RVC: Local Training Tutorial (How to make your own custom vocal a.i.)

[@LucidFirAI]

I dunno if you can add subtitles or notes to a video that already exists but... KimVocal2 + VR DeReverb + VR DeEcho Aggressive + VR 5HP Karaoke work amazingly to isolate the main vocals. I can't seem to use ensemble to do all 4 simultaneously to get one output (I can do 2 at a time, and I get multiple outputs rather than an output that used all 4 of the above effects) A good tip I don't think you mentioned that is awesome is that you can load as many audio files as you want, so if you're taking a voice from an album you can load the entire album in and apply kim vocal 2 to all songs, then apply dereverb to all the kimvocaloutputs, etc. (I'm using UVR5 separately, it doesn't seem to work in my RVC gradio)

[@p3tro]

best i can do is pin this comment!!

[@bogobogobogo542]

Hey, would you mind sharing what aggressiveness did you find to work the best on the VR models?

[@LucidFirAI]

@@bogobogobogo542 it's not one size fits all apparently. To streamline it do like 10 songs at a time for testing

[@TheCommentVEVO]

Speed info: 3090 24gb - batch size 36 allocates 22gb - around 17 seconds per epoch

[@lostdreamer50]

hey sorry for replying 4 months later, but do you perhaps remember how many sound files you were using when you did this? I used around 370 (less than 10 sec) clips of me singing and set epochs to 1000. Left it for the whole night and more and it's still "processing" (9+ hours). That's normal right?

[@BigToro]

@@lostdreamer50 im late but generally you dont need much above 200 epocs, also the console window should show which epoch its processing on currently

[@BlueprintBro]

Fantastic tutorial, straight to the point with things. Now I'm just waiting for the training to end. 15 minutes of cleaned up voice, 500 epochs been 24+ hours 😅😅 Massive EDIT (If you are facing some issues like thinking the training is taking too long/stuck/suspended) At first I thought the training just needed a long ass time. I left it to "work" for over 48 hours, only then did I check the command line, where it clearly stated that it was still at epoch 13, after this long, the program must have bugged out or something. What I am doing now is saving a checkpoint every 5 epochs instead of 50. Since my first attempt did not even reach 50, it did not save, and I had to restart my model from scratch. The point of this post is that YOU CAN RESUME training a model anytime you choose, as long as you have those checkpoints. (I did not know this, which made me hesitate to turn off my PC or check the command line before) i.e. I want to train my model for 300 epochs. If it bugs out / stops for whatever reason at epoch 273. Stop the program, and restart it, using the same parameters you input when first training the model. It will load up the checkpoint at epoch 270, since I am saving a checkpoint every 5 epochs, so the last checkpoint would be at 270. I noticed it would bug if I use too many threads, and a large batch file size. For my specs 9 threads, and a batch size of 5 is optimal. 1 epoch would take around 4.5 minutes Trained a model for 300 epochs using around 50 minutes of sliced up audio. It took around 23 hours. Specs for those who are curious: i7-8700 GTX 1080 Using 8 threads to process

[@JamesT777]

A little late, but thank you for your service sir o7. Us low-spec users thank you.

[@eto38581]

There is no .pth file after training. I see there are index files only, no ckpt either

[@MusikIndieNesia]

Hi P3tro... I was wondering if it is possible for us to train a model, then stop... then continue the next day... I was aiming for 1000 or more epochs. I guess my question is, does training have to be done in one go? Or can you stop and continue? Thanks P3tro!

[@sebastianszwarc4162]

"Hello. Today I created my first model thanks to this tutorial. I used a 5900x processor, an RTX 3060 12GB graphics card - as training material, I used 25 minutes of an interview, resulting in around 150 samples of about 10 seconds each. The model took about 7 hours to train with 500 epochs. Unfortunately, I didn't notice an option to stop the training anywhere."

[@pujabanchu8239]

Even after completion of all processes, i cant see any .pth file for vocal i trained...i can see only index file..help me with that

[@Facha115]

same issue here

[@Thyronix]

for those of you having trouble, he skipped over a bunch of important shit and got a some things wrong. First off, you can't load those pth files (he shows that it generates in that folder) in a voice changer client, you have to load the one in the "weights folder.." you'll notice it's actually named what you named the experiment... second, when you click "train model," that doesn't generate the index files. You can either click "One-click training", or train it, then click "train feature index." Some more tips, I've had great quality using 700-1000 epochs and 10 minutes of clean audio separated into 10 second chunks. Don't use version 1 like he does, use version 2. Using more than 10 minutes of audio seems to slow the training by a lot and doesn't help with quality much. You can't use too many epochs though, it just won't get any better past a certain point. RTX 4070 12 gigs Vram took about an hour and a half to do 700 epochs.

[@thegtlab]

thanks for the tips man

[@Thyronix]

@@thegtlab np

[@AIImagePlaymates-r7h]

I can't seem to get the index file even doing your method.

[@PACKAGEMODS]

@@ThyronixHi so after training we use the path file in weighs folder? And for index you said to click train index to get the index file? Where would this file be located once completed ?

[@Thyronix]

@@PACKAGEMODS the model once completed saves to the weights folder, the index file saves in logs/project name

[@LosantoBeats]

Can you make a video on how to continue training the model. Covering things like adding more audio files to the dataset, and running more epochs on the model. Thanks!!

[@CyPhaSaRin]

with the dataset, whats the rule on the clips, and the ten seconds, is it strictly 10 seconds, should i go 8 seconds if the last two seconds are scream where the first 8 were sung.. should i aim to divide the different vocal tones or ways of singing the singer can/does sing clip to clip or should i just cut a studio vox into 10 sec clips til the song is over.. Looks like it's kinda answered for me, UP TO ten seconds, but still keen to know if there is a benefit to cutting and sorting the vocals based on tone, sing, scream, that kinda thing..

[@Mowgi]

Currently in the process of training my first model, thanks to you. Thanks P3tro

[@amcerngaming502]

Thank you so much for this video! I spent days researching on KZbin on figuring out how to generate and train AI voices because I wanted to create my custom made voice, but this one had the most clear and straightforward instructions that finally got my project to work. I almost gave up until I saw this video ngl lol

[@GXboy12345]

Is there a way to do this on CPU-only (AMD device)? I tried setting the GPU argument to blank, but it's just frozen in the console on the python runtime arg

[@slevpy]

when i press train model, no .pth file show up. do you know what i might be doing wrong or where they could have exported?

[@Himalaya-ri4kj]

Bro....any solution,have same problem

[@magmad1ver]

no one's answered so far

[@slevpy]

@@magmad1ver yea bro I completely gave up on ai voice models stuff a while ago. To much bs to work with and no one giving answers on how to make it work🤷🏻‍♂️

[@RoyalMintz]

@@slevpy the .pth is in the weights folder

[@niloytesla]

u know what... best YTchannel ever!

[@pijoctisgamer]

Could anyone out there help me please? When I click on "Train Model", this message appears: 训练结束, 您可查看控制台训练日志或实验文件夹下的train.log (Translation: After training is complete, you will be able to view the console training log or train.log in the experiment folder) Does this mean it is training the model? Because nothing appears in the console.

[@SherlocMagico]

Hey p3tro! Shout out to you btw, great tutorials. I've been working on a few A.I. songs just to listen to personally. How do you actually get the best quality vocal takes from the svc? Is it a certain way the vocals should be recorded? Should I upload the vocals with/without FX? Because I've found a bit of compression gets rid of those silent bits the A.I. generates for the quieter parts of the vocals. How would you go about getting the best quality out of it? Loaded questions, I know.

[@p3tro]

Always upload the vocals as clean acapellas! The volume matching feature in RVC can also be turned down for quieter vocals to more exactly follow the vocals, for instance it starts at 1 by default - turning it around .35 or .45 for quiet vocals will make it follow them more specifically! The style of vocal recording depends most on the style of the model, try to match what you want the model to sound like in your own delivery so the model has to do less work for you! For example, the best lil Uzi AI songs use lil Uzi covers as the input since the artist (Uzi Clone) sounds so much like Uzi already!

[@hanynagy8969]

@@p3tro I m really need your help! If I have my own custom trained model but I want to add a new data to it to improve it,How can I add these data (New Audio Files) without the need to create a new model from scratch? PLEASE anser me because every time I want to add a new audio I go to mak a new model. Thanks In Advance!

[@nswij123]

@@hanynagy8969 Did you get the answer?

[@ItsJustAdrean]

A note for everyone using this and getting an error, make sure there are NO SPACES in the names of your speaker, folder, etc! I don't know why it worked for p3tro and not for me, but this stumped me for a sec

[@martinfleiva]

I've been watching your tutorials, you're a god! Thanks a lot.

[@p3tro]

Glad I could help!! :)

[@EntropicDecayGaming]

After I hit the "Train model" button, the "Output information" box displays "After the training is completed, you can view the console training log or the train.log" so there's no error and I think I've done everything correctly but after 10 hours, STILL no .BAT or .INDEX files. How long does this thing take?

[@Mowgi]

So my first training took 12+ hours. Changed it to 24 threads (I have a 5900x) and the next one was done in 5 hours lol. Man, my fans were going crazy for 12 hours that first time 💀💀💀

[@AmanKhan-bj7im]

What gpu you have ?

[@davidplesa]

what do you mean you changed the threads? how do i change them?

[@lostdreamer50]

I used around 370 (less than 10 sec) clips of me singing and set epochs to 1000. Left it for the whole night and more and it's still "processing" (9+ hours). That's normal right? I guess I kinda overdid it with the amount of clips? I do have an rtx 3080 though. In the lower right corner it now says "processing | 46185/42.3" the left number's increasing quickly

[@hungvominh2490]

Hi! How long did it take you to finish the training? Mine is currently in the same situation, it's taking so damn long.

[@lostdreamer50]

@@hungvominh2490 hey can't remember exactly but it took the whole day or so. I may have overtrained it. Must try again with tensor

[@BeeBN]

now that google is cracking down on rvc this is a good alternative that they CANT take from us.

[@aschenixy4529]

For those having problems with step 2a having a blank preprocess note or processing data with blank output information... I think you have to make sure that the training folder doesn't have spaces or probably even symbols... Hope this will help as it did for me.

[@justinchey2281]

I have a folder with a single .mp3 that I wanna train my model from but when I copy the file path and click process I get some error about "\RVC/preprocess.log'" doesn't exist (which it doesn't. Should it?) Anyone know how to fix this?

[@HOTTAZHELLENT]

I get the same error no still haven't found a fix

[@Goddessღ]

@@HOTTAZHELLENT Did either of you find a fix? I have the same issue.

[@blabbilizer]

my gpu wont work because it's amd, what do i do about amd?

[@jammysnewadventures]

help! It says Errno2, no such file or directory! what do I do?

[@realweifi]

i didnt get any index files and i got two weirdly named pth, D_2333333 and G_2333333

[@VideoGameAcademy]

Nothing happen when i click on "process data".....got these error in the consol "runtime\python.exe trainset_preprocess_pipeline_print.py F:\Voice AI\train 40000 8 F:\Voice AI\RVC-beta\RVC-beta0717/logs/rainzo False Traceback (most recent call last): File "F:\Voice AI\RVC-beta\RVC-beta0717\trainset_preprocess_pipeline_print.py", line 8, in sr = int(sys.argv[2]) ValueError: invalid literal for int() with base 10: 'AI\\train'

[@Sullykato]

could you solve it?

[@DJAdalaide]

To get copy as path on windows 10, you have to hold shift for the menu to come up

[@datdus5524]

For some reason I get a filenotfounderror [Errno 2] No such file or directory.

[@Antonsetiady]

Blessed your heart P3tro; i really wanted to do this locally; sadly i dont have a window laptop yet or pc? I know that i need a window laptop with a nvidia gpu ge force at least 12ggbvram... any brand ok? Is 8gb vram is doable too?? Id like to do several model? Or PC is still the bestway as it is more futureproof? Excited about your website service as well :) Is it possible for trial or at least you test hearing the data set before we purchase? If the dataset is perfect to go train or not? Once again; thank you thank you THANK YOU !❤

[@p3tro]

Pc is definitely the best way, and 8gb of Vram will work for both so-Vits and RVC! I am going to release a video on using a google collab to train/covert vocals as well! As for the website service, I’m not sure what the plans for releasing examples or showing off the datasets Is, hopefully it’s updated soon, I agree we should be able to see the datasets before purchasing anything and also what if someone sends a dataset that isn’t good enough?

[@samuelepanecaldo1326]

dk why but it keeps saying my nvidia drivers are old, even though i updated everything 20 mins ago... help

[@LinkinV]

I just made a model but there wasn't any Index file, any ways I could find/fix.

[@JimBaker-ks4io]

Thank you, great tutorial! hard to find information about the local version of RVC. Do you need to hit the "Train feature index" button too in order to get the proper files created in the folders?

[@p3tro]

It’s supposed to be a feature of the train button, but the f.a.q reports it’s often necessary to train index feature regardless when the index file doesn’t appear! For me rather or not I need it is hit or miss

[@JimBaker-ks4io]

@@p3tro Okay thank you for all the knowledge you're making available! In case I have to click it, is it a long process too?

[@p3tro]

@@JimBaker-ks4io Not at all!

[@JimBaker-ks4io]

@@p3tro Thank you! Don't know if you can help with this. My dataset is 1 hour long. Once RVC starts training, the program crashes after 3 minutes. It even kicks me out of the windows desktop, and when I come back to the session, RVC tells me there was an error. My settings are 200 epochs, GPU batch size set to 4. I tested two other models today and the whole process worked fine... I never used more than 30 minutes of audio though, and didn't go higher than 50 epochs. Perhaps the amount of data is too much to begin with? Is one hour overkill?

[@p3tro]

@@JimBaker-ks4io one hour can work, it just has to be prepared properly, but to be honest your best bet is 10 minutes or so of the HIGHEST quality, anything even slightly low quality should be removed, it will work great with 10 minutes of high quality acapella. For me I aim for 500 epochs at 10 minute data sets

[@AtlasGeneticist]

theres a realtime gui bat and I'm curious if youll ever cover that?

[@SomedayTomorrow369]

Hey great vid thx , Should audio clips be in mp3 or wav ? Or it dont matter ? thx

[@jeetchheda8916]

Suppose if i am done creating the model, can i make use of that model for becoming a TTS Voice pack ?

[@communionhiphop]

man, thanks so much for this informative video!! just one question, why did you choose 40K as the sample rate over 48K?? I think 48K would deliver better quality sounding vocals?

[@p3tro]

48k sometimes has problems processing when I do feature extraction, atleast in my experience!

[@communionhiphop]

@@p3tro what exactly is feature extraction? btw I would like to run an experiment and try with 48K just to try to get better sound quality...

[@adallaire6511]

Hi p3tro! Great stuff works for me as explained. I'm blown away. Thank you for this. Will you do a tutorial about the 'ckpt Processing'? I'd be very interested to hear what this would sound like. Tx

[@xParanoiiia]

Hey I had an error, that I think I fixed very stupidly. It would not start processing my audio samples. The error chat GPT couldnt help me with. What fixed it was renaming both the source folder that the whole program was in and the folder that my samples were in. Both names previously had spaces in them "AI Voice" etc. just left out the spaces and it worked without an issue

[@Catcarp8]

Dude thank you! This fixed the errors I was getting during processing too.

[@UdawalaweWildAdventure]

Whens last step running ( making epoch) my display turned off.. But pc is running. Numeric and caps locks a working.. Is that normal or what?

[@SquooshyShark1000]

your display went to sleep lmao has that never happened to u before

[@johnbob70]

Great video! Where can I find the index file? For me it is not saved in the log folder.

[@johnbob70]

Ah you also need to click on train feature index...my bad 😆

[@p3tro]

@@johnbob70 nah that’s my bad I totally forgot to mention this 😭🤣

[@Lowbarful]

@@p3tro I'm having the opposite problem: It's creating an .index file for me but no .pth file. Any ideas?

[@p3tro]

@Tech Crew Guy I have seen this issue be fixed by using “one click training” I believe, but one issue could be having it installed on a improper drive - like say your external hard drive as opposed to the same drive as your operating system. Also - check your weights folder, this is where your shareable model will be. The model in logs is for use with your Rvc only

[@Lowbarful]

@@p3tro Dang... for whatever reason my weights folder is empty even after everything finishes

[@blackhand9581]

If you DO have a path file but no index file, don't worry. A fork of RVC called 'Replay' makes it so that you don't need an index file, just the path file.

[@chriscrowe11]

THank you! i was very worried and about to restart the entire process

[@-nishizumisan-]

Только что подумал о том, что неплохо было бы, если петро запишет гайд по rvc, и тут же видео выходит. Магия.

[@p3tro]

Я, должно быть, чувствовал твои мысли отсюда! :)

[@-nishizumisan-]

@@p3tro Definitely! I just mastered svc from your guides, and I have already heard about the advantages of rvc and am ready to study your new content using it. Thanx for your work!

[@qwerty0yt]

Anyone has the same problem as mine at step2a? It hasn't appeared "end process". This happened to some of my audio datasets, but some didn't. I so have no clue what's actually wrong :(

[@p3tro]

This usually has to do with naming conventions, the size or volume of the audio, or the location of the files! Try and put the dataset folder in your RVC folder itself, or try and make sure they are longer then 3 seconds or so and shorter then 10, then make sure they are all atleast -6db type of volume!

[@qwerty0yt]

@@p3tro wow thanks again. i have tried changing name (but still not sure what's their criteria), and already tried to change the location (but it seems like not my problem because I also had tried the dataset that has been worked and it still works, except the latest ones that i met the problem). About the length, I'm not sure, is the inconsistency (some are shorter than 10s, some are 10s, and some are longer than 10s or even some are about the 20s) of them affected? however; in the later tryings, i had already erased some that exceed 10s, and it's still not working, Does it really help?). But The size or volume of the audio, hmm, I never know I'm unconsciously avoiding thinking of it. Is AI not fully support belting or loud voice? Give big THANKS to you for giving me hopefully helpful ideas!!

[@p3tro]

@@qwerty0yt glad to help as much as I can!! These are just tips that helped me, I agree sometimes it still just randomly ignores some files or won’t end process 😭the ones above 10s are fine they will just get ignored! I know that when I used a limiter to change the volume of all my samples it accepted them and it fixed the preprocessing for me. But that isn’t always the case, just something that worked well for me ! It can be very finicky, like you said sometimes it will run just fine and other times it won’t! Even switching browsers by pasting the URL into another browser randomly helped me before 🙏 Hope you get it working!

[@joerick8384]

When I go to logs folder, I dont have all of those audio files, I just get a preprocess.log file, even though I got all "success" in the command prompt and "end preprocess" lol.

[@matttdk4888]

i got the same problem, did you find a solution?

[@joerick8384]

​@@matttdk4888 Yes, one of my files; for example "audio samples" had a space between the words, so I just changed it to like "audiosamples" but I think that "audio_samples" would work too, just no spaces. I dont know the rules for spaces but generally I try to avoid them now.

[@diegolopez-xz8pg]

Hi P3tro! How many gb of VRAM has your RTX 360 Ti model? 8, 12 gb? Regards from Argentina!

[@froggustusthe3rd69]

It says my gpu is not supported and I can't use feature extraction

[@ThisIsntmyrealnameGoogle]

One of the best youtubers and one of the few I have notifications for thanks! Any good quality mics you recommend for hobbyists (non professionals)?

[@p3tro]

What’s your price range! There’s some good condenser mics for cheap, under $50 even! I like a dynamic mic like the Shure SM7B but at $399 it can be pricey for a hobbyist’

[@ThisIsntmyrealnameGoogle]

@@p3tro Id say around 150 but if it's really worth id be willing to go up to 200!

[@opiniggao-opn9638]

Hi, how are U? Nice video. Where can I find the Download of training program?

[@Angus-Ghoul]

hi! when I am training the dataset, on my pc pop up an blue screen and said "our device ran into a problem and needs to restart" . I have a geforce 3060 4gb + 16gb of ram so idk what is happening :(

[@Arrowkraft21]

I can generate 500 epochs fine, I end up with a pth file but no index files at all. I have cut down the dataset and run it again but it still gives me no index file. Any idea why? :(

[@p3tro]

If the training doesn’t give you an index file, you can use “train index feature” to generate one!

[@Arrowkraft21]

@@p3tro Thanks so much! I managed to get a result! although the pitch does jump around the place haha

[@yucci5311]

​@@p3tro Hello, I may be late, but maybe you can help me... The same thing happens to me, but when I use the “index train function” it generates an empty folder instantly, do you know why?

[@alexfonseca5076]

Ran a model, trained it for 500 epochs with no errors. .pth files created but no Index files. Any reason why?

[@Twiddle_things]

This guy chill asf damn. Thanks for the tutorial man!

[@Melvitos]

Maybe I'll get the answer later but, Why every tutorial I've seen are talking about Google Collabs struggling with GPU limits, when you can just do it locally as you did???

[@equrantutors]

whenever I hit "one click training " I see no error, it says process completed but I don't get .pth file, I only have log files. can you please help me regarding this issue?

[@Freeak6]

Is it really necessary to cut it in 10s clip ? I thought the preprocessing was taking care of it.

[@cmyk8964]

When I train my model, I get two .pth files: one starting with D_ and the other starting with G_. It also generates two .index files, one starting with added_ and the other with trained_. Which ones do I use?

[@vonckjenl3633]

i had the same thing, but i found the right pth file in the weights folder and then my rvc worked

[@Neophage]

Does the dataset audio actually have to be divided into 80-100 clips or can the audio be one long, continuous track?

[@AmanKhan-bj7im]

Segmented clips are better.. they'll save up some memory..

[@CL3NRcProductions]

I have no clue where the files could have been spat into. you uh.. did not show us the file location

[@frattuncbas]

How can i continue my pre-trained model later? I want to train it 20000 epochs for most realistic quality but i need to run day by day my same pre trained model..Is it possible on Easygui Colab? I have Pro colab

[@xunknownartist7311]

Hey guys, I'm running Step2a of training in the RVC webpage under "train" and I keep getting the error "FileNotFoundError: [Errno 2] No such file or directory: 'C:\\RVC-beta\\RVC-beta0717/logs/kara/preprocess.log'" - I copied and pasted the path of my vocal sample (folder is called "KaraVoice") and this error keeps coming up. I move the folder and copy the new path and the error still comes up. Do I need to do some sort of preprocessing?

[@CricGeek-bb9ii]

Watched many videos but clipping video into 10second file and low batch side fixed my error on GTX 1650 ti (2 batch size )

[@vidneypopples]

is there an easy way of mixing vocal & instrumental back together? Doing it in audacity is a bit of a faff. Something more automated would be good.

[@sosososo4348]

What is the solution to the problem ? FileNotFoundError: [WinError 3] The system cannot find the path specified: 'G:\\RVC-beta\\RVC-beta-v2-0618/logs/myvoice/3_feature768

[@babyyjmusic]

hi ive been finding a problem for the past 2 days now i dont know why but everything pops up accept from my pth file in the logs

[@grandlos7278]

Do the dataset clips need to be cut into 10s or can they be just one long clip?

[@JimBob_Joe77]

Do you have to use clips of singing only or can you use clips of the person speaking also when making a model?

[@SquooshyShark1000]

speaking works just as well

[@AbonneerOpRijk]

After training it gives me a PTH file but not an index file.. How to fix this?

[@beesechurgertf2]

What should i do when it says "connection errored out"?

[@laszloopoczki7509]

Hello! How can I set the RVC when I want to teach so that it doesn't use the GPU if not the CPU? Because it doesn't support my GPU, that's why I'm asking..thanks

[@VuLe-sd4un]

Hi, I have tried to train a model followed your instruction. But when it finished, I can't find the .index file anywhere... Can you kindly help me out..?

[@sebastianszwarc4162]

Click „train index” button

[@AIImagePlaymates-r7h]

@@sebastianszwarc4162 It doesn't work for me

[@Tyulenin]

"No eligible .pth files found". Am I dumb? upd. at least in my case, the proper ".pth" is in the weights folder.

[@thecakeplays]

Running into this error on step 2a - "ValueError: invalid literal for int() with base 10: 'set'." Can anyone help?

[@thecoolrich]

Yo can I dm you bro?

[@acshighlightsff]

dude i am getting file not found error when processing output after pasting my dataset path.. Help please!!!

[@LeGnocchi]

7:04 I have 2 .pth files, "G_2333333" and "D_2333333"

[@Tyulenin]

same. Does it go "No eligible .pth files found" if you try to use any of them as a model? upd. at least in my case, the proper ".pth" is in the weights folder.

[@LeGnocchi]

@@Tyulenin Yes, I also found the right file 👍

[@rmdir]

well where is it?@@LeGnocchi

[@RedouaneAouameur]

is it mandatory to have a gpu for training?

[@datdus5524]

For some reason I get a filenotfounderror [Errno 2] No such file or directory. Any one any idea how to fix this?

[@rmdir]

Thanks for the tutorial, I think I might be able to get a model working here, but I still have one question. How long does it normally take to train a model with 500 epocs or so?

[@elskepode2]

200 with a 3060 and i7 12700f is averaging 160 minutes per model. 500 would be around 4-5 hours with similar hardware

[@sebastianszwarc4162]

"Hello. Today I created my first model thanks to this tutorial. I used a 5900x processor, an RTX 3060 12GB graphics card - as training material, I used 25 minutes of an interview, resulting in around 150 samples of about 10 seconds each. The model took about 7 hours to train with 500 epochs.

[@ced.studios]

can I use RVC trained models in Tortoise TTS and vice versa?

[@Lightning-uv6ri]

hey i was going to download a rvc model but I dont know which one to download which one out of these is better Slump god (RVC) 220 Epoch, Slump god 60k, slump god 10k?

[@SubhadeepGayen]

Why do you need 10 sec? Why not 1-2 hr single file?

[@717MERCURY]

is there anywhere I can find .wav files to see examples about how would be the ideal type of samples I should provide my AI so it learns a more wide register of my voice?

[@elskepode2]

Theres a python program you can download called instant dataset maker. You load your audio files into the audio file of the download and then run the application and it will create 10 second clips of all the songs you put in the audio file and you just save the zip to wherever you want then extract it and point RVC at it in the training tab

[@Epicfuzz]

when i train my own models, i dont get any index files, just pth files. anyone know why

[@p3tro]

It should auto generate, but often times you have to "train feature index" to get it to spit out, which is right next to "train model". If you already made the model just make sure the name and settings are the same, and it will still generate the correct index file

[@roguebinks8985]

Thx I had the same issue and was very disappointed when I didn't see them. @@p3tro

[@bigdaveproduction168]

problem, there is nothing when I click on process data at the first and second steps

[@ced.studios]

getting this error on Training ValueError: invalid literal for int() with base 10: what is this?

[@ced.studios]

figured out to remove spaces on folders and fixed it.. now I have this error RuntimeError: Failed to load audio: ffmpeg error (see stderr output for detail) end preprocess can someone help?

[@wesing7406]

ValueError: invalid literal for int() with base 10: 'Video' Help Me

[@hyperlaserrebrn]

do you have to manually do the bat command everytime you want to train or can you install it

[@Dhruv_hazard]

after training the model the path file is not showing

[@AVoices]

"RuntimeError: The expanded size of the tensor (12800) must match the existing size (0) at non-singleton dimension 1. Target sizes: [1, 12800]. Tensor sizes: [0]" - such mistake on last step...

[@furious_eros]

Am I able to continue training an already trained voice?

[@LuciferSamaelMorningstarLight]

Good day! Awesome tech and videos! One thing i do run into issues with is with the training, my model trains and succeeds every time, but when i go to convert and audio file using the finished model, it loads and converts, but when i play, no sound, no voice. Now i have gotten it to play before once or twice but that was hundreds of times and days ago. I do try over and over but with the same result mostly, where it finishes and succeeds and everything loads and converts fine, just no sound. Not sure what would cause a model to produce no sound after training, as i know it can work, it just doesn't tho, at least for me, once or twice getting it working using exact same ways it's supposed to be done over many many other times, but once or twice over 100's or 1000's of time isn't really much

[@locmonstr]

I cannot find the .pth file. It just doesn't give it to me

[@p3tro]

I fucked up this part by not mentioning it, but if you refresh your voice list you will see the pth file, but the actual shareable version will be in the "weights" folder! When i made my models I wasnt sharing them so i didn't realize there were seperate files!

[@locmonstr]

@@p3tro I refreshed my list and the path file still doesn't spawn

[@p3tro]

@@locmonstr are you sure the training was successful? It should already appear in weights after your initial training - the refreshing voice list was so it would appear in the GUI as an option for available models

[@locmonstr]

@@p3tro Yeah the training says it was completely successful

[@p3tro]

@@locmonstr did you use one click training or just regular training? And are the D_ and G_ files and index files in the log folder?

[@VegaLyrae]

Super curious to find out how well this works with speaking clips instead of singing ones. I know most use cases for that are probably not good ones, but I have a custom edited Azure voice I use for a AI that I want to train this to sing like. I’m gonna try it but anyone with advice on this would be greatly appreciated.

[@VegaLyrae]

note sounded awful with 300 clips. i fed it diffrent tones, pitches, and settings and it didnt work well. i'll rety with a lot more but so far no dice

[@VegaLyrae]

few months later. training data was the issue. voice is actually awesome now. however I ditched AZURE. Azure sucks.

[@RoyalFlush007]

Hey, thanks for the video. I ran into an issue where in step 2a where it says "Enter the path of the training folder:" no matter what file path I give it it will not process data. It cannot find the file path in the terminal every time.

[@p3tro]

This happened to me when I left writing in the example file path area! Make sure you leave the area blank where it is has the example file path if that makes sense

[@RoyalFlush007]

​@@p3tro Thanks for the reply, I've been searching all day but I'm still not exactly sure what you mean. I also have been using chatgpt when I first started with this stuff, but it wants me to edit the code in the terminal which I am not allowed to edit anyway. I end up with a "FileNotFoundError: [Errno 2] No such file or directory:" I simply can't figure out how to upload my .wav and get it to actually work.

[@p3tro]

@@RoyalFlush007 there’s a way to type your file path for your wav in, copy and pasting the address instead of actually uploading the file should fix it if that’s the only error

[@RoyalFlush007]

@@p3tro I have been copying as path and putting that result in the box under "path to training folder." Is there actually a place to upload the file as well?

[@SomedayTomorrow369]

Tried on 3 different pc's and everytime i hit the process button im just getting these same type messages each time ' File "C:\RVC-beta-v2-0618\my_utils.py", line 14, in load_audio ffmpeg.input(file, threads=0) File "C:\RVC-beta-v2-0618 untime\lib\site-packages\ffmpeg\_run.py", line 325, in run raise Error('ffmpeg', out, err) ffmpeg._run.Error: ffmpeg error (see stderr output for detail)" Any idea bud ?

[@p3tro]

If it’s not working on all 3 pcs it’s definitely a problem with the installation, unless all 3 pcs somehow don’t have powerful enough GPUs. But the processing features part uses CPU anyway. Is it installed on your main hard drive? Sometimes even moving it to the very root of the drive instead of documents or other folders can be helpful. The threads = 0 seems to imply it set your CPU threads to 0 which would mean it isn’t using your cpu at all. Where it says “threads” try adjusting that number to 8

[@SomedayTomorrow369]

@@p3tro thank you for replying , ill try them keep up the good work ))

[@p3tro]

@@SomedayTomorrow369 hope you get it working! Let me know if your still getting the error I can try and think of a few more reasons why it could be happening!

[@maradota1]

10+ hours training, 124 files from 5 to 10 sec. Processing 55200/139.7 . Im close to the ending? haha . Thanks for the tutorial bro

[@sebastianszwarc4162]

"Hello. Today I created my first model thanks to this tutorial. I used a 5900x processor, an RTX 3060 12GB graphics card - as training material, I used 25 minutes of an interview, resulting in around 150 samples of about 10 seconds each. The model took about 7 hours to train with 500 epoch

[@zukushimeku]

is possable to run this on mac? I have silicon mac so I can't bootcamp

[@Harry_See]

How do you detect and delete noise in FL Studio? I only see an option to denoise a noise sample. Thanks!

[@p3tro]

You use edison, and highlight the noise you want to remove - then you click the toothbrush and click "detect noise" - then you can select all of it and click the tooth brush again and "denoise". I cover it in this video as well: kzbin.info/www/bejne/jJKnmXt_ebSjhrs

[@Slinky]

Is it possible to load the model you trained as a base model and then add more data and increase the total epochs of a model?

[@p3tro]

Yes ! Just make sure the file paths and settings are the same and it should automatically load it and keep going

[@Slinky]

@@p3tro Interesting, thanks!

[@Slinky]

@@p3tro Would the model sound weird if it has been trained on a speaking and singing data set? Really curious if it has been done with success.