hi, Jarod! 👋😊
could you guide me about how to manage my own voice to swapping another language? Ive also sent an email
to you Jarods.
Thanks for your shares...
I have blutooth headphones can I give input and output same? For Microphone and To listen output plz reply
Yes, just make sure they're connected before you open the voice changer
Is there a 32-bit version of this application?
is it correct that it doesnt save the modules and you have to redownload it everytime?
Correct
can I use it on discord then?
how the frick you change it from cpu to gpu, help fr mine wont show up my gpu its annoying
is AMD GPU still not supported at the app?
how do i fix connection refused when join the ngrok link btw
Free google collab is just like 5 minutes
okay real quick question i just need to train my own model FROM RVC correct that RVC?
Correct, RVC models are the ones compatible.
@@Jarods_Journey why the colab keep disconnecting me saying there codes that weren't allowed and my free computing has run out.
Hi there, I want to sound like a female in my videos using this AI Voice Changer.
I only have an Astro A50 headset. I don't have a separate microphone like you.
I use OBS Studio to record my videos.
Are you able to tell me what settings to use on this program and OBS so that I don't hear myself in real-time, and also so that I only use the AI voice once I hit the record/stream button please?
At the moment I hear myself first through my headphones, and then a second later I hear the ai voice I have chosen. I don't want to hear myself.
Or do I need a microphone like yours to do this properly? I want to be a Vtuber. I can buy a Shure SM7B if needed, I have the funds to.
Your microphone will likely work fine, although the quality will be better with a better one. The SM7B is probably a waste of money, you could get a decent USB (Rode, Yeti, etc) and a pop filter and use the rest to upgrade your computer.
Delete your microphone as an audio input in OBS. Get a virtual audio cable program (Voicemeeter Bannana is probably simplest) and add one of its virtual outputs to OBS. Have the RVC input from your mic, and output to a virtual input, then route this virtual input to the virtual output you sent to OBS earlier.
The hearing yourself is likely due to listen mode being enabled in OBS - in the audio panels settings you can switch this off. If you deleted your mic, you likely won't hear it anyway.
Thanks, i'll give it a try.@@whitey4986
the sign up of ngrok page can't be reach
yay, finaly i can prank my friends on discord)
this work on mobile:(?
Does not work with amd gpu?