Optimizing Settings in AI Voice Changer Client
ฝัง
- เผยแพร่เมื่อ 24 มิ.ย. 2023
- Links referenced in the video:
w-okada's github repo: github.com/w-okada/voice-changer
AI Voice Changer Client Installation: • Realtime AI Voice Chan...
Get it working in Discord: • How to Connect AI Voic...
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/3NBAsIq
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 PC recommended:
Cyberpower 3060 - amzn.to/3XjtZoP
Come join The Learning Journey!
Discord - / discord
Github - github.com/JarodMica
TikTok - / jarodsjourney
If you found anything helpful, please consider supporting me and the content I am trying to produce!
www.buymeacoffee.com/jarodsjo... - วิทยาศาสตร์และเทคโนโลยี
![НЕЙРОСЕТЬ учится БЕГАТЬ (Создатель Существ) [Перевод Code Bullet]](/img/n.gif)
Great video. thank you
I'm one of the VCClient and RVC contributors. There are some additions to the content of the video.
Regarding the difference between the f0 estimator harvest and crepe, in addition to the sound quality, harvest uses a CPU and crepe uses a GPU. Crepe can improve latency if you have a good GPU.
In sever mode you can choose the sound driver. VCClient measures latency within VCClient, but additional latency is added when connecting to other devices.
Besides MME, WASAPI and ASIO can be selected, so if you can use them, I recommend using them.
For the protect item in advanced options, if protect is set to less than 0.5, the ratio of retrieved features will be reduced in cases where f0 estimation is unsuccessful (silence or breath sounds).
I've seen, appreciate your work and thank you for the additional information!
Would be better if next to every option will be ? icon when on hover you will see popup with explanation. It will help a lot.
Yo app just mines BTC stop the cap.
@@hired_meo proof?
@@paradym777 My Premium Kaspersky version 😀
From a musician experience: if you have ASIO supporting soundcard - use ASIO instead of MME. It decreases the audio delay provided by audio tract (e.g. on my PC guitar/mic recording delay for 1024 samples chunk is 180ms for standard MME, and 14ms for ASIO). Theoretically WASAPI can also work fast however I don't have WASAPI supported hardware.
It stutters for me when use my asio soundcard or just doesnt work at all
@@realxdey same here
Where do I change the settings for asio?
ASIO4All an any other ASIO doesnt work.
Thanks for all the help! You have responded to all of the comments and provided everything Ive needed. Anyways keep up the great work and keep doing what you are doing 👍
I'm so lost. My window is just empty when I click on the native client exe...
Thank you so much for making this video. I have a RTX 3090 and was wondering why no matter how much I messed with the Extra and Chunk settings, the voice still sounded distorted, but it was probably because I had the pitch too high where the voice probably wasn’t trained (I was using above 20 pitch) I didn’t know that most voices are trained no higher than 12. I’ll mess around more with this program tomorrow with what I learned from this video.
Tech is getting so cool, great video!
This worked so great on my first Voice Changing experience. Other videos on your channel are also great! Thank you very much.
Great video as always glad you went through everything with a good explanation for everything! Keep up the great work, and i am excited for what will come after RVC!
hearing senchou speaks english just feels weird, not in a bad way it's just feels like im hearing something im not supposed to hear in my entire life
i literally just upgraded from my 1050ti to a 4070 today just to use this + other AI tools. love these tutorials
GZ! Very solid card right there :D
i'm probably going to buy an rtx 4090 for AI stuff too. otherwise i'd buy a 7900xtx. it costs half as much, has the same amount of vram and the performance of a 4080. too bad amd sucks with these things.
man im using intel HD card 💀, will it work?💀
I'm still using rtx 2060
I'm using 920mx😀
you're a legend!! insane video quality and tutorial, can't believe i found pure gold at 4 am.
i guess youtube can also be a chad and recommend really good content wow
Great video! I'm glad you showed what this program is capable of on a 4090. It seems we're not quite there yet with AI voices. I wonder if this is a small hurdle that will be overcome soon or a insurmountable mountain like hands are to AI art.
Ah, actually some AI voice models that I've tried are actually pretty scary accurate, meaning my models need a bit more training lol. I would say we're not that far away from indistinguishable voices
Great! can't wait to try this out to see if it improves performance. If it doesn't I might just install windows 11 to match your settings exactly.
Thanks, I really need to try this one.
If anyone has issues exporting an ONNX file and getting an error message in the GUI (it usually just says error message: no error message), but if you check in the console it says that pytorch has tried to allocate VRAM and has failed. A quick workaround for this that worked for me was changing in the GUI to use my CPU instead of my GPU and then exporting the ONNX worked. Afterwards you can change it back to your GPU.
Smart, didn't even cross my mind xd
i still get an error in the GUI..
if it helps,this is what is shown on the console : "[Voice Changer] get_onnx ex: Can't get source for . TorchScript requires source access in order to carry out compilation, make sure original .py files are available."
yooooooooo thank you so much
What do you mean changing the gui?
@@Nebularban changing the setting in the GUI
Very helpful 👌 . Just where can you get models?
i have to figure out a way to offload the processing to my laptop so it doesn't bug out when playing demanding games. is there a way to access it from a lan client?
ps: i thought the directml version worked for my 6900xt, but it's processing on the cpu. but, on the plus side, converting to omnx literally halves the load on the cpu. when running this exclusively, i managed to get a total lag of 400ms buffer + processing. not bad for an octacore cpu (5800x3d)
question, how to train? what does train do?
i followed all previous tutorial and the voice output only sounds distorted repetitive. im using ryzen 3600x with gtx 1050ti 16gb ram.
If you've already trained, you might need better audio samples, more samples, or adjust settings to better fit your system via chunks/extra
Hear Botan/Marine speaking clear English is kind of weird&awesome at the same time XD
Using a RTX 3070 with:
Chunk 256
Extra 131072
Sounds perfect on these even with the half second delay!
Hi Jarod. I'm trying to use my bluetooth mic as an input for voice but it is not showing in the input options! Any help in this regard will be appreciated!
Thanks for the video, but when I browse and select the onnx file in one of the slots, do I just leave the index blank? Because if I select the onnx file, then click upload and try to use the voice it doesn't work. Also, with piper it seems to require a json file as well but how is this generated? rvc doesn't give this to me.
@Jarods Journey do you have any resources for voice models, or do you train your own? All of your voice models sound much better than what I could find online. What settings do you use to train your models at?
I train all of my models using RVC v2. No real special settings, I just clean and curate my data so it's crystal clear input audio.
@@Jarods_Journey Trained a couple models already, came out great. Thanks for all the help between comments and videos.
HOW DO U CLEAN DATA@@Jarods_Journey
great video! keep it up!
Hi im the one on your discord server who created a ticket about the gl not loading and you said to change the chunk to max, i just switched the audio to server and it worked perfectly for me, i guess theres a weird thing going on with client option. Just wanted to put it out there
Damnn, the intro transition was goood
hey, is there any difference in method which rvc works here or in any other fork ?
i seem to have a problem with delay, theres a long delay before the voice starts working, is there a way i can fix it?
Hi! just wanted to ask, can I use this app to change my pre-recorded voice?
I hope they will make it in VST3 format so i can just put it on the daw track which my microphone is routed through. it would be so amazing holyy
Considering how calculation heavy this thing is it's unlikely.
I might try to make a VST3 client for this stuff.. but I couldn't find any API description on their page.
why cant you just use VAC and use VAC Line in as your input mic in your DAW?
@@PsychicType hmm ill try that ty!!!
Hey I tried to use this today but while my mic is on sometimes out of nowhere there is a robotic static.. is there any way to fix it?
Do you think we could get a download for the Marine voice?
Great video !
HI, I have an idea to use the RVC client on a karaoke party . I'm struggling to make the delay short enought to sing in real time. It's just not possible.
I bought top graphic card - Zotac RTX 4090 with 16500 cuda cores. The delay (even in passthru mode) is ca 0,3 sec and theres no way to get rid of it. Only sound driver that work is MME , but the latency is enormous.
I can't make it work (in server mode )with any of the ASIO drivers I have. It's mute. .I've Downoladed ASIO4All. Mute. Bought Focusrite 212 interface. Asio is dead. Realtek ASIO with built in sound card - doesnt work, as well. What I'm doing wrong ? \
My CPU is Intel i9 12900K
I have an AMD CPU with nvidia graphics card so idk if it matters which one I use right?
you saved my day !
Thank you Jard!!!
please do you have a fix for:
[Voice Changer] Pipeline is not initialized.
[Voice Changer] Waiting generate pipeline...
pls it would be much appreciated
Bro. You are too good . A wizard
I have an AMD RX 6950 XT gpu and a i9 12k processor, are these considered good specs for the voice changer?
Is the onnxgpu-cuda for NVidia gpu's or different gpu's? cuz i have a RTX4070ti i just want to make sure i download the right thing.
thanks s omuch helped a lot !`
one issue though, when i press start and i speak it works fine but then it cuts out and then comes back smooth again, then cuts out and then back. how do i fix?
I'm very sad, I've been trying to configure this for days, but my pc doesn't have any good gpu, only cpu, and I think that's why my voice is laggy too much, and it takes a long time to come out, I can't find a perfect cpu setting, please help me, because the voice doesn't even want to come out right, is just laggy 😢😢
i get some repeated when trying to hear it myself, the first time its fine and its nice to hear, the second time and the following is just distorted sound of the first one, how do i fix that?
Whenever i run it on my pc the voices sounds TOO robotic, training them would make a dif?
hi, is that possible to run the server voice changer on a rdp(windows server) and use it on our computer?
Hello sir, do I need to download any extra files prior to downloading these as shown in this video, or will everything work just fine when following the same process in this video? I have heard you talking about downloading RVC first or maybe I am mistaken.
Thank you.
You will need to download or train AI models for voices. It only comes with 4 preinstalled
Thank you, sir, for replying. But I am done downloading the software but I think the voice changes often say things that I didn't say and weird noises and twist echo. I have an AMD potato laptop with just 4gb vram and 32gb ram, in the gpu drawer I see just gpu on/off option no room for choosing between GPU or CPU not sure why that is like that on AMD played with it all day yet the software hasn't been able to say a single word of mine only making a weird noise and short utterances. Not sure though, do you think upgrading the AMD drive would make any much difference? Or maybe time to change to Nvidia.
Note. The laptop comes with a hybrid gpu AMD Rx / Vega 8. I think the Vega is interfering. 🤯
@@ivw1286I have a rented A100, a $10,000 GPU, and it still struggles a little. I don't think you meet the minimum requirements to even run the program.
Is there any way I could possibly hook this voice changer to something like discord? I think it'd be fun to mess around w/ friends of mine w/ a model!!!
Can you help me out? With when im recording I hear voices in the background? Even tho Im not even speaking
I assumed that the clear setting button at the top would clear out a slot for instance I've been testing a bunch of slots and I want to set them back to the default blank but apparently clear setting does not do what I thought it did in fact it seemed to screw up the whole program.
For High cpu usage problems....
1) Buy $10k PC setup. jk
2. Set index to 0. Even setting it to Index: 0.1 maxes out CPU
3) Set Extra to a lower amount. Higher Extra uses more CPU. Lower Extra uses more GPU. ( use 16384, any lower and doesn't decreases cpu usage, any higher and cpu usage doubles.
4) Use crepe_full (uses most gpu)
4) S.R. to 48000. (dont go beyond as echo forms.)
this works for amd gpu, i was getting crazy delays. followed this settings and now its only 2 seconds delay. thank you!
thx bro now im only getting a 1/2 second delay on amd gpu
Can you tell me the voice model and settings you used in the opening? please.
how to reduce the res? when I start my res it actually reaches more than 1000ms, I think that's what makes my voice not come out, is there a way to reduce it?
amd pcs dont work with this unfortunately, is there a way to use rvc real time cloning with colab? and use it in game voice comms, etc
You'd have to try to directml version, but yeah, lots of instabilities there I've hear. Can't run this specific one in colab due to it requiring an exe, but there is one built into RVC that I have yet to take a look at.
Thanks so much for explaining this! What parts of my computer would I need to upgrade to make this go faster?
GPU 100%
so whenever i boot it up and do all the settings and whatnot, it never changes my voice even if i follow every step, i can hear my voice yeah and ive already fiddled with the tune, yet nothing changes even if i turn it on
hi is this possible to work on laptop? i tried but it doesnt read my voice and all i hear are cracky sounds. is it bc i dont have gpu, what is the solution for this
Hey I'm little confused but i wnna install it so which vid of urs i should follow thx
do you have any idea what happened? The voice changer was running pretty well a couple weeks ago but it looks like they had an update and now all the voices seem slurred and choppy, it was perfectly fine a couple weeks ago and I've made no changes to the settings, i don't know what happened, it affected all my models
Not sure, this would only have happened if you had installed a newer version of it. I would recommend just use and stay with the version that worked for you previously as there seems to be issues using other versions
@@Jarods_Journey I’ll try and play around with it, does it remove your models if you install it again?
@@kongk5772 yup, you gotta start from scratch
@@Jarods_Journey so far I've redownloaded everything, and everything is up to date, I optimized my settings the same as the older versions but for some reason, the millisecond per response keeps stacking, all the way up to 30k ms per response all the while it's eating my CPU, this didn't happen before and I'm not sure why, do you have a fix for this?
which version is the one that works for you?
even at highest chunk i cant get it to sound good
hey wanted to ask where you can all that extra models ?
sometimes it works when i use an audio file (i have no gpu so its slow) but other times it just creates popping noise? it seems really inconsistent why is this?
Hello and nice video! i have a big problem with my output lagg (I hear my voice after few seconds like 10-12) i have buf:300ms and res:like 10k or something how can i fix that?
Sometimes it lags a bit, give it a minute or so and the res will go down. I turn off my mic and it speeds up the process tremendously
Anyone know why, when I use client, all my input and output options disappear? It was working fine, and then I just couldn't select them anymore.
Edit: I figured it out but for anyone with the same problem, Ill leave this here. It only occurs when you make a shortcut version of the start file and try to run it. The way to fix it is to open the original start file, and switch between server and client on one of the sample voices. If it doesn't work just repeat until your input and output options reappear.
my voice changer worked perfectly when i just installed it, but after reopening it, it started to give errors like "GL is not supported" and refuse to work. any suggestions how to fix this?
1650 Super and even 1060 are doing surprisingly fine using this tool.
Awesome to hear! I had great success on my 2070 super so I'm glad to hear the pascal cards are still chugging along
hello mind telling me what settings you are using? i have a 1660 super and its super laggy nothing is understandable
@@raidentatsunoko same, i have 1660, but voice very bad and laggy
@@soluckymoon sad, still same to me.
@@raidentatsunoko have youn found a fix?
Why did I download it, turn it on and go to the voice changer client, but the GPU section shows (cpu, gpu0, gpu1, gpu2, gpu3) while my computer has a Geforce RTV 3050 Laptop Gpu card?
there is no way to mute the output feedback, right?
so that, when you talk, you don't actually hear yourself back but still can be recorded or heard by other people....
hey, i just want to ask how I can use my trained RVC model for text to speech instead of real time? I tried using tortoise-tts but it says the model is unsupported
Different architectures, you need a tortoise model for tortoise
Great instructions! I saw you don't have any of the original models that come with the software loaded up anymore, how did you delete those? I can't seem to figure out how to get rid of them lol
You can just overwrite them by uploading new ones :)!
@@Jarods_Journey Gotcha, it didn't seem to work when I did that so maybe I'll try reinstalling or using an older version. I downloaded whatever the most recent version was as of yesterday (7/10/23) and the UI looked a bit different than yours too so maybe some changes the dev made broke overwriting. Thanks man, appreciate you!
Hello, I have a GPU but for some reason it’s not selectable in the processing option
is there another link for this, it says too many people have downloaded it yet or something
hey, whats is the reason to high res in okada and how i can fix it? is the mic reason?
When I use the output Digital Audio Cable driver and select it as my microphone in discord it works just fine except that when someone else speaks it creates an echo. Adjusting the tresh hold in the client does not help, anyone having any idea on how to fix it?
I've found using this in games causes your mic to cut out and stutter, as well as spiking CPU usage. I hope in the near future it's even more optimized ^^
It's to be expected. The results outside of games are when it's able to fully utilize your GPU.
While playing a game, the game is eating some amount of resources. Ecspecially so if the game isn't able to run at a steady framerate, which very well causes the voice client to have unreliable processing times.
Have you figure a solution for that yet? Bc recording or even ingame mic check, it sounds fine. But people ingame hear it stutter alot
I notice my voice cuts out a lot while speaking. So it misses a few words sometimes. Whats the best way to solve that?
I'm also having this issue, so I'm tagging on to your comment.
it feels like the voice changer is picking up too much background noises and saying things I didnt say is that normal?
can i use it while playing valorant or would i get any problems with vanguard (valorant anti cheat)
Cool video. Ive been wondering if i can use it to talk to other people as i dont really want people to hear my voice. Like in discord or something.
Mb i found the vid
@@AxooD1 what’s the title of the vid
Why is my voice changer says no embedder? When i changed into a custome voice it says failed idk why it says fail i cant even hear the voice changer
I find that the DirectML version (AMD) tends to randomly stop working and sometimes my GPU driver would crash, so that could use some more work to make it stable. And thanks for this guide! I barely know what each option really means.
Wait is your gpu recognized in the software?
@@jesuisunroi5543 It is, it's an RX 6900XT.
how do u get it to use ur gpu mine only uses the cpu even on the directml version 😭😭
@DesuVR wtf my rx 6800xt is not how did you do 💀
@@jesuisunroi5543 Honestly, I don't know if it's even using the GPU at all, I get ~50-80% CPU usage when RVC is active. I thought you meant if the software worked at all on AMD, my mistake. ^^;
However, I got it running okay-enough on my 5800X3D with these settings; INDEX: 0, F0 Det.: harvest, S.Thresh.: 0, CHUNK: 320, EXTRA: 4096
Hey , how i can fix if my input not appear my mic and the select vc button is disappear ?
😍 Thank You
Im having issues where everything comes out in short breaths or stutters, barely comprehensible, I have pretty good specs and I tinkered around with as many settings as I could but I always get the same result, how do I fix this??
Same
hii. I have an rx 580 and i installed the DirectML version. But on GPU section i have something like CPU-GPU0-GPU1-GPU2-GPU3. Software doesnt find my GPU probably. So the anime voice is with delay, low quality ect...
Can u help me? Maybe i downloaded the wrong versione for AMD GPU?
Where did you get your Marine model?
Hey there Jarod, I have a question regarding "Cuda Cores" / Cuda core compatible graphic cards.
If i get this straight : Your GPU must be a good one , i.e: a newer and faster one (That much is clear) but it also must have cuda cores, since all the models are cuda core based. So a 1080 ti which is a strong gaming card and a favourite of many gamers cant handle the ai voice changer program. The program defaults to CPU usage. Thats what happens on my end with a 1080 ti (which has no cuda cores). The solution would be to go with the OTHER download (what you called "for amd users".
Is my assessment correct?
CUDA cores are included in all modern GPUs actually. the software will default to CPU always. The benefit to having a higher end card would be the faster GDDR6 memory, larger amount of VRAM, and many other factors. I am not sure if RVC uses CUDA cores in any way, but a GTX 1080ti should still work, just increase the delay, run it through server, and have the audio encoder set to ASIO or WASAPI if you can.
Note : Only use the alternate AMD version if you have an AMD GPU, if you have an NVIDIA graphics card, use the standard download you already have. I tried out the software with my GTX 1650ti on my laptop and it worked pretty well, once set to a 5sec delay. Your GTX 1080ti should do far better then my mobile GPU.
@@calebpeters191 I have done exactly that, but for me the delay needs to be really high to sound any good. For some reason, the GPU still sits at only 30% at max. The sound is still robotic and just sounds like i am a 8 year old kid with a generic voice morpher.
Besides, sometimes (rarely) my CPU goes to 99% and on the more common times : 20-30% (5800x3d), which made me think that the GPU was ignored and cpu used instead
@@calebpeters191 thanks nonetheless dude
I believe the 1080ti does have cuda cores... But has 0 tensor cores which allows for much more efficient cycle usage of the GPU.
CUDA should still be your best bet, but I'm not sure why it's not noticing your card
@@Jarods_Journey I just bought a 4070 ti a few hours ago, and let it run. The performance is LEAPS better, thanks for the answer Jarod
vb cable causes popping when i talk through it voice changer is clear but audio that comes from the cable to talk on discord sounds robotic
and idk how to fix it
Mine sometimes has a weird voice for half a second before it says what i say. I thought it was just background noise but even with all the suppression and sensitivity being lower, it still happens. It kinda sounds like glitchy breathing
I’m using a i3 12100f with a 6600xt 16gig of ram however whenever I check task manager it only uses my cpu with little to no gpu usage I downloaded the direct ml version and I’m using gpu 0 should I just try redownloading it to make sure I have the correct version or are there any fixes and is my hardware to slow to use it?
This sounds correct to select GPU 0 for AMD. However, many users report issue with AMD cards and getting it to work with the voice changer, so you may want to leave an issues tab on w-okadas github in this instance or check if anyone has a solution there.
because the RES comes out with 8000 up to 29000 you can lower it because it takes a long time to load the voice and sometimes you don't even hear anything
I have tried to install it and get it to run, but something is missing and I have not seen a video on installing and getting the client to run.
Great Video as always! but im having a problem with my gpu, i cant seem to find my current gpu (AMD) on the options for GPU (and i mean cpu is the only one appearing), how do i make it appear to be an option?
You will have to download the directml version of the package, it should appear there so I've heard after this version is installed.
You ever figure out how to get it to work on AMD?
@@faded8975 download the directml version not the gpu cuda one
Is there any way to fix the bug where the voice changer use the system sound as well?
Help, how to turn off playback? when i talk it repeats after me to hear what i sound like but i want that off
Just a quick question, the voice changer works perfectly but it always cuts off at the end of the sentence. Any clues why?
+1
same
Can you tell me why when I say something through the AI it keeps repeating it?
How do you do it with obs cause I can't record microphone (that voice) it seems to detect the voice but i can't hear it on recording
Have you solved this problem yet?
Setup with vb audio cable, then set up microphone source in OBS as vb audio cable
@@Jarods_Journey thx for advice but i still can't manage to record it. I don't know what i'm doing wrong :
I tried this tool out. Honestly... I can see this kind of tool being put on a list of banned software at some point in the future. I gave it a go mostly to see if the DirectML implementation works on intel Arc and... well it didn't. Which is fair. I have a 1080ti as well so that's no biggy.
Ah, I think that's a slippery slope. If it does get banned, that leaves only bad actors to use it as nothing will ever stop them lol.
It's still hardware intensive atm so newer Nvidia are still needed
I have a rtx 3050 laptop gpu 4 gb i usually runs in on 384 chunks and 4096 extra and the voice comes out after 1 second and it was only using 2.6gb vram so what should i do to decrease the response time? I tried to reduce the chunks but it resulted in choppy voice
I think your gpu just ain't fast enough
bro, mine same like u but even worse cause my voice wont let out and after while it get error message.. btw my res is 0 and i think my gpu not detected bruh but im not using amd (on average amd doesn't detect gpu)
im on a 3050ti laptop and i can run mine at 96 chunk, 131072 extra perfectly
I have successfully put on voice models nad played around with them a little. Next time I opened the program it waits for web server indefinatelly and the voice changer native client pop up is permanately all stuck white. (Yes I deleted "stored_setting.json" it changed nothing)
hi, i wanna ask. why my res ms is high? hope you can help me with this. thank you