(COLAB PRO ONLY) AI Voice Cloning with RVC in GOOGLE COLAB - Guide and Setup
HTML-код
- Опубликовано: 4 июн 2023
- Links referenced in the video:
RVC Github - github.com/RVC-Project/Retrie...
Previous so-vits-svc vid - • Complete Guide: AI Voi...
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... |
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:
Corsair Vengeance i7400 - amzn.to/3p64r22
MSI MPG Velox - amzn.to/42MnJHl
Cheapest and minimum specs recommended:
Cyberpower 3060 - amzn.to/3XjtZoP 
Наука
CURRENT Issue 9-15-2023: I was made aware that Google has started banning RVC usage on free accounts, similar to what it did to stable diffusion. There is no fix for this ATM other than to get the PRO version of Collab.
IMPORTANT: You MUST click "Train feature index" at 12:07 in order to get the IVF index file you'll need later. As noted by another comment, this can be done before or after training.
Sorry about that guys!
Thanks !
what is the difference between the two? is there anything different about it?
good lookin out
what is the consequence? Is it train model -> train feature index -> one-click training?
@@user-wy7qe9zh2z whats the diff between them
Man, thank you so much for such detailed guidance! I've watched about 10 videos on the same topic but did not understand the process. You really helped me, thanks!
You did humanity a huge favor by making this tutorial , many thanks ! So much detail and well explained ! Liked and subscribed !
Appreciate it!
thanks for the video bro, i've just spent 5 hours on a model of my friend and it works relatively well without much training, totally worth it
I just want to say thank you so much Jarod for making this video. I can feel the passion and kindness and you were doing it to help people not taking it for money. It really worked for me the collab I was dancing crazily when It worked. I was so happy. Thank you, I subscribed!
Haha appreciate it, glad I was able to help you get it all up and running :D!
@@Jarods_Journey Thank you. All the best. I will be supporting your future videos! Thank you for replying too, It feel great when the author of this video replies to their fans! :D
My last programming was in 1997 with basic (i was 12). This is whole new universe. Thank you!
I just want to say publicly, that I appreciate you Jarod for creating all of these guides. I can tell that you’re not doing this for views and that you truly have a generous spirit. These videos will create businesses..generate revenue. I explore anyone that does so pays it forward or pays it back. I know I will… ☯️🥂🚀💯
Thought I had left a comment on this one, but going back through comments again, really appreciate it Gerard 🙏!
@@Jarods_Journey I need your help please.I just want to Update my trained model,I mean if I want to add more date(Audio files) to the model,Is it possible? Because every time I have new data I make a new model from 0 so I m tiered from that.Thanks in advance!
thank you so much for full detailed tutorial , after watched so many videos finally i found your video with full detail step by step. thank you so much 🙏
Appreciate it, glad I could help :D!
I've watched a different tutorial which didn't help me at all. Sound was robotic, and generally no much explanations whatsoever.
But this one helped me, and the trained voice sounds perfect! I am extremely thankful for this.
You are the best! :D
Appreciate it 🙏!
This dude is a legend for these videos.
Thanks for your videos on RVC, they were very helpful to me to get started with this. One important detail, in the video you put up a graphic telling viewers to click the train button, not one click training. This was good because I think one click training is slightly bugged, but you forgot to mention that if you don't use one click training you MUST also run feature training (2nd big button) or you won't have an .index file.
Appreciate it! Man, I had to do a double take real quick, but I do say it to click it at 8:16 😅. I realized after editing that one-click training does all of the previous steps before, so it was redundant to click one-click training. If you run all the previous buttons before and click one click training, it just redos all of the previous steps.
Edit: misread the comment, looks like an oversight and a missed step on my end!
It’s now banned tho 😢
Thanks for this tutorial! I tried doing the RVC on my laptop locally, but it seems like my 1650 ti gpu is not as strong as I thought. Good thing I found this tutorial!
Appreciate it! All these "AI" training tools are VRAM hogs, so colab is a great alternative.
tip: i use google colab for "Train" but "Model inference" & "Accompaniment and vocal separation" i use on gtx 1650
Hey! I've been following your tutorials for quite a while, and I must say - they are helping me A LOT. But in this one, I need a little help!
When I run the "start web" code that you show on 6:37 I get the error message saying:
/content/Retrieval-based-Voice-Conversion-WebUI
Traceback (most recent call last):
File "/content/Retrieval-based-Voice-Conversion-WebUI/infer-web.py", line 7, in
import faiss
ModuleNotFoundError: No module named 'faiss'
Have you seen something like this and how we can fix it?
Try re-running the cells in the Collab, faiss may not have installed fully or correctly
You earned a new subscriber ❤
And hopefully I bestowed some new knowledge :)
So, theoretically, if it disconnects because of 12 hours period, how am I supposed to finish extraction and training faster? Or there're some checkpoints for data, so I can keep going from the previous one?
('Cause it threw me error at first try and after that collab didn't want to start again)
If anyone is looking to use RVC locally, it’s worth it. Saved me hours of training and it sounds better. Haven’t touched sovits since I installed Mangio-RVC.
I had troubles with installation but after 2 days it was worth it. The CHAT channel in the ai discord helped
Interesting fork! RVC is definitely worth it and in my experience, it just trains faster and produces better results. I gotta take a look at the architecture to see why lol.
Thanks! Trying to decide if I should go local or colab. Will the local speed depend on the graphics card? I only have an RTX 2070 on my Razor Pro from 3 years back. OK its not that old but might be slower than colab? What are you running it on?
@@Cameron787 shouldn’t be bad at all. My 3060 ran fine. Even if it takes a few minutes this is crazy technology either way lol
I don't care much for google collab, but I'm here for the like, comment, and the view.
Appreciate it 🤟
nice!
Thanks!
TYS
10:38 you say (and do) "click one-click training" but you display "do train model". so what is it now? which one is the right one?
Do train models then after training, click extract feature index
@@Jarods_Journey isn't that what one-click training does? (training + feature extraction with one click)
Very helpful - thanks! Whats the difference bw pm, harvest, and dio?
Not too sure, but harvest produces the best results in my testing.
There is a difference, I just haven't looked it up extensively lol.
@@Jarods_Journey No worries, thanks! Do you happen to know what "Search feature ratio" does? I tried setting it all the way left & right, without much difference
@@theAIsearch I think the affects accent of the voice so if at 1, it may retain more of the accent
I have to say your tutorial is very useful so thank you for it but I have a question
the colab notebook seems to run an older version of RVC where there was no rvmpe
so how to run the latest version on colab?
i was getting lots of bugs on my windows laptop with amd hardware, so this is a good alternative
18:00 - that would be a good glitch voice for different Sci-Fi effects!
Most definitely 😂, I'm sure its highly possible to find some use cases for it
Hi Jarods, I am using Colab Pro, and didnt have issue at first, but now everytime I want to train model, it always shows me error, because two log folders are missing: 2a_f0 and 2b-f0nsf. If I add them manually, training will proceed to the end, but they will remain empty and my model won't change voice on cover song successfully.
I'm trying to solve the issue past three days, but without success.
Can you help me with that? Why are these folders missing now, and not before, and how this can be solved? Am I doing something wrong?
Thanks in advance!
Constant connection errors on every step in the public web RVC, for a second they seem to work and then give error messages. I managed to get to feature extraction but each time it just halts the process, says there is a connection error and sometime after colab disconnects
Question when executing "restore pth from google drive": After re-importing from GDrive to goolgecolab the *.index and the *.npy file are uploaded into the "content" folder of googlecolab. But is this the correct folder where those file should be in the end? Because afterwards when I am in the "model inference" tab, I have to choose then the "Feature search database file". I should then link to the *.index file which is now lying under "content", is that correct?
and: how many epochs do you think are perfect? is it possible to overtrain and get bad results when putting the epoch too high? let's say eg. 200?
Thank youuuu!
Yes, when you reupload back to colab, you have to copy all of the correct file paths to the spots where they belong at. There is no perfect epoch count as it is dependent on data, I haven't seen any instance of overtraining and it's not that severe when training on these small datasets.
This is really cool. Tho one thing, trying to download the multiple audio files at once completely crashed my Chrome. Luckily the Colab session was still going so I didn't lose anything.
Yup, it'll stay active for 12 hours as that's the runtime length unless you deactivate it!
[Edit: I searched the comments and found your response to someone else! Youre the best. Patiently waiting for my epochs to finish :) first time ever doing anything like this and im honored to have found/used your help)
Original: This was such a wonderful video. My issue with the training is that it gives me the error message "filenotfound error: [erno2] no such file or directory: pretrained/f0G40k.pth" any help with this step will be greatly appreciated. It is the default load pre trained model option in step 3.
Hello Gradio file is not found in temp. I got error so i went to the temp folder. But gradio folder not found inside it. What should i do?
Great video, thanks for that.
I saw in the description of the Voice Conversion WebUI that there is a link to hugging face space.
Will you also make a video on how to get this to work there? Many thanks in advance.
Appreciate it! Unfortunately, it's not hosted on hugging face and when you click the button, it just brings you to the page that you would need in order to manually install the models needed to run it locally
Is it possible to do this on the windows terminal because whenever I try to do it, I keep loosing connection so is there a more permanent solution?
Trained IVF file didn't appear in log -> me folder. How could I solve this?
Thank you so much for this helpful tutorial! It's very detailed and easy to follow along with. I trained my model with the default total training epochs(7), save frequency(5), and batch_size for every GPU(7). The result has some static. I plan to add more samples to train with. Can you recommend the number of epochs I should do for the best results? Or is it like the higher number the better type of thing?
Appreciate it! It's all data centered and a case by case basis, so really hard to generalize
A good rule of thumb though is quality over quantity. If you can get 1 hour of high quality audio samples vs 10 hours of mixed quality, I would just save the time and do the 1 hour of samples (faster and higher quality). Then you can run it for more epochs if you wanna try and smoothen out the noise, or add more high quality samples (though a lot will be experimentation and seeing what works best for your voice!)
@@Jarods_Journey Thank you for your help. I will give it a shot :)
Thank you so much mate. But I've got a situation here. after accompaniment and vocal separation and press CONVERT, I always got Connection errored Out. and can't convert. what is that? what i gonna do?
Is there any way to move GPU work for AI voice changer in to Google Colab? Cuz im getting 50-60% usage on RTX 3070 with Voice Changer 😂
You should cover the new mangio-crepe fork, it’s way better at making models and has RVC v2
Your video was so helpful!!! I wanted to ask if we come back another day to use our model, after loading her up on colab, do we have to train her again or can we just jump into the model interface step?
As long as it's trained, just inference :)
Can you explain what the "dataset" at the beginning is for? Do I need this to define in the google colab section or is it just the same like defining the folder in the training tab under step2a > "input training folder path"? Thanks for the great content, it's a big help!
dataset contains all of the data to be copied over to the colab repo, to then get trained on. You'll just set all of the things as shown in the vid for paths.
Hi, your tutorial helped me a lot (i just need to find a way to get the perfect model training to avoid all those "glitch" stuff, i tried to set the epochs on 200 actually).
The thing is, can i download the voice model (or is the files ado-japanese_D_935.pth, ado_japanese_G_935.pth and ado-japanese935.pth + the added_IVF file and total_fea.npy) to then use the model on the RVC-web (the .bat method) and to also be able to use the voice model on other websites / software?
Appreciate it! So I'm slightly confused with your question, but if you want to be able to use these models anywhere, you need the one that's in the weights section, and the trained index file in the logs section. You don't actually need the G/D ones in RVC except for training and doing training at a later time
Hi I am on mac, I don’t run it locally , I trained 2 models , now I can’t find them, I didn’t download them, do I have to start all over again? Or they are somewhere on my mac? Thank you
Im gonna need some clarification.. everything is on the Google drive. If you didn't download anything, nothing will be on your Mac. If you didn't save them after training, they will have been lost and you'll need to retrain them
@@Jarods_Journey thank you so much for everything u do, yes I had to retrain, my fault.
Always appreciate your videos. Question: is it possible to implement a additional rounds of training on previously trained models? I know you run another round of training on the model you create, but how would you get it to recognize a model that's already be created after having completed and closed out the process? would you just re-insert the .pth and index files back into the weights/other folder and run a new training process on a new dataset?
Hmm, are you talking about adding additional training to another model, say you downloaded it off the web and you want to train it with other data? Unfortunately, I think you would run into some issues with this, but it's 100% possible as the pth files are just parameters or weights for that voice. However, even if you re-instert the .pth files in their respective folders, I believe you also need all of the other parts of the training process as well which includes the previous processed logs and all those other things.
Never hurts to try though! I've never done it myself for rvc xd.
@@Jarods_Journey I guess he means if you have already started a training session with let's say 20 epochs. And then you realize you want to have actually 100 epochs. Can you somehow continue training on the previous model and then just add the missing epochs from 20-100 or do you have to start all over from 0 again? Thanks for this incredibly helpful video!
@@Jarods_Journey yeah, so to simplify, can I have multiple rounds of distinct trainings (i.e., dataset A, dataset B, dataset C) and have the same model retain what it learns from one to the other, or can you only train a model on one dataset, even though there may be multiple rounds (i.e, dataset A1, A2, etc...).im just trying to make the most banging model I can, if ya feel me
hi the video is so straight but i have an issue i dont have the "trained_IVF201_Flat....." but i have train.log what should i do?
Is there TTS for RVC?
Traceback (most recent call last):
File "/content/-EVC-/extract_feature_print.py", line 13, in
version = sys.argv[6]
IndexError: list index out of range
['extract_f0_print.py', '/content/-EVC-/logs/Cristi', '2', 'crepe', '115']
how do i fix this
Wait, just to be sure, is the "vocals" part for singing or something like that ? If I'm only interested in voices that talk, can I skip this ? Because i don't have singing recordings of the voice i want to clone, i only have the voice samples in the zip file where the voice is talking.... I'm confused
EDIT : AH OKAY, yeah so the video file is not a voice sample to be cloned, it's a voice that we want to replace with the voice from our trained model!! Ok I get it now, so yeah because I want to do the microphone real time thing of the other video, I don't care about inferencing my trained model on this additional video file like you do. Good.
Please Make a tutorial on mixing two rvc models using ckpt.
is this rvc or rvc 2? because in other web ui i saw u could select between v1 or v2 but this doesnt have that option. Thanks!
Colab (at the time of the vid) only goes up to v1, however, the local installation has the v2 option.
So do you train like for example 1 1/2 hour of your own voice with the harvard sentences(or maybe something better I'm unaware of) and then use 1 1/2 hour of a mp4 vocals or whatever voice you're trying to clone to go into the realtime AI voice changer?
You can do that, but the voice samples don't matter as long as they're complete words and there's no background sound. Once an RVC model is trained, that is what goes into the voice changer/
@Jarods_Journey I have a question about the dataset that you used to train the main model. Is it big? How big? I train french voices and they sing with an english accent! Cute, but sometimes weird!! I was wondering how to have a big french model and how many data was needed to do so. BTW huge congrats, your work is amazing. Love it.
Appreciate it! The datasets I use range from 10 minutes to 3 hours, but my average is around an hour of audio. In my experience, the index file controls the "accent" so you can try adjusting it to 1 and seeing if it results in a better accent
@@Jarods_Journey that's correct, english accent has almost gone. thx
Sir, this is the error i am getting from content data set
Could not parse variable and value from ""/content/drive/MyDrive/dataset/AKALEYO_NEE_vocals.zip"". Expected the line to start with a variable assignment. please help me
I am having this same issue, I've tried everything, including just naming my zip file what the program seems to be looking for. No idea what went wrong. I don't even have the same opportunity under the "Dataset location" to edit the path as seen in the video. It is just all red saying "Could not parse variable and value from ""/content/drive/MyDrive/dataset/lulu20230327_32k.zip"". Expected the line to start with a variable assignment" and the edit symbols are greyed out and inaccessible. Not sure what has gone wrong, I followed everything precisely. Would love some help! Glad to see I'm not the only one!
Is there a reason why the program can't locate my dataset folder in my drive? Does it have to do with it being banned?
Hi man! Just a query! I have trained my voice model for narration yesterday. Now, I would like to convert a TTS voice to my voice. I followed your instructions restored the path before going to web interface. Now, as you mentioned we have to put the path address of IVF file in database file path but now, I only see G and D file of my model when I am going to RVC - logs - my project. Can you please help on where to find the ivf file when we are coming on the next day.
You want to grab the weights file, inside of assets/weights. That is your voice model. Inside of logs/ is you're index file
how to continue training my model ... I had trained it for 200 epochs ... but it still not that good ... Is there any way to continue training it ... Or should I train a new model ???
hey, after you mmake one voice clone ai how do you make another one? Like, i wanna make another one do I just make another data folder but name it data set instead of dataset or do I put the files with the other data set, or do I delete the files from the last voice clone from that data set and replace it with my new audio files
You can just use another name or delete the old one (rename) and start over
When i want to make a model crepe is Best or harvest?
Hi, i see error " can't open file '/content/Retrieval-based-Voice-Conversion-WebUI/infer-web.py': [Errno 2] No such file or directory" at start the web step. How can I fix it?
I install all of the cells and when i click on the web Portion it says Python 3 cant open file or directory, What am i doing wrong?
Hey, everyone. I've trained a model with 400 epochs with voice samples of 10 min audio. When i'm doing the voice conversion, the words are not even pronounced correctly. The sound looks more like humming instead of speaking. What am i doing wrong? Appreciate your help.
The python console in the Collab is throwing me errors about non-existent modules whenever I click on the "start web" cell. Why is this?
I'm running on pro version, I'm getting the following error:
I added a single .wav file to the /content/dataset-2/ folder inside my collab
start preprocess
['trainset_preprocess_pipeline_print.py', '/content/dataset-2', '40000', '12', '/content/Retrieval-based-Voice-Conversion-WebUI/logs/me', 'False']
/content/dataset-2/.ipynb_checkpoints->Traceback (most recent call last):
File "/content/Retrieval-based-Voice-Conversion-WebUI/my_utils.py", line 14, in load_audio
ffmpeg.input(file, threads=0)
File "/usr/local/lib/python3.10/dist-packages/ffmpeg/_run.py", line 325, in run
raise Error('ffmpeg', out, err)
ffmpeg._run.Error: ffmpeg error (see stderr output for detail)
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/content/Retrieval-based-Voice-Conversion-WebUI/trainset_preprocess_pipeline_print.py", line 75, in pipeline
audio = load_audio(path, self.sr)
File "/content/Retrieval-based-Voice-Conversion-WebUI/my_utils.py", line 19, in load_audio
raise RuntimeError(f"Failed to load audio: {e}")
RuntimeError: Failed to load audio: ffmpeg error (see stderr output for detail)
/content/dataset-2/record.wav->Suc.
end preprocess
The preprocess seems to have worked, but make sure there are no other files in there and no spaces in your path. This is usually the fix that I see for people.
Is there somewhere we can download voices that others have already trained?
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.
I don't see anything under the inferencing voice tab. I have refreshed the voice list. I do see the index logs file under the "path to the .index" file tab, however. Process data step ended with "end preprocess" and the feature extraction step ended with "all-feature-done". Can anyone help?
Thanks for your content, Jarod.
I might need a bit of clarification, but I believe you have everything finished that you need to train. Once the training is done, in the inference tab, you gotta click the refresh timbre button.
One additional thing is check to see if the training outputted a .pth file in the weights folder, if there's nothing there, the training didn't finish correctly!
@@Jarods_Journey Yeah I when I check the weights file there isn't any .pth file. It is weird.
Anyway, I watched your Google Colab tutorial for SVC and I managed to make it work! Thank you so much, mate. Keep it up.
How do I get the pro version? Will there be a free version?
Wil it work with the Google Colab Pay As You Go options?
weird. my comment got deleted. Anyways , put this right at the start of Install dependencies if you keep failing to build pyworld and/or not be able to find faiss
!pip install pyworld==0.3.2
!pip install numpy==1.23.5
My comment was deleted too, I sent whole installation line which should've fixed that
Yt doesn't like full code comments or links, unfortunately, so it seems like it filters those out.
Appreciate the fix, but we'll have to wait until it's pull-requested over on the repo or updated 🤟
Im receiving this error
ERROR: Failed building wheel for pyworld
Failed to build pyworld
ERROR: Could not build wheels for pyworld, which is required to install pyproject.toml-based projects
any idea why?
Can I use on My Mac running, Os Mojave?
I m using Imac
Hey, at 15:28 i can't find the file in the model folder. Can u help me why
I have the same problem... any solution?
If you clicked the "Train" button and not the "One click training" button you have to also click the middle button "train feature index", that's what makes the .index file you need. You can do it before clicking training or after, it doesn't matter.
@@mikecameron2327 thanks :D
@Jardos_Journey While I'm in Accompaniment and vocal seperation. I am not able to get any model to select like (H5 model). Can anybody please help me find my error.
I had an interesting error come up when I started feature extraction.
This is what I was met with at the end
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/conv.py", line 309, in _conv_forward
return F.conv1d(input, weight, bias, self.stride,
RuntimeError: Calculated padded input size per channel: (1). Kernel size: (2). Kernel size can't be greater than actual input size
all-feature-done
I believe I saw this error somewhere in the GitHub page, you might wanna check there on the RVC issues tab
Hey! Need some advice; my Google Colab keeps constantly disconnecting at the end during the training process. I don't know exactly why, but I have a feeling that the GPU is becoming overloaded with the request and simply just crashes. I don't know anything about Python and pretty much follow your tutorial step by step, so trying to resolve how to fix the issue is something I need a little help with. Maybe a way to dedicate more of the GPU to the task to avoid these crashes?
I think google colab disconnect you if you use any webgui applications, they did it with stable diffusion. So you either have to pay or run it on your own machine.
Hi! Thanks for the video. It's very useful. I faced an issue during the process. I trained a model and it appeared in the model inference. However, when I wanted to copy the path to my index file, it wasn't there. On which step does this file have to appear in the folder logs/model_name? I didn't extract vocals from a video as I already had an audio file. What should I do, if I have the trained model and there is no index file? Please help me with this issue. Thank you.
Check out the pinned comment!
@@Jarods_Journey thank you for peeling my eyes =)
Is there a way to change the voice in real time using Google's colab?
Do you have to train it on multiple voice sample files, or if I had say a 5 min compilation of voice samples in 1 file would that work as well?
You gotta split long audio up as 5 minutes will cause out of memory issue for any consumer level hardware.
for me it worked well even with one single 15min file, so dont worry
First of all, thank you for the video.
some questions:
1. I can't download files from google colab, and I have no such icon in my chrome browser.
2. when I try to restore the paths of my models, it says that the directory couldn't be found... I'm not sure why. everything seems to be correct.
3. how do I take the model that I've trained in google colab and run it localy? where do the model files need to go?
Thanks again :)
1. Not sure what's happening, right clicking on the file and downloading can work in this case
2. Make sure the path is being entered in correctly and there are files in the google. If they did not export, this may be the cause.
3. To run local models, you might wanna see the local version of this tutorial to get insight on that process
how do we train more than 1000 epochs for the model?
I have a 10 minute voice dataset but it's 1 file.
Do I need to split it like 1 sentence per file or 10 sec like that?
Yes, you need to split it into 10 second audio samples or you will run out of memory during training.
Is there a way to use the model you train here in a normal text to speech engine? If yes, can you point me in the right direction for that?
Sorry, unfortunately not. You'll wanna check out my videos on tortoise TTS to RVC pipeline video
Does anyone know how to stop the voice changer from saying what I say twice?
I understand what everything here is except for what the MP4 file is, and is for. Can you elaborate on that?
Ah nevermind. I understand. I was wanting to use this to do the realtime with my microphone. You wouldn't need the MP4 file for that I guess.
Tell me how many i have to recording my voice and how long i should record on each? 15 seconds or longer? I really confused and wanted to know it so badly
When creating sample data, make sure the audio is 10 seconds or less when recording. If you have one long audio file, it will give you out of memory issue. If you want to know how to curate voice data, check out the part in the beginning of this video here: ruclips.net/video/xgvT7UnUTng/видео.html
What is the difference between 2 audio.wav file in TEMP folder and why we need to wait both of them?
Those are the outputted files from the webpage. If you don't download them, they get deleted.
I trained it for 200 epochs, how do I reuse that model to train it to 400 without having to start it from 0 epochs again?
Which one do you think is better for singing voice, so-vits-svc (and fork version) or RVC?
RVC
@@Jarods_Journey Nice! have you tried the mangio-RVC-fork?
You may be executing code that is disallowed which may terminate your runtime without warning. Colab prioritizes interactive notebook compute and disallows some types of usage when executing code without compute units as outlined in the FAQ.
Your compute unit balance is 0.Purchase more
I am continuously getting this message
Please help
why did I just set the epoch to 100, but it appears G and D files have up to a few thousand? which one is better?
The RVC epoch is actually weird in the way they named it, 100 actually means 10k epochs. The G and D files have a few thousand due to this and are named for the step count it takes to go through the data. If you're familiar with sovitssvc, this might make a little more sense. You use the G for any type of inference.
Hey! At 14:05 when I click on refresh the vocals, mine doesnt get refreshed. More files arent appearing like in the tutorial and im also getting another error besides the me_zip. I also get a "/content/drive/MyDrive/dataset/.ipynb_checkpoints: Is a directory" error. Can someone please help me?
Do you have any files that are .ipynb_checkpoints in your datasets folder? Not too sure unfortunately if files aren't appearing, could be possible some parts of the process never finished.
how much colab pro credit might be required
Hello . Thanks for sharing very helpful video. I follow your every step very seriously. I tested the training with prepared dataset consisting of two files ".wav" (human voice A). Total time is 8 minutes. The total training time I observed on Colab is only 10 minutes, which is very fast. After converting I see the best results are from female to female (quite similar to A voice about 70%). And from male to female, it is very bad. I want to ask: how many minutes should the total time of the dataset be to have the best effect? Thank you .
You may want to train longer, but 10 minutes of high quality data is good to start with, but you might want to add more if you're seeing inadequate results.
@@Jarods_JourneyI think it makes no sense to record only about 1 minutes for the dataset and then copy that 10 times over, right? :)
I have a fully completed model with 270 epoch but couldn't build on that to have more epochs ?
If you could help me how can I do that ?
Sorry looks like my original shorts response was for local installation. For colab, I'll have to get back to you on it.
@@Jarods_Journey thanks in advance ..
i got this error on export message when i tried to process data.
runtimeError:failed to load audio: ffmeg error(see stderr output for details).
end preprocess was success,but features file didn’t show up in my log.
i used audacity to convert my audio to wav file, i’ve been stuck here for quite sometime and i will appreciate any help
There is probably a cell that errored out or didn't finish running in the google collab, specifically the one that has ffmpeg in there. Try rerunning that cell and checking to see if this resolves your issue.
As long as it's a valid wav file, it should be fine. One way you know on the surface is if you look at the size of the wave file vs the size of the original audio file, the wav file should be much bigger.
what happens if the load package dataset doesn't work? it didn't work for me and im unsure why
NO I FIGURED IT OUT, I DID NOT NAME MY DRIVE FOLDER DATASET
Great Tutorial. I'm having some problems with the last part, even after using a 2 min. short audio (inference) I'm getting the timeouts and there is also no folder with the name Gradio in the TEMP folder. There are only couple of INFO messages in colab and nothing else.
Hmm, not sure what's happening here, it might not actually be processing then. Have you tried restarting runtime?
@@Jarods_Journey actually not, I only tried rerunning the cell. I'll try restarting the runtime as well, then post the results here
@@theentirecircus6623 hi ! I have the same issue. I've restarted Runtime a few times but it doesn't help. My TEMP folder is empty. Did you solve the problem?
@@obeyoutube I've just tried again with another notebook and it worked (I've waited couple of minutes after getting the timeout error). @Jarods_Journey I can link the notebook if it's okay, but it's not from the original repo
@@theentirecircus6623 Link it please
do i delete old *.pth files if my latest file is D_98000.pth and G_98000.pth??
You can to save space. Some people like to go back and listen to them to make sure they didn't "overtrain", but most cases it should be fine
i've trained a model and didn't give me the index file, how to fix it
Click the train feature index button
What's the option at the bottom left of train tab that says "Load pre-trained base model G path" and "Load pre-trained base model D path"?
Is it possible to change this to expand or update an existing model with even more epochs?
You know, since free Colab has a daily limit, you sometimes can't finish 200 epochs in time...
If this is possible, how could I go about doing this?
And would it be possible to continue said training with an expanded dataset (i.e. same audio file but updated with additional data) or would it negatively effect the results of the training?
These are the pretrained models that the authors used to start a "foundation". From what I know, you pretty much need them unless modify some things in the code etc. Since colab has a daily limit, I believe you have to export all of the trained data to your Google drive. I'm not sure if it does this using the export option, but you might have to manually do this (I might have to look into it).
Yes, you can add more data, you just need to re process and re feature extract. It has to be the same voice though
Hlo its says file not found error at pretained / f04D0k.pth pls solve my problem 😢
I dont have the window for dataset location...