📰│dev-updates

[tribal ingot]

i just shit my pants

[sweet sun]

#1159383618389540864

[lyric oracle]

Once the renovation is done, we will reset everyone's level to 0

This also included us, the staff members.

[civic veldt]

Everything is looking up at the moment and a lot of work has been done. Thank you for your patience and support!

[lyric oracle]

@here

ModelAI is now fixed! You can now use /usermodels to obtain Model Maker if you're already one on the old server.

As for this error "I'm sorry, the user you mentioned does not have any models uploaded in my database.", please wait for a bit and read the pinned message in #1159514067187277865, Thank you for your patience!

[pulsar wharf]

I'm working on people who say they don't have models, give me a few hours and it will be fixed

[lyric oracle]

For the new users, /submit and /submitkits is currently on maintenance!

[pulsar wharf]

FIX: I'm sorry, the user you mentioned does not have any models uploaded in my database..
If you have the problem that your models are not there, please use the /resync command
It is likely that due to this notice, it is somewhat slow but it is normal, but you will be able to recover your models.
(It is possible in remote cases that they do not recover, tell me about those cases so I can see them, they are isolated cases) @here

@here

[lyric oracle]

BOT IS DOWN, GIVE US A MOMENT

We will be right back

Anddd should be back now (pls don't break the bot)

[sweet sun]

New English documentation from AI HUB FRANCE

We are proud to announce that we have added an English section to our website!

You can now find guides and FAQs at
https://us.aihubfrance.fr/

We'll keep this updated with new info and guides on various Colab, Huggingface and local options (Applio, Replay, Mangio-RVC etc.).

[vague spade]

<@&1159332519167144039> we have re-enabled embed perms for ya'll on #1159289700490694666 , so that it doesn't look so ugly, feel free to make edits

[hollow zephyr]

Announcement 📢

I'm happy to announce that @sweet sun is now back online. If you wish to get the Model Maker role to be able to post in #1159289700490694666, you can use the /submit command (or /submitkits for Kits.ai models) in #1159514067187277865. You can find more info on how to submit your model in this guide: https://rentry.org/fdg_guide_new.

In case you already had the Model Maker role in the previous AI Hub server, you can use the /resync, then /usermodels command of @vague jackal to directly get your role back.

⚠️ REMINDER: For audio demos, DO NOT post copyrighted songs. Only upload the AI vocals, not the instrumental. Rule-breaking is subject to timeouts or bans.

[pulsar wharf]

I am aware of the hosting database crash, I am repairing and using a backup soon

[fiery fable]

📢 Automaze November Update

Hey there! The bot has been updated with new features and improvements, so you can have an even better experience.

New /guides command

Prefix commands for accessing the guides, such as -audio, -rvc, and -realtime, have been organized into a single /guides command. You can still use the prefix commands, but /guides is a more convenient way to access them in one place, so you don't have to memorize many commands.

Context menu commands

You can now send a guide to a user by right-clicking (or long-pressing on mobile) their profile picture, selecting Apps, and then choosing the command for the guide you want to send. For example, to send guides on how to upload models to Huggingface, you would select the "Send upload guides" command.

Final words

That's all for now! To learn more about the slash commands, use /help, and for the prefix commands, use -help. Thanks for being a part of the AI HUB community!

[ruby plinth]

Hi, im aware of the problem with Ilaria RVC.
Im so sorry this is happening and im trying so hard to understand whats wrong, seems to be an issue with rmvpe itself on Huggingface.

However, even in times like this, Ilaria is always here to bring the solution so, for now, i temp fixed it.

Ilaria RVC its using a new version of PM wich is not as good as RMVPE but its faster and generally better than other options, also, on some tests ive done it seems to be very fast (around 15-30s per conversion).

This is the link for Ilaria RVC: https://huggingface.co/spaces/TheStinger/Ilaria_RVC

If you want a faster and private Ilaria RVC space i suggest you cloning it: https://huggingface.co/spaces/TheStinger/Ilaria_RVC?duplicate=true

I hope you will have a great day, this is Ilaria, heading out.

[ruby plinth]

Working on a fix for the rmvpe, its currently broken on everywhere.

[ruby plinth]

Okay, rmvpe issue resolved, for now it seems that the only free web based working RVC its mine, Ilaria RVC.

Im sorry if for a bit yall had to use PM but it was the best solution while there was apparently a fix for the normal rmvpe.

Since i got it working and everything is up and correctly working again im pinging yall <@&1165417817814863985>, let me know any potential issue, but again, its working correctly 100%.

Please follow the links above (ill never stop reccomending cloning the space and not using the public one) and if you need any help, text your question/issue in #✨│ai-help and our staff team will be more than happy to help.

Do not also forget to live a like to the Huggingface Space, its one click, free and its the least you can do.

Love yall, have a great day/night, im trying my best.

Ilaria out

[visual osprey]

🎉 Introducing our new closed beta API!

You can now gain access to over 500,000 voice models completely free with incredible loading speeds, less than ~2 seconds for 1,000 models!
This API has enabled us to create the new version of the ultimate model search Applio bot, available both on our website and soon integrated into all our systems.

Here's how you can get started:

• 🤖 Invite Applio Bot to your server: https://bot.applio.org/
• 🔎 Explore the Applio website: https://applio.org/models
• 💾 Sign up for the closed beta API access here: https://forms.gle/gwC1ivkfxyBunNYDA

Join us in this exciting journey as we push the boundaries of voice model technology. Don't miss out on this incredible opportunity to access a wealth of voice models for your projects. Get started today!

<@&1165417631969448057>

[gray dragon]

Check out more information here:
https://applio.org/api

[visual osprey]

Try new search feature with @silent trail in #🔍│find-models

[visual osprey]

🚀 Now @silent trail is open-source, don't forget to leave your star!

https://github.com/IAHispano/Applio-Bot

https://github.com/IAHispano/Applio-API

[ruby plinth]

Sorry for the ping but its kinda important <@&1165417631969448057>

All RVCs are broken again with an issue on rmvpe side, exept Ilaria RVC wich is now FOR THE MOMENT using PM as the pitch extraction algorythm... AND IT WORKS.

Its not the best but its a quick solution, if you want to use rmvpe (little to no difference in some cases tbh) you sadly have to wait, im gonna ping again only when the issue is fixed.

Link for Ilaria RVC: https://huggingface.co/spaces/TheStinger/Ilaria_RVC (duplicate the space for no queue)

As always, Ilaria RVC>>>>>>>>>

also if you can like the space while youre there.. pwease

[ruby plinth]

For people confused, im sorry, i worded poorly, i was busy.
I meant public instances like HuggingFace Spaces and/or Colabs.

Obviosly local instances or private istances (Paperspace, Runpod,..) should (not tested personally) work fine (as they told me).

Again, sorry for the confusion.

I always talk about public ones like HF Spaces so i thought it was obvious and it was not.

Have a great rest of the day, ill update you more once the situation will get fixed.

[ruby plinth]

Hi <@&1165417631969448057> how to make ai cover?

Issue with rmvpe seems to be fixed on Ilaria RVC, dont know about others tbh but Ilaria RVC is better because what do you expect me to say? its my RVC its obviously better because its made by me owo

Joking aside, i hope you have a nice day, i love you, kisses and smooches, cuddles even

pwease pwase like the space

Link to Ilaria RVC: https://huggingface.co/spaces/TheStinger/Ilaria_RVC

Link to clone the space (reccomended): https://huggingface.co/spaces/TheStinger/Ilaria_RVC?duplicate=true

PS: in this week i will update Ilaria RVC with faster speed, new tts system (Ilaria TTS) integrated so you can use a good tts with RVC models and trying to fix Wav2Lip for the 3 souls in earth who use it lol

[ruby plinth]

Hewwo <@&1165417631969448057> its a me your favourite italian gurl!!

For anyone having problems with Ilaria RVC even with a cloned space, its possible that since it wasnt restarted you still have an older version.

To restart the space please follow the arrows on the screenshot below.
It will take around 1-3 minutes.

If you have any problem even after restarting, please tag an helper on the #✨│ai-help channel.

PS: if you liked the space, like me or idk for the love of god please please i beg you put a like on the original space (last screenhot), every like a free pizza for everyone

[formal pendant]

kalomaze is my favorite italian girl im sorry Ilaria

[soft harbor]

Ilaria is my favorite Italian girl

[humble hinge]

Hi everyone - working with @visual osprey to add a dataset maker functionality to the Applio UI. This approach takes the UVR / Demucs approach a step further: it identifies speakers and makes sure that the end result only contains the person speaking - no laughter, no claps, no random sounds etc. This works with audio files that have multiple speakers as well.

Check out the Colab Notebook and samples here: https://github.com/dubverse-ai/rvc-data-prep

Do add suggestions / bugs as issues so I can keep a better track there.

[humble hinge]

Have created this quick video on how to use the notebook on community demand: https://youtu.be/QLQ8eSGZDi8

[visual osprey]

Hi, I've been optimizing various parts of the code and organizing everything to make it easier to understand for people who want to collaborate.

We have also updated the installer so that it should now give much less problems when using a conda environment, we will continue to optimize the code, our next goal is to update and reduce dependencies.
Any feedback is appreciated privately and any contribution to the code as well.

https://github.com/IAHispano/Applio-RVC-Fork/releases

[visual osprey]

We have just updated the dependencies, now there are less dependencies than even in the original RVC, if you have any problems please let me know.

[ruby plinth]

Hi cuties, i made a program that can tell you specifics regarding an audio such as bitrate, spectrogram and so on to understand the audio better and try to make better datasets.
check it out here: https://huggingface.co/spaces/TheStinger/test

do a screenshot and send it in #🔊│ai-development with some feedback, if you would like to implement something more (its wip for now) or if something needs to be fixed!

<@&1165417817814863985>

[humble hinge]

Hello everyone,

You can now use the colab to clean local audio files and create music acapellas - all powered by AI!

Github: https://github.com/dubverse-ai/rvc-data-prep
Colab: https://colab.research.google.com/drive/1NA2GuJ2y-zRfoG3NearNiMCQa8NbidSh

[ruby plinth]

Hello babiessssssssss, Ilaria Audio Analyzer is officially OUT!
You can easily check every info about your audio simply on this page, so you can understand better if its a good quality audio for the dataset, or wich settings to use when training!
It also gives you a spectrogram, so you can see your audio analyzed.

I hope you like it and i hope it will be helpful for any new projects you have in mind!

Let me know if you want some useful application to help your trainings, your covers or even something completely unrelated, and *don't forget to put a like to the space!*💖

Ilaria out, byeeee <3

<@&1165417631969448057>

[ruby plinth]

ILARIA RVC WILL BE DOWN 30 MINUTES TO UPDATE TO THE BRAND NEW VERSION!!

Ilaria RVC is getting revamped with the new support for Ilaria TTS (best tts at the moment for HF), fixed support for ElevenLabs and Google TTS, faster inference and better looking UI and tons of small fixes.

See you in 30 minutes!

[soft harbor]

sorry if I allow myself to write here......ilaria is the best

[ruby plinth]

Hai hai hai cuties!!!!
The brand new version of Ilaria RVC is now out!

This new version supports 3 TTS, including the best one, Ilaria TTS!
It's faster, better and i removed a bunch of bugs here and there too.
Don't forget to heart the space to help me be in the main page of Hugging Face and help more people possible with the best avaliable tools for RVC!

Here is the list of changes made:

• New support for Ilaria TTS (best tts at the moment for HF, resources speaking)
• Fixed support for ElevenLabs and Google TTS
• Faster inference
• Better looking UI
• Various bug fixes
• Removed Herobrine

Until other news, tons of kisses and love, let me know if you like and i hope you will have a wonderful day!!

Byeeeee

<@&1165417631969448057>

[ruby plinth]

So many people are using Ilaria RVC that it literally got taken down for too much usage lmfao, we reached the sky

Clone it to have private and faster! https://huggingface.co/spaces/TheStinger/Ilaria_RVC?duplicate=true

its up again :3 <3

[ruby plinth]

ILARIA CONVERTER IS FINALLY HERE!

Do you have in mind when you browse Google Image and you finally see that image that is PERFECT for you?
You immediately download it and... its a webp image.
You cannot use it.
You start swearing ancient words so much that Chthulu emerges from his eternal sleep.

Well... not anymore!

With Ilaria Converter you just need to drag and drop an image or an audio (there are 2 separated tabs!), choose the new extension you prefer and done!
In one second you have what you want, immediately, super fast, for free with no limits and uhh... free!

Theres also opions if youre a nerd and want an audio in a different bitrate or if you want to change the resolution... the limit is the sky!

Try Ilaria Converter and dont forget to put a like on HuggingFace, it will help me a lot!

<@&1165417631969448057>

https://huggingface.co/spaces/TheStinger/Ilaria_Converter

[ruby plinth]

haiiiiiii Ilaria Audio Analyzer now offer support to download audios from Google Drive directly from the application,
so you dont have to upload them manually anymore, just put the public link of the audio and automagically it will be inserted into the app to check every aspect of it.

Hope you like the new feature!

https://huggingface.co/spaces/TheStinger/Ilaria_Audio_Analyzer

[hollow zephyr]

Hello

Friendly reminder that you can use the </report:1165908028445429860> command of the @molten tiger bot to report any rule-breaking / problematic behavior.

For transparency reasons, this command can only be used by <@&1159345504442925056> (level 1).

That's all for now

[visual osprey]

Hello, we are aware that the .pth generated in Applio can generate incompatibilities and that is because the rest of the programs are running older versions.

In Applio we are betting on updating dependencies to improve performance and by updating Torch can improve performance x2 in many cases, the new way of saving that has can generate incompatibilities, so we also added in the Processing section, an option to convert your model to the version without optimizations and compatible with other programs. We will also look for ways to make it more compatible without losing optimizations.

[ruby plinth]

Fixed an issue on Ilaria RVC.

Should work as usual if not better.

[visual osprey]

@stiff elk has published a new update of Applio, in which now when saving the model it will apply directly the conversion to the format without optimizations, soon we will add a checkbox for those of you who want to have your models with optimizations.

I invite you to also ask other developers to consider updating the dependencies of their programs if you really want to have a better and better AI!

For those of you who already had a model within Applio, go to Extra -> Processing -> Model conversion

[past marlin]

<@&1159332519167144039> <@&1165417631969448057> Yo, a new update guys! My name is SimplCup, and i'm the new developer of RVC pretrained models for the RVC. I work together with @west reef @visual osprey and @urban grove to make RVC better and faster :3

I'd like to present you a new test version of the pretrain Ov2! I've made it myself improving overall quality of it.

It's made on 40k sample rate with 18 new voices, which gives it a lot of variety.

To use it, you'll need to copy and paste both .pth files into pretrained_v2 folder of your RVC. Then, in your rvc you'll need to choose 40k sample rate V2 (all sample rate version will be released once i make the final version, this is just a test version). After you've preprocessed your dataset and extracted features, you'll have to put in the names of the pretrains as shown on the second screenshot (don't mismatch the G and D). After that, you can begin training your model.

Here are rough guidelines for the model training on this pretrain:

• Minimum length for a ideal model is 1 minute, but you can train good models even with 10 seconds datasets on this pretrain
• The models trained on this model don't require huge epoch count: usually for 3-5 minutes 40-60 can be enough, for 1-3 minutes, 60-100 is enough and for 10-60 seconds 200-300 is enough
• Use clean datasets to get the best sounding results :3

Also please give your reviews in my dms once you train some models with it, so that i'll be able to make it even better! Hope you'll like it :3

Here's the huggingface rep where you can download the test pretrained models: https://huggingface.co/ORVC/Ov2Super/tree/main

[visual osprey]

Do not hesitate to consider supporting @past marlin, training models costs a lot of time and resources. Soon we will bring new models and in more frequencies!

https://boosty.to/simplcup/donate

[past marlin]

[sweet sun]

<@&1159332519167144039> <@&1165417631969448057> sorry, his ping didn’t go thru so I am pinging you lot!

[past marlin]

Btw, here's how my pretrain sounds in comparison with the regular V2 pretrain. As you can hear, my test pretrain has a lot more clarity nad less noise :3

[past marlin]

do remember tho, that this is a test version, so if it'll have any flaws you can report it to me in dms so that next time i can improve it even more, ily y'all

[fierce storm]

RVC Disconnected Update
Alright, you made yourselves loud and clear!
I've added an option for the new Ov2Super 40k pretrains to RVC Disconnected. Enable the use_new_pretrains value (which is enabled by default), set your sample rate to 40k and v2, and the trainer should automatically use the new pretrains.
Please let me know if there's any issues regarding this update in #help-rvc .
https://colab.research.google.com/drive/1XIPCP9ken63S7M6b5ui1b36Cs17sP-NS

[visual osprey]

Applio's new generation

Hello everyone, I have good news. Applio is soon going to undergo one of the biggest changes in RVC.

We've been working for a while on developing a new RVC client from scratch based on a Gradio interface, as we've done before. This will allow us to have better code, enabling us to bring more frequent and improved updates.

As you know, Applio was a Fork of Mangio, which itself was a Fork of RVC. It may sound a bit confusing, and the code is indeed quite complex. The code has passed through the hands of many developers, making coding practices challenging. That's why the Applio interface will now be powered by CORE, which is a library containing all the RVC code.

We will soon open beta testing, and your help will be crucial as many parts of the RVC code have been enhanced. We expect the performance to be better than that of the RVC mainline due to improvements in coding practices.

So far, we have taken into account all your feedback (much of which couldn't be addressed due to the codebase we were working with), and everything will be applied to the new version, Applio V3. Initially, it will be released with fewer options than the current version, but at least the essential ones. Gradually, we will add all the features.

[past marlin]

32k test version of Ov2Super is up! You can use it the same way as the 40k but with 32k datasets, cus many people been asking specifically for this sample rate :3

https://huggingface.co/ORVC/Ov2Super/tree/main

[past marlin]

it looks like the one that i posted doesn't work on mailine or mangio rvc, but @runic dust made a fix, so if you want to use the 32k version download this instead

https://huggingface.co/poiqazwsx/Ov2Super32kfix/tree/main

[visual osprey]

Ready to get to know Applio V3? Next week, beta testing starts 🚀

bye mangio 😢

[visual osprey]

All Google Drive download systems are down due a recent Google update, fixed on Applio

For developers, consider use and apply the Applio-RVC fixes https://github.com/IAHispano/gdown

we are working on improve the library

[visual osprey]

Re-named repo https://github.com/IAHispano/Applio

[ruby plinth]

google link no work ilaria rvc for work i dont know if you help i happy thank

[visual osprey]

[visual osprey]

We start beta testing soon, small spoiler of what's coming... 👀 it is not yet available

[gray dragon]

2023 has been a good year, let's start 2024 well.

Log in right now and review everything you have achieved this year with Applio!

https://applio.org/tour

Tour 2023 will close on January 22nd.

[visual osprey]

see you soon

[visual osprey]

Hello everyone! As you may know, the release of Applio v3 is just around the corner. If all goes well, this <t:1705705140>, we'll be unveiling the trailer, followed by the release. It has been many weeks of hard work, and I'd like to clarify some key points about the update.

• In terms of performance, the change won't be drastic. The entire interface is more fluid, but in terms of RVC, the speeds are similar. The initial phase of development focused on adapting the code, with the next step being optimization (some things have already been applied).
• It will be launched in about 20 languages, so we'll need people who can correct translations as it will be done using an automatic system.
• The Gradio interface and RVC code now function separately, making it easier to implement new features more frequently.
• Approximately 20 libraries have been reduced compared to RVC and around 140 less than Mangio. This has resulted in a lighter and easier-to-install version.
• We are working on incorporating TTS on the release day with some improvements and considering the option of adding other extra features. I leave it in your hands to give us ideas.

On Friday, we'll delve into more details, and I'll share more information. I just hope you appreciate all the effort behind this update. We had to reconstruct the RVC code from the official repository, improve it, and adapt it to an interface, adding some features that many of you will enjoy. Applio will be released with some less features than now, but better made and will be incorporating everything necessary over the days completing the beta testing. Meanwhile for those who like the previous versions will still be available

Starting from the release day, Applio will no longer be a fork and will begin to take its own initiative in development without depending on any other repository. See you on Friday!

[ruby plinth]

Ilaria 64 will be offline and/or having problems for a bit since im adding some new stuff.
Apologies

[visual osprey]

Hey, I remind you that tomorrow at this time <t:1705705140:F> the Applio update will be released. Version 2, like version 1, will continue to be available for those who want to continue using it. It is possible that Applio will not turn out perfect since there may always be some factors that we miss but we will try to correct them quickly. I am excited because you can see the trailer and we can start a new path in the development of this technology that we love so much

You can activate the notification https://www.youtube.com/watch?v=nxqo6kOhsBI

[visual osprey]

Applio V3 - Premiere now on YouTube
https://www.youtube.com/watch?v=nxqo6kOhsBI

<@&1165417631969448057> <@&1159332519167144039> <@&1159537820352917516>

[visual osprey]

Regarding installation problems, it happens on some computers. In a few hours we will release a pre-compiled zip

There was a small problem when uploading the release, everything should be solved

For those of you who say "it should have been tested" it was when uploading it to GitHub, so there is little we can do

[visual osprey]

UI Colab by @midnight dagger ❤️
https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb

[visual osprey]

For all of you who have any error, it is very important that you file an issue, so we can identify and solve the error. And from there the stable version of the new version will begin, https://github.com/IAHispano/Applio/issues

[stiff elk]

There is an ongoing issue with training models on the latest version of Applio; the final .pth file is not generated correctly, causing a break in the process. As a result, the inference is limited to static audio. We are actively working to resolve this issue.

[visual osprey]

Hey, all the bugs reported so far have been fixed. We are investigating some issues with training, we recommend to wait for the moment.

Thanks for all the feedback and reception you are giving to the update, we will keep working on improving all those things that may go wrong, sorry for the inconvenience.

Encourage developers to participate in the repository, we are 2 people and there are many things we have ideas to do!

[west reef]

Ladies a gentlemen get ready for the next challenger - RIN_E3

We didn’t get the RVC V3, so we made a replacement for it.

Yaharo! my name is MUSTAR, I'm just a random guy, I've made a pretrain
Working together with @past marlin @visual osprey @urban grove @ashen root

RIN is a totally new English pretrain made from a scratch.

Specs
dataset: 140 hrs
epochs: 50
steps: 1040928
sample rate: 40k

hardware trained on: rtx 4080, a100

i didn't done a lot of tests\models, so feel free to compare pretrains and experiment

Download link-https://huggingface.co/MUSTAR/RIN_E3/tree/main
zip link-https://huggingface.co/MUSTAR/RIN_E3/resolve/main/RIN_E3.zip?download=true

Enjoy

now we need a ping

[ruby plinth]

<@&1165417631969448057>

hey i tagged you becaus the new pretrains are good, i did it for you :(

i love you btw

[stiff elk]

We finally fixed the training issue with Applio it should work perfectly fine now, please download the new release https://github.com/IAHispano/Applio/releases/tag/3.0.3

[stiff elk]

We have created Applio precompiled with all the dependencies installed to ensure a seamless experience.

Download the zip file, extract it, and then open the 'run-applio.bat' file.

Download Applio V3.0.3 Precompiled: https://huggingface.co/IAHispano/applio/resolve/main/Applio V3 Precompiled/ApplioV3.0.3.zip?download=true

[fierce storm]

RVC Disconnected Update
Heh, late as always.
I've added support for @west reef 's RIN_E3 pretrained model to my notebook. You can use them by setting pretrain_type to RIN_E3. (OV2Super and the original pretrains are still supported!)
Please let me know if there's any issues regarding this update in #1192011222023950368 .
https://colab.research.google.com/drive/1XIPCP9ken63S7M6b5ui1b36Cs17sP-NS

[humble hinge]

Hi everyone!

We have open-sourced the 🔥text to speech models🔥 for English and major Indian languages. This supports elevenlabs-like zero shot voice cloning as well.

Expect a scaled model packed with 1000s of hours of training data coming to your doorsteps soon!

This uses a Hubert encoder like RVC, so the plan is to use this for RVC text to speech in the coming weeks 🫰

Do take this baby for a spin 👀 There's a colab link for y'all: https://github.com/dubverse-ai/MahaTTS

[ruby plinth]

Ilaria RVC on Colab really soon..

[gray dragon]

Small reminder that Applio can be used in Google Colab with UI, If your pc does not meet the requirements to use RVC you always have this option:
https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb

[ruby plinth]

Ilaria RVC.

On Colab.

Now.

Fastest inferece.
Easiest Interface.
Always updated.
New features added weekly.
Made by a woman.
Zero Problems.

<@&1165417631969448057>

Try it now: https://colab.research.google.com/drive/16LkwvFZeudTpUOsE_6bMjOq2qkxFo8Hr?usp=sharing

[stiff elk]

I uploaded a new Applio Precompiled with a lot of new fixes!

Download it now (Version 3.0.4): https://huggingface.co/IAHispano/applio/resolve/main/Applio V3 Precompiled/ApplioV3.0.4.zip?download=true

[ruby plinth]

ILARIA RVC.

COLAB.

NOW.

Not like the others.

More feature coming soon.
Request new features here #🔊│ai-development.

<@&1165417631969448057>

https://colab.research.google.com/drive/16LkwvFZeudTpUOsE_6bMjOq2qkxFo8Hr?usp=sharing

[patent sphinx]

Applio bot is down, the devs are investigating the issue, please be patient

[visual osprey]

is back

[ruby plinth]

Ilaria RVC had planned new interesting features such as faster load times, training, ilaria audio analyzer included, importing personal pth and index files directly by drag and drop (so you dont need downloads), different UI fixes and updates, and general improvements.
It was set to release this sunday as a small beta update.

Unfortunately tho my mental health rapidly declined, taking me to a really, really low point.
so from now on ilaria rvc update is being suspended for the time being.

i know this may be upsetting for everyone and i apologise and promise to work on it as soon as i can and i start recovering.

in the meanwhile i ask you for me to be in your thoughts and prayers.

I apologise again, we will come back stronger and better than ever, it cant always rain on the same place.

Until now, have a great day and be the awesome person you are!

• Ilaria 💖

[stiff elk]

We've created a Hugging Face space just for you, allowing you to seamlessly plug and play with Applio online! To enhance performance, consider duplicating the repository. Also, remember that training on Hugging Face may not be possible due to the absence of GPU support.

https://huggingface.co/spaces/IAHispano/Applio

[visual osprey]

Hello everyone!
I would like to know what things you would like us to introduce in Applio, currently it is already in a quite stable phase where we can focus on some features, I will tell you the next news and listen to my DMs in case you have any ideas!

Planned features

• Plugin system (You will be able to add plugins made by the community)
• Discord RPC that will change according to what you do in Applio
• Autotune
• Training assistant for training issues & overtraining
• Theme customizer + Updater
• Installer .exe with UI
• PDF to TTS
• TXT to TTS
• Integration of more TTS models
• Pause training
• Model fusion

[stiff elk]

I uploaded a new Precompiled with a lot of new fixes!

Download it now (Version 3.0.5): https://huggingface.co/IAHispano/Applio/resolve/main/Applio V3 Precompiled/ApplioV3.0.5.zip?download=true

[alpine swallow]

To all colab users: Gradio is Down! you can check the uptime here:
https://status.gradio.app/

[visual osprey]

Applio API is under maintenance due to an update, it will be back soon.

[alpine swallow]

Gradio is Back up!

[stiff elk]

Exciting news! Applio's latest and enhanced version, 3.0.6, is now ready for download. Explore the improvements we've made and elevate your Applio experience!

Choose the right option for you:
🚀 Release: Download now

or

📦 Precompiled: Download now

Upgrade to the latest version and unlock an even better Applio experience!

[stiff elk]

We've fixed the issue that incorrectly indicates your local version and prompts you to update to version 3.0.6 in the precompiled version. If you've already downloaded it, simply grab the updated precompiled zip again. Alternatively, you can edit the 'assets/version_checker.py' file and change the line from 3.0.5 to 3.0.6 yourself.

[runic dust]

Huggingface may be down right now you can check the status at:
https://status.huggingface.co/

[past marlin]

[stiff elk]

👀

Applio Elevenlabs TTS Plugin
This plugin allows you to use the Elevenlabs TTS model to synthesize text into speech. Remeber that elevenlabs is a multilingual TTS model, so you can use it to synthesize text in different languages. Languages supported by the model: Chinese, Korean, Dutch, Turkish, Swedish, Indonesian, Filipino, Japanese, Ukrainian, Greek, Czech, Finnish, Romanian, Russian, Danish, Bulgarian, Malay, Slovak, Croatian, Classic Arabic, Tamil, English, Polish, German, Spanish, French, Italian, Hindi and Portuguese.

[stiff elk]

Introducing our latest feature, the Elevenlabs-TTS plugin! Get it now by clicking the link below:
Download Elevenlabs-TTS

To install, follow one of these two methods:

1. Quick Install:

   • Download the plugin zip file.
   • Go to the plugins section in Applio.
   • Upload the zip file directly.

2. Manual Install:

   • Download the zip file.
   • Extract its contents.
   • Move the extracted folder to the tabs/plugins/installed directory in Applio.

Remember to restart Applio after using either method. If Applio was already open, the second method requires two restarts.

[nimble tapir]

https://twitter.com/huggingface/status/1756401140495679871

Hugging Face (@huggingface) on X

We're having some issues on the Hub and looking into them! We should be back soon.

Please send hugs🤗

Twitter · Feb 10, 2024, 7:33 PM

...a minute later

Huggingface is apparently back, in case you wanna check, here's the status page: https://status.huggingface.co/

UPDATE, it's back down

[nimble tapir]

UPDATE: Huggingface is back again.

https://status.huggingface.co/

[jolly vault]

IT'S DOWN AGAIN 🦀

[sweet sun]

Best bet is to wait a day or so to let HuggingFace get back to it's usual state, by then they should've fixed their stuff. <3

[stiff elk]

Since Hugging Face is currently experiencing issues, I've uploaded the precompiled Applio to Mega. If you'd like to download it, you can find it here: Mega Link

[nimble tapir]

UPDATE: Huggingface should be working now!

[visual osprey]

@silent trail is being updated...

[visual osprey]

Update completed, the /search command now has some new features and should be faster

Maybe it's a bit bug, throughout this day I will improve it

[visual osprey]

gdown is down

https://github.com/wkentaro/gdown/issues/325

[visual osprey]

If you are experiencing some problems in Applio Colab right now it is normal, we are working on adding some new features :)

It should be working now

[shadow wigeon]

Well, it looks like google fixed the problem I had, bringing gdown back to life.

[visual osprey]

gdown isn't down

[soft harbor]

Ilaria rvc and other colabs may not be working at the moment

if you have a error similar to "module 'numpy' has no attribute you can download a old version of numpy for example:
!pip install numpy==1.23.5
or you can go into the code and change all the things that have depreciated like np.int into int

here are some temp fixes from @runic dust until they are updated (It may take a week or so)
Ilaria rvc
Aicovergen ui

[quick whale]

Alternatively, you can replace np.int in the vc_infer_pipeline.py file (right inside the Ilaria RVC folder, in the src subfolder in AI Cover Gen) with int)

You can do this command (after you've cd to the RVC folder):
For Ilaria RVC:

!sed -i 's/\<np.int\>/int/g' vc_infer_pipeline.py

For AICoverGen:

!sed -i 's/\<np.int\>/int/g' src/vc_infer_pipeline.py

Now, depending on the fork, the file name can vary, but it's mostly the same.

[quick whale]

Here's the root cause of the error: https://github.com/numpy/numpy/pull/22607

TL;DR: np.int is deperecated and REMOVED from Numpy (the library that RVC depends on), and the builtin int type made it redundant.

[ruby plinth]

Hi everyone, Ilaria here.

Sorry if lately the colab/space are having problems.

Unfortunately im very sick and im bouncing off hospitals in the last weeks, so forgive me but unfortunately in my life Ilaria RVC is not my top number one priority right now.

I apologise if this is an inconvenience to everyone, and while i get it can be upsetting please be patient while our wonderful team is working on temp fixes for the problem.

I am very, very sorry, i wish i was on PC to learn the problem and fix everything but i cant phisically do that for the time being.

In the meanwhile i want to thank Hina for the wonderful work and port and everyone who is currently looking for fixes.

Im sorry for not being here, i promise i will once i get phisically and menrally better.

As always, stay fresh and i love you.

• Ilaria 💖

[stiff elk]

The latest Applio version, 3.0.7, is now available! This is a compact release with improvements in training (fixing the ckpt issue), introducing new inference modes such as fcpe and hybrid, along with some general fixes.

Choose the right option for you:
🚀 Release: Download now

or

📦 Compiled: Download now

Upgrade to the latest version and unlock an even better Applio experience!

There's been a bug that says "your version is outdated" to fix it just update the config.json in the assets folder and change the version number form 3.0.6 to 3.0.7

(You can also download again applio to fix it)

[native granite]

UPDATE

@midnight dagger Fixed AICOVERGEN UI & Ilaria RVC Colabs, so dont use the temp fix ones that were in the previous message of angetyde anymore!

[visual osprey]

Training in Applio's Colab may be broken for the next few hours due to the new update we are preparing, we recommend waiting to train until we release the update 😄 - fixed now

[stiff elk]

🚀 Exciting News! Upgrade to the latest and greatest Applio version, 3.0.8, now available for download!

Choose your preferred option:

• 🚀 Release: Download now
• 📦 Compiled: Download now

Unlock an enhanced Applio experience with these fantastic updates:

Applio Changelog Highlights:

• 🎤 New and improved Voice Blender option, now better than ever.
• 📋 Detailed parameter descriptions in both training and inference for better clarity on actions and their impact on results.
• 🔊 Advanced audio cleaning system introduced for enhanced clarity in spoken audio or text-to-speech (TTS).
• 🔄 Additional inference options based on valuable community feedback.
• ⏩ Inference times significantly reduced! Optimized code for faster results.
• 🚫 Stop training directly on the web UI.
• 🛠️ General Fixes

What features are you hoping to see in our upcoming updates? 🔍 Stay tuned for more exciting improvements!

[runic dust]

Huggingface may be down right now you can check at https://status.huggingface.co/

[stiff elk]

A recent issue with the model blender feature in Applio has been identified and promptly addressed. The bug has been successfully resolved, and to take advantage of the fixed functionality, we recommend reinstalling Applio. Your experience with the model blender will now be seamless and free from the previously encountered issues.

Additionally, please be informed that Hugging Face is currently undergoing maintenance. To ensure uninterrupted access for users of the compiled version, we have provided an alternative link for downloading Applio.

[stiff elk]

Ignore the alternative download link, mega has a downloading quote so we recommend waiting till huggingface is up again.

[native granite]

It's back online!

[native granite]

Everything should be fixed

[ruby plinth]

Hello cuties, Ilaria here.

HuggingFace contacted me congratulating me for being in the Top 100 impactful AI developer!

im on the position 46.

I want to thank everyone personally for helping me reach this amazing milestone, it was and it id an incredible journey that i hope we will continue together.

I can’t wait to get better and come back to do other amazing things together!

Per aspera, ad astra.

• Your friendly neighbour girl 💖✨

[visual osprey]

We have updated Applio's Colab, now better than ever with an extra section ported by @runic dust, and the encryption method has also been strengthened

https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb

[visual osprey]

Soon in Applio all your models will contain a hash that will act as the unique ID of your model. This will prevent anyone from stealing your models since each model will have its hash that will be built from the number of epochs, steps, date and model information. I am also adding additional information so you can better see how a model is built (no one can lie to you anymore)

[ruby plinth]

ILARIA RVC VIDEO TUTORIAL

Since a lot of people asked for a video one instead of a written one, here it is.
Keep in mind the video was translated, dubbed and subbed by an AI from an italian video i made, so it can contain word mistakes.

Hope you like the video and find it useful :) <3

https://youtu.be/wh4cLMLLkj0

[ruby plinth]

Ilaria RVC will be soon avaliable with Gradio 4.2.0 allowing users to login to their instances to keep their models and audio downloaded.

[native granite]

UPDATE FOR ALL GOOGLE COLAB THAT HAS UI (INTERFACE)

Currently, google colab is detecting the ui again, so when you are using colabs like ilaria rvc, applio, hina wokada, etc. For around 10 mins, it will disconnect you from the runtime, we are working on a fix for this, meanwhile you can use no ui ones such as rvcdisconnected

[visual osprey]

https://tenor.com/view/axolot-cute-smile-gif-13234340

[visual osprey]

should be fixed on applio, @shadow wigeon 🙏🏻

[ruby plinth]

update on Ilaria RVC;

it appears that the down is currently only for some geo-locations.

in some countries its up and running, in the meanwhile we are investigating the issue.

[ruby plinth]

I'm trying to understand the problems reported by few users regarding a disconnection happenin in Ilaria RVC after 20/30 minutes.

Do you currently have problems with Ilaria RVC?

Vote below please

[sweet sun]

<@&1165417817814863985> ^^

[ruby plinth]

for those who have problems: text your problem in #✨│ai-help so i can address it immediately

[ruby plinth]

Ilaria RVC Small update

1. It now seems that the limitation is about max 30 minutes per day, per google account (you can switch it without problems) wich is super okay since its only inference.

2. Small changes in the GUI.

3. Downloads from Google Drive are now fixed.

4. The first process (the installation) is faster by 50% at least.

5. the "torch" problem seems to be fixed.

if you encounter any other problems, let me know in #✨│ai-help

[ruby plinth]

You have Paperspace?
you wanna still use Ilaria RVC?

pre-compiled notebook for you

[vague spade]

New Text-To-Speech (TTS)

Introducing GSVI, an experimental new fork of GPT-SoVits by @bronze cradle . This text-to-speech [TTS] voice cloner offers an API as well as added features and optimization for easier use.

Do try it out, and feel free to leave feedback in #🔊│ai-development. We’d really appreciate testers, contributors and translators as well.

Link: https://github.com/X-T-E-R/GPT-SoVITS-Inference

Demo: English / Chinese

Special thanks to @patent sphinx for the english translation. <@&1165417817814863985>

edit; oops wrong channel link, should be fixed now

[bronze cradle]

I am currently editing the document. Please do not directly follow the methods on the readme for installation. Instead, please directly install prezip (I am in the process of uploading it, and it will be sent to Hugging Face by tomorrow):

[bronze cradle]

Finished Editting! Have a Try Now!

[bronze cradle]

https://huggingface.co/XTer123/GSVI_prezip/blob/main/GSVI-2.2.4-240318.7z

[visual osprey]

Due to lack of time, motivation and little improvement margin, Applio will stop receiving updates so frequently, the project is at a good point and we will take the opportunity to focus our efforts on a new product that will change the way we use RVC.

If you have any suggestion of something you would like to add to Applio let us know (I have the DMs open), soon I will tell you more about the new project.

[visual osprey]

about this, we were considering the option of integrating it, but the version of Gradio higher than the one we use has problems when loading the audios, for our part we will wait for them to solve it...
more info here

[ruby plinth]

^i can confirm, and i forgot to update that its put on hold until it will be found a way.

[ruby plinth]

OKAY OKAY ILARIA RVC NEWS

i have one bad news and one good news (and the stupid news)

bad news: ilaria rvc (3.0; the current one) will be deprecated and not updated anymore

good news: im switching directly to mainline, so this means even faster inference and training with the fasted avaliable

stupid news: it wont be called 4.0 because in italy 4 is a bad luck number so it will be called ilaria rvc insertcoolnamehere *

*tba

[ruby plinth]

I ░ L ░ A ░ R ░ I ░ A ░ ░ R ░ V ░ C ░

░ I ░ N ░ ░ B ░ I ░ O ░

(im gonna try a small update tomorrow, ill let you know if successful)

[ruby plinth]

Forgot to update,

Ilaria RVC: The Mainline Update is currently being worked on.

[ruby plinth]

update: it works!!!!!!

but its still a long way on the road.
it miss the ✨ ilaria touch ✨

btw consider supporting me if you like ilaria rvc (link in bio) and i will be forever grateful and i will consider putting the nicknames in the program :p

[visual osprey]

Upgrade to the latest and greatest version of Applio, 3.0.9, now available for download!

Choose your preferred option:

• 🚀 Release: Download Now
• 📦 Compiled (Just double click it): .exe / .zip

Applio Changelog Highlights:

• 🔑 Unique model hashes now included in each save for better traceability.
• 📊 Step saving in the model name for detailed tracking of the training process.
• 🔍 Added search function for improved navigation and model access.
• 🖥️ Training tab automatically disabled when compatible GPU is unavailable.
• 🆕 Version 1 of pretrains no longer required.
• 🛑 Fixed stop training button.
• 🛠️ General fixes.

What features are you hoping to see in our upcoming updates?

[ruby plinth]

Ilaria RVC Mainline Beta 1 was given to a bunch of Beta Testers!

[ruby plinth]

For those who are interested and wanna follow the developmente step by step, i made a Trello, so im more organized and you can watch the development unfolding :3

https://trello.com/b/tU4k9jFn/ilaria-rvc-mainline

[ruby plinth]

Ilaria RVC Mainline coming very soon

Passed every inference test, now working to perfect the training part, will take some time but we are working to give the users finally the best option.
I hope you will appreciate our effort <3
I will stop updating so frequently, i apologise.
I will update only for important things.

That being said...
A tab with all the people who helped the project will be made, to rightfully credit every person who helped me on this nightmare journey.

I love you all,

Ilaria out. 💖

[visual osprey]

For the future update we are experimenting with a system so that you can know at what point in the training the lowest point on the tensorboard occurred. Although this is not a super precise indicator, it can be useful to be able to easily identify how it is progressing your training!

[ruby plinth]

Ilaria RVC Mainline was given out to all staff members interested (if i missed someone DM me)

Public beta very soon :3

[visual osprey]

In the next update (already available in GC) in Applio you will be able to have more control of your trainings.

Set the number of epochs from which you want to stop training if no improvement is detected and go have a coffee while Applio takes care of the rest.
This new system compares your current results with the best result you obtained to prevent your model from overtraining. As always, this can vary a lot since regular graphs are not always available, but it gives you an overview!

I will be improving this system constantly, right now it is in beta. My idea is to automate as much as possible and soon I will achieve it. Leave me your feedback by DM!

[ruby plinth]

HEWWOOO ily all ✨

very happy to announce that Ilaria RVC Mainline will be ready for a public beta very soon!

we are still findinf ways to better the program and make it easier for users!

Ilaria out 💖

[visual osprey]

Hey, for all of you who use Applio it would be very helpful if you could answer this form. It's only 1 minute and it will help us to improve the current version of Applio and the big update we will have before summer.

https://applio.org/feedback

[ruby plinth]

Ilaria RVC Mainline Beta 1 is here! 💖

Get ready to turbocharge your RVC experience with our lightning-fast beta program!

After rigorous testing by our dedicated staff members, we’re thrilled to unveil Beta 1.

Remember, it’s a work in progress, so keep your version up to date using the “Ilaria RVC Assistant” one tap, and you’ll have access to all the pretrain and updates you desire!

Windows, Mac, Linux: Download and instruction
Google Colab: Link

Small list of features:

• Fast Inference Page
• Streamlined Model Training
• Improved TTS
• Training Helper, Model Information, Audio Analyzer
• Better Model Import & Model Downloader

PS: Don't forget to check the github release to see the new features!

PPS: Sorry! The previous message was accidentaly deleted, heart this message too again to receive my eternal gratitude and keep this project FREE!!

*Exciting updates are on the horizon, but for now, here’s a sneak peek of what’s brewing! Stay tuned

[alpine swallow]

*Google drive isnt available in colab yet, we're working on that!

[civic veldt]

Quick follow up, @runic dust managed to implement an auto back-up system. Enjoy :3

[alpine swallow]

Autobackup system now implemented in Base64 encoding, see "Base 64" section of colab :D

[alpine swallow]

Another quick changelog: Added drive downloads from backup whenever you boot the colab :D (currently in rot13 only)

[visual osprey]

Applio's audio analyzer has been updated, now you will have more information about your audio!

[ruby plinth]

In Ilaria RVC Mainline now the refresh button also refresh the audio folder, so it can scan for new audio files!

We are working on small quality of life updates, so feel free to DM me with suggestions on how to improve it for your likings!

To update to the latest version, run the Ilaria-RVC-Assistant.bat and select 1 to update, it will automagically update every file for you!
The change is automatic for Colab users!

PS: later today we will also implement the upload dataset function to the train tab!

Hope you like the Beta so far! 💖

[stiff elk]

🚀 Exciting News! Upgrade to the latest and greatest version of Applio, 3.1.0, now available for download!

Choose your preferred option:

• 🚀 Release: Download Now
• 📦 Compiled (Just double click it): .exe / .zip

Unlock an enhanced Applio experience with these fantastic updates:

Applio Changelog Highlights:

• Enhanced audio analysis tool, offering a comprehensive array of data insights. 🎧
• Implementation of an overtraining detection mechanism: Specify the desired number of training epochs; if no advancements are detected, Applio automatically halts training, allowing you to step away and enjoy a break while it manages the process. ⏳
• Facilitated model export and file management: Seamlessly download your trained .pth and index files directly within the Applio interface, or conveniently store them on Google Drive (exclusively available in Google Colab). 📤
• Various optimizations and bug fixes for improved performance and stability. 🛠️

What features are you hoping to see in our upcoming updates? 🔍 Stay tuned for more exciting improvements!

[silent trail]

🚀 Exciting News! Upgrade to the latest and greatest version of Applio, 3.1.1, now available for download!

Choose your preferred option:

• 🚀 Release: Download Now
• 📦 Compiled (Just double click it): .exe / .zip

Unlock an enhanced Applio experience with these fantastic updates:

Applio Changelog Highlights:

• ⚡ Significantly upgraded inference time, now 50% faster.
• 🔊 Rectified output audio format.
• 🛠️ Various general fixes and improvements.

What features are you hoping to see in our upcoming updates? 🔍 Stay tuned for more exciting improvements!

[ruby plinth]

ITAILA IS OUT!

Itaila is a 32k pretrain that allows users to make models with super small datasets!
The results as of now seems to be amazing.

Link: https://huggingface.co/TheStinger/itaila/tree/main

Attached we have two audios;

1. 2 minutes dataset, 150 epochs, 8 batch size, very noisy (almost not cleaned)
2. 2:30 minutes dataset, 150 epochs, 4 batch size, very clean (podcast like audio)

For now only italian models were tested, since the pretrain was trying to achieve that but it possibly works well with spanish and other languages.
Just try i and give me feedback on #🔊│ai-development

Ilaria, out! 💖

[west reef]

SnowieV3 RU-JP Pretrain!!

1. SnowieV3 40k https://huggingface.co/MUSTAR/SnowieV3.1-40k/tree/main

2. SnowieV3 48k https://huggingface.co/MUSTAR/SnowieV3.1-48k/tree/main

3. SnowieV3 32k https://huggingface.co/MUSTAR/SnowieV3.1-32k/tree/main

4. SnowieV3 X RIN_E3 40k https://huggingface.co/MUSTAR/SnowieV3.1-X-RinE3-40K/tree/main

5. SnowieV3_JP_Hubert https://huggingface.co/MUSTAR/Snowie_V3.1_JP_hubert

**Pretrain "Snowie" helps improve a model's Russian and Japanese speech \ makes English-speaking models speak Russian and Japanese **

Dataset size: 58 hrs
34 womens speech
24 males speech
20 different voices

How to use:
1 Download G and D pth files from Huggins
2 Place it in the pretrained_v2\custom pretrained folder
if you have Applio, upload it directly (but you can also send it to the folder)
3 Paste the path to the pretrain in the fork (standard fork or mangio fork)
4 You're ready to Rock!

СнежокV3 РУ-JP претрейн!!

1. SnowieV3 40k https://huggingface.co/MUSTAR/SnowieV3.1-40k/tree/main

2. SnowieV3 48k https://huggingface.co/MUSTAR/SnowieV3.1-48k/tree/main

3. SnowieV3 32k https://huggingface.co/MUSTAR/SnowieV3.1-32k/tree/main

4. SnowieV3 X RIN_E3 40k https://huggingface.co/MUSTAR/SnowieV3.1-X-RinE3-40K/tree/main

5. SnowieV3_JP_Hubert https://huggingface.co/MUSTAR/Snowie_V3.1_JP_hubert

Претрейн "Снежок" помогает улучшить Русскую и Японскую речь модели \ заставляет англоязычные модели говорить на русском и японском

*Претрейн нужен для создания моделей

Размер датасета: 58 hrs
34 женской речи
24 мужской речи
20 разных голосов

Как использовать:
1 Качаете G и D pth файлы с хаггинса
2 Кидаете его в папку pretrained_v2 \ custom pretrained
если у вас Апплио - закидываете напрямую (но можно и в папочку)
3 Вставляете путь к претрейну в форке (стандартный форк или мангио форк)
4 Вы готовы жечь!

Quick guide how to Change Hubert:

Disclaimer: inference and training wont work on different models and berts. Jp bert will work with model that trained on it and wont work on default one.
Transfer to the same folder where the standard Hubert is located.
Necessarily! When transferring a new Bert, rename the file to hubert_base.pt

Дисклеймер: Инференция и тренеровка не будет работать на разных моделях и бертах. Jp bert будет работать с моделью, которая обучалась на ней, и не будет работать с моделью стандартной.
Для смены Хуберта:
переносите в туже папку где надится стандартный хуберт.
Обязательно! При переносе нового Берта переименовать файл на hubert_base.pt

Japanese Hubert by rinna
https://huggingface.co/rinna/japanese-hubert-base

[fierce storm]

RVC Disconnected Update

Quite late on announcing this one, aren't you Kit?
I've added ItaIla, SnowieV3, and SnowieV3 x RIN_E3 to the pretrain selector! Be sure to select and download a pretrained model before you start training by using the new "Download Pretrained Model" cell. (If you don't care what pretrain you use, simply download the "original" pretrain, which is default.)
As always, please let me know if there's any issues regarding this update in #1192011222023950368 or #✨│ai-help .
https://colab.research.google.com/drive/1XIPCP9ken63S7M6b5ui1b36Cs17sP-NS

[ruby plinth]

Ilaria RVC Mainline got updated to the Beta 2!

Try it out now!

Github: https://github.com/TheStingerX/Ilaria-RVC-Mainline/releases/tag/beta-2.0

Colab: https://colab.research.google.com/drive/1mHKTGH5e3SAyDSBss1KtiYRbDdQzwSMs

[pure bloom]

🎵 UVR5 NO UI for Google Colab 🎵

This colab includes:

• All VR Arch Models
• All MDX-NET Models
• Demucs v4 Models
• MDX23C Models
• BS-Roformer ViperX Models
• VIP Models
• Batch Separation

Now you can use the best models without having to wait in a queue or use up your CPU if you are low on resources or you are on mobile, all this for free!

Try it out now!

** Colab:** https://colab.research.google.com/github/Eddycrack864/UVR5-NO-UI/blob/main/UVR5_NO_UI.ipynb

How To Use: https://github.com/Eddycrack864/UVR5-NO-UI#how-to-use

Made with by @pure bloom for **AI HUB **

[native granite]

# HUGGING FACE IS HAVING ISSUES ~~
~~ They are currently having problems, sometimes going up, 5 mins after going down again, etc.
~~You could not be able to download the AI Models from there. ~~
This also means YOU CANT USE WOKADA VOICE CHANGER EITHER LOCALLY OR ONLINE
~~Check their status at: https://status.huggingface.co/~~

HUGGING FACE IS BACK UP

[hollow zephyr]

Bot notice 🛠️

@sweet sun is currently experiencing some issues. The /submit command will be unavailable until the incident is resolved.

<@&1159434617451978802> To avoid data losses, please also avoid using the following commands:
/approve
/reject
/cancel

Thank you for your patience and understanding

[hollow zephyr]

Update 🔔

More complications have occured. Unfortunately, @sweet sun will be fully down for an unknown amount of time. This means submissions for the <@&1159332519167144039> role via the /submit command are temporarily suspended.
Apologies for the inconvenience

[hollow zephyr]

Update ✅

A fix has been implemented and @sweet sun is now back online. All of its commands should be restored. If it breaks again, feel free to ping me.
To avoid cluttering this channel, the bot's status will be updated here if I need to do maintenance on it in the next days.

QCBot's current status: 🟢

[nimble tapir]

Huggingface is currently down.

https://status.huggingface.co/

[nimble tapir]

Huggingface is back online.

https://status.huggingface.co

[silent trail]

🎉 We made a small update to the /search and /searchuser command, now you will have the help of an autocomplete to make your searches easier!

[visual osprey]

🚀 Introducing Applio NoUI

In response to Google's recent blocks, we have created a perfect alternative to use Applio without the UI complexities.

Training and inference available!

• https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio_NoUI.ipynb

[visual osprey]

Experimentally launching Custom Hop Lenght for Rmvpe available on Google Colab! ok, ~~totally ~~partially broken. Reverting for now

[visual osprey]

Introducing TITAN: The Versatile, Robust, and High-Quality Pretrained Model for RVC Training

Today I am pleased to announce TITAN, a state-of-the-art pre-trained model designed specifically for RVC training. TITAN is a robust solution for RVC, providing high quality results with minimal training effort. This is just the beginning of a collection of pretraineds that I am sure will bring a lot to our community.

Model Details

Titan-Medium
TITAN-Medium is the first model in the series, trained using a RTX 3060 TI on Applio v3.1.1 with a batch size of 8 over the course of 3 weeks. The model underwent 1,010,588 steps and 467 epochs, with a sampling rate of 40k and 32k (still in training by @arctic comet). The fine-tuning process involved using an RVC v2 pretrained model with pitch guidance, leveraging an 11.15-hour dataset sourced from Expresso. You can find the dataset used for fine-tuning in the TITAN-Medium repository.

Titan-Large
Details for Titan-Large are forthcoming. Stay tuned for updates!

Collaborators

I would like to extend my gratitude to our collaborators who have contributed to the development and refinement of TITAN:

• @west reef
• @past marlin
• @arctic comet

Beta Testers

Also thanks our beta testers who provided valuable feedback during the testing phase of TITAN:

• @past marlin
• @mortal aurora
• @sweet sun
• @ember quail
• @sinful relic
• @civic veldt

Available as both model and pretrained

More info, samples and ckpts here! ➡️ https://huggingface.co/blaise-tk/TITAN/

*(works on applio, rvc, mangio...) *

[visual osprey]

<@&1165417631969448057> 🔔

[visual osprey]

~~Gradio is presenting some difficulties internally, this could be noticed in actions such as having the public link of a UI, so it could affect the use of tools such as Google Colab.

⏬ Follow the status here
https://status.gradio.app/~~

back again, more or less

[native granite]

Added Ngrok Tunnel option to FaceFusion UI Colab, now you can use it when Gradio is down!

[native granite]

#1228715105127235646 IS IN MAINTENANCE RN (EXCEPT THE HF SPACE, LOCAL & NO UI COLAB)
#1228715105127235646
Added:

• See as 3D Model Feature
• Ngrok Tunnel for the WEB UI Colab
• No UI & UI Kaggle Notebooks

Enjoy!

[native granite]

GRADIO API COULD STILL HAVE SOME PROBLEMS

There could be problems into having the Public URL Link in the google colabs, Check the status in https://status.gradio.app

[visual osprey]

Exciting update on TITAN! 🎉

Introducing the 32k version of Titan-Medium, courtesy of @arctic comet training efforts. ❤️
All the familiar features, now in the version you've been waiting for.

➡️ Check it out here: https://huggingface.co/blaise-tk/TITAN

[native granite]

#1221468455723733094 IS CURRENTLY IN MAINTENANCE, IF YOU GET ERRORS, PLEASE WAIT!

#1221468455723733094 is fixed & updated!

[fierce storm]

RVC Disconnected Update

As per (exhaustive) request, I've added the TITAN-Medium pretrain to Disconnected's pretrain selector. I also added a catch such that 32k models wishing to use Ov2Super will now download and use the Poopmaster fixed version.
This will probably be the last unique pretrain I add to the selector; the only changes from hereon out should hopefully just be updates like Ov2 2.1 and TITAN-Large.

I also added an experimental toggle for something to Training, courtesy of Sonphantrung. Let me know in #🔊│ai-development if things go awry with it.

https://colab.research.google.com/drive/1XIPCP9ken63S7M6b5ui1b36Cs17sP-NS

[pure bloom]

New feature added to UVR5 NO UI 🚀

Now you can also separate an audio from a YouTube video/short using the link.

Try it now!

Colab: https://colab.research.google.com/github/Eddycrack864/UVR5-NO-UI/blob/main/UVR5_NO_UI.ipynb

How To Use: https://github.com/Eddycrack864/UVR5-NO-UI#how-to-use

Made with by @pure bloom for AI HUB

[native granite]

~~ https://discord.com/channels/1159260121998827560/1218902018673086524 is currently in maintenance!~~

https://discord.com/channels/1159260121998827560/1218902018673086524 is fixed!

[silent trail]

🚀 Exciting News! Upgrade to the latest and greatest version of Applio, 3.2.0, now available for download!

Choose your preferred option:

• 🚀 Release: Download Now
• 📦 Compiled (Just double click it): .exe / .zip

Applio Changelog Highlights:

• 🌟 Embedder Model: Choose between hubert or contentvec for optimal performance.
• 📥 Pretrain Downloader: Easily download the most popular pretrains!
• 📈 Sync Graph: Now you can link Tensorboard graphs to your actual training model with this new option!
• ⏩ TTS Speed: Adjust the speed of the text-to-speech to your liking.
• 🔊 Upscale: Enhance the quality of your audio!
• 🔄 Port Changer: Applio now automatically switches to another port if 6969 is unavailable.
• 🛠️ Pretrains: Improved compatibility for creating pretrains, now compatible with other RVC repos like RVC Mainline using older torch versions.

We skipped some versions to bring you this big update. This version is highly recommended for training pretrains. What features are you hoping to see in our upcoming updates? 🔍 Stay tuned for more exciting improvements!

[visual osprey]

Exciting update on TITAN! 🎉

Introducing the 48k version of Titan-Medium, courtesy of @west reef training efforts. ❤️
All the familiar features, now in the version you've been waiting for!

➡️ Check it out here: https://huggingface.co/blaise-tk/TITAN

[stiff elk]

The Gradio Share Links are currently down. If you are using Applio Colab, please enable the "share_tunnel" option for now.

[native granite]

ILARIA RVC MANGIO IS DISCONTINUED

The ilaria rvc mangio which was the one in our docs, won’t get updated anymore and it doesn’t work. It got removed from our docs.
The whole ilaria rvc project is not dead, ilaria rvc was a fork on Mangio RVC before, but as mangio rvc was not updated anymore, ilaria decided to instead fork Mainline rvc, making Ilaria RVC Mainline which is a better version with training and other options and similar to the original ilaria rvc.
So you can either:

• Try ilaria rvc mainline #📰│dev-updates message
• Check the other colabs in our docs https://docs.aihub.wtf

[visual osprey]

Sharing an important message from the RVC team:
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/2109

Unfortunately, I have to tell you the truth that the initial RVC developer, whose name is RVC-Boss, basically refuses ANYONE to do ANY changes (including PR) on HIS project, which is, in his mind, "dirtying" his code. That is the main reason of why this project progresses so slowly. I had a chat with him and decided to REVERT the commit of the main repo to what he like it to be and put the newest code to my repo as below.

https://github.com/fumiama/Retrieval-based-Voice-Conversion-WebUI

My repo will continue to pull the algorithm update from the main repo (this repo), and, is open for everyone to contribute new features. I hope that with this effort, RVC will be better than before by the help from you alls.

[pure bloom]

New UVR5 NO UI Update

Changelog:

• Model UVR-BVE-4B_SN-44100-1 was fixed and now works correctly
• New models added: BS-Roformer-Viperx-1053 and Mel-Roformer-Viperx-1143
• Colab setup time reduced to just 2 minutes!
• Separation time of the BS-Roformer and Mel-Band RoFormer models reduced by half.
• Improved separation quality of BS-RoFormer and Mel-Band RoFormer models

Try it now!

Colab: https://colab.research.google.com/github/Eddycrack864/UVR5-NO-UI/blob/main/UVR5_NO_UI.ipynb

How To Use: https://github.com/Eddycrack864/UVR5-NO-UI#how-to-use

Made with by @pure bloom for AI HUB

[native granite]

https://discord.com/channels/1159260121998827560/1228715105127235646 Now has:

• Mac
• Linux
• Windows

Local support

[native granite]

https://discord.com/channels/1159260121998827560/1218902018673086524
Update:

• Reduced the minimum max new tokens to 50 instead of 250
• Fixed a bug where you couldn't put repetition penalty on 2 for the web ui

[ruby plinth]

ilaria huggingface in beta, now with an A100 80GB, this means ultra fast inference

try it here https://huggingface.co/spaces/TheStinger/Ilaria_RVC

give me feedback on #✨│ai-help tagging me if something is wrong - NO TUTORIALS YET BUT TIS PRETTY EASY

message will be deleted again once finished to make the full release announcement

[ruby plinth]

🎉 ILARIA RVC ZERO IS NOW AVAILABLE! 🚀

Check out these amazing features:

• ✨ Ultra Speed with GPU A100 80GB, making your inferences lightning fast!
• ⏳ Queue System to manage multiple inferences simultaneously.
• 🎨 Enhanced Inference Page with a more intuitive and visually appealing design.
• ⚙️ Advanced Settings offering greater customization.
• ⏩ Faster Conversion for all your needs.
• 📥 Rapid Model Download & Upload, super easy to use.
• 🔊 UVR5 (still in testing) for separating or enhancing audio.
• ℹ️ Model Information to learn more about your selected model.
• 🕒 Training Time Calculator for an approximate training duration estimate.
• 🔀 Model Fusion to effortlessly blend models and create new voices!

And the best part? It's F R E E! 🌟

https://huggingface.co/spaces/TheStinger/Ilaria_RVC

[native granite]

# FACEFUSION ONLINE PORTS BUG
Currently the online ports are having a bug, gonna fix it soon!
Fixed it! But kaggles are gone as i got banned off kaggle.

[ruby plinth]

Ilaria RVC Zero is hard for you?
Different UI so its confusing for you?

Thanks to @native granite we have a guide: https://docs.google.com/document/d/1YbXcLFPaGjhOdG5NFkK3QrucCEpHZBwFUxkeMO8aB18/edit

Smooching and kisses from your local italian queen <3

[visual osprey]

During these months we have talked a lot about new versions of RVC. In response to the abandonment of the constant development of RVC, some initiatives were born, such as the creation of new pretraineds, seeking to bring us closer to this technological advance that we all want.

With Applio we want to mark a point where we can start a new path on the fundamentals of the RVC architecture that until now was little different from that of VITS.

This involves the de-alignment of those standards that RVC set some time ago to start introducing some more significant changes, some of these have been introduced gradually (or will be) and others will be somewhat more drastic.

Among future changes involves updating the architecture of the RVC technology, which is currently based on HIFIGAN and is what we know as V2.

The change of the version number will be necessary, simply because the models will have another architecture, the main idea is to call it V3, but if in the future RVC will also release another V3 it will be adapted.

We start a process of polishing the architecture version and preparing the pretraineds, the latter in particular, is a task that costs money so any kind of donations are always welcome, as it requires computational power. Also, anyone with a lot of computational power can write to me.

Soon we will be updating and giving more news in order to see if finally an update of the architecture that we use so much, is worth it.

[silent trail]

🚀 Exciting News! Upgrade to the latest and greatest version of Applio, 3.2.1, now available for download!

Choose your preferred option:

• 🚀 Release: Download Now
• 📦 Compiled (Just double click it): .exe / .zip

Applio Changelog Highlights:

• 🌟 Custom Embedders: Choose the embedder model that best suits your needs.
• 📥 Cache Dataset in GPU: Cache the dataset in the GPU to improve performance.
• 📈 Custom CPU Cores: Select the number of CPU cores to use for processing and extraction.
• 🎵 Pitch Curve: Extract the pitch curve for the audio file.
• 🔊 Optimization: Enhance processing performance and improve the speed of updates.
• 🔄 Python 3.11 Compatibility: Applio is now compatible with Python 3.11.
• 🐛 Minimal Error Fixes: Addressed minor issues for a smoother experience.

What features are you hoping to see in our upcoming updates? 🔍 Stay tuned for more exciting improvements!

[ruby plinth]

🌟 Welcome to UVR5 UI! 🌟

🎶 Are you tired of waiting in long queues to separate vocal and instrumental tracks? UVR5 UI is the perfect solution for you! ⚡

✨ Why Choose UVR5 UI? ✨

🎛️ User-Friendly: Separating vocal and instrumental tracks has never been this easy!

🎨 Multiple Separation Models: Choose from a wide range of models to find the perfect fit for your needs.

📁 Versatile Formats: We support various file formats to ensure maximum flexibility.

🚀 Blazing Fast: Powered by our A100 80GB, experience lightning-fast separations!

💸 Completely Free: Enjoy all these features at no cost!

🔗 Links

🤗 Huggingface - Use an A100 80GB (Faster than Colab; No Hours limit)
Colab - Use a T4 (4 hours daily limit per account)

*UVR5 UI is made by @pure bloom completely, the space is hosted by Ilaria. 💖 *

[ruby plinth]

⚡ NEW IMPORTANT FEATURE IN UVR5 UI ⚡

What is it?

You asked, @pure bloom delivered and Ilaria hosted! 💖
📥 Download Audio: From now you're able to download audio directly from the UI without the need of you having to do yourself!
A complete list of avaliable services (YouTube included!) is avaliable here.

🔗 Links

🤗 Huggingface - Use an A100 80GB (Faster than Colab; No Hours limit)
Colab - Use a T4 (4 hours daily limit per account)

🤗 TRY UVR5 UI NOW FOR FREE AND UNLIMITED

[native granite]

The RVC v2 Pretrains Guide just got updated!

• Added Table Of Contents to make it easier to read
• Added info about the original pretrain dataset (being the english-only VCTK one)
• Added RVC v2 Disconnected Google Colab section for downloading pretrains
• Added Applio's automatic way to install pretrains via the Download tab (before only the custom pretrain checkbox was explained)
• Added Applio Google Colab UI
• Added Applio Google Colab NO UI
• Added Applio Kaggle
• Replaced KLMv7s old deleted pretrain with KLM 4.1
• Fixing typos/misspelling
  And thanks to @west reef :
• Added DMR v1 & Nanashi v1
• Added Best GPUs for training a pretrain (tierlist + recommended one by him)
• Batch size info
• General Recommendations

Check it out here: https://docs.google.com/document/d/1j9J8A8Oop9bMOHmCs3jDXzPujuD6TQ0Q396rJ0MyuIc

[native granite]

EASYGUI GOOGLE COLAB IS BROKEN & DELETED!

The easygui google colab was broken since some time, waiting for @errant sorrel to push the fix, it stayed stay like this for both ui and no ui as the problem is in the installation.
But recently the Google Colab got deleted. RIP

You can use other cloud ways, like:

• Ilaria RVC Zero (INFERENCE ONLY) (ZeroGPU Hugging Face Space, A100 faster than Google Colab T4)
• Other Google Colabs in our docs/

[native granite]

FACEFUSION ONLINE PORTS BRANCH UPDATE!

Update

Now in https://discord.com/channels/1159260121998827560/1163571683848900629 the Google Colabs and the Hugging Face Space, you can change the FaceFusion Github Branch!

Branches

• master: the latest stable release!
• next: the latest beta release with in-development features!

[silent trail]

🚀 Exciting News! Upgrade to the latest and greatest version of Applio, 3.2.2, now available for download!

Choose your preferred option:

• 🚀 Release: Download Now
• 📦 Compiled (Just double click it): .exe / .zip

Applio Changelog Highlights:

• 💨 Enabled GPU support for RMVPE and CREPE extraction processes
• 🗃️ Streamlined and optimized the extraction workflow
• 🌐 Implemented global code enhancements
• 💻 Added compatibility for Linux plugins
• 🎯 Introduced a toggle to force FP32/FP16 precision
• 📶 Added a progress bar between epochs to track the status of your steps

What features are you hoping to see in our upcoming updates? 🔍 Stay tuned for more exciting improvements!

[native granite]

~~ GOOGLE COLABS UI HAVE PROBLEMS, GRADIO IS CURRENTLY DOWN! ~~

~~ Gradio Share API which is used to share the public url link in most google colab uis, is currently down, you will need to wait! ~~
~~ Check the status in: https://status.gradio.app/ ~~
All solved!

[native granite]

~~GOOGLE COLABS UI HAVE PROBLEMS, GRADIO IS CURRENTLY DOWN! ~~

Gradio Share Links, the public url link in most google colab uis, is currently down, you will need to wait!
~~Check the status at: https://status.gradio.app/ ~~
All solved!

[native granite]

ILARIA RVC MAINLINE GOOGLE COLAB IS BROKEN!

The current ilaria rvc mainline google colab that replaced the old ilaria rvc mangio google colab, is broken currently!

You can use other cloud ways, like:

• Ilaria RVC Zero (INFERENCE ONLY) (ZeroGPU Hugging Face Space, A100 faster than Google Colab T4)
• Other Google Colabs in our docs

[ivory goblet]

# RVC AI Cover Maker WebUI v1.0.0 Release!
Today I'm announcing the launch of my “new” gradio program.
What does it do?

• It does the complete cover generation process with the best vocal/backing vocals/reverb separation models
• It runs both in the Colab and on your own PC (Windows or Linux) !

Note: To run on your PC, I recommend an RTX card, AMD doesn't support it unfortunately.
Note 2: I only tested the default settings, which I consider to be the best for most cases
Any errors you find in the program, feel free to ping me!

• Colab Link
• PC Version link

[native granite]

FaceFusion New Online Port: Lightning AI!

You can now use FaceFusion (UI) on Cloud, Lightning.ai offers for free 15 credits monthly,
Which can be in terms of GPU & Vram from the most to the least hours (& power):

• 22 hours monthly of T4 16gb
• 21 hours monthly of L4 24gb
• 8 hours monthly of A10G 24gb
• 6 hours monthly of L40 48gb

(This has been calculated by the price per hour of the gpus, and meant if u use only one gpu for all 15 credits, ofc u can swap between GPUs)

The default (and best one personally) is the L4 gpu for being faster than T4 with double vram, for the price of one hour less than T4

It has persisten storage meaning you won't need to reinstall everytime (unless you change from cpu to cpu or viceversa, or change branch)
Also, there's NO risk of ban or session stopping for disallowed code as Lightning.ai has nothing against UIs or Deepfakes

Check it out https://lightning.ai/nick088/studios/facefusion-ui

[native granite]

EASYGUI GOOGLE COLAB IS BACK!

@errant sorrel came back and uploaded a fixed/updated version of the google colab!

Check it out!

[native granite]

RVC CAN RUN LOCALLY ON ANDROID NOW!

You can sucessfully inference (use models) on your android phone!

It runs on CPU which means its slow and the performance depends on your phone, but it's able to do it even **offline **after setup! (except the TTS)

As of right now, you can use Applio & EasyGUI

Some examples on Applio:

• A Galaxy A9 2018 took 146.67 seconds to convert a 10 seconds audio!
• (shown in the screenshot) an Honor 90 Lite took 69,08 seconds for converting an 8 seconds audio

EasyGUI Example:

• It took around 59-61 seconds for converting 8 seconds audio on an Honor 90 Lite

The fastest one for inference seems to be easyGUI for sligthly 8-10 seconds less.

Technically it would be able to train on CPU too, at a very slow speed, but there was no powerful enough phones that tested it sucessfully training a model right now.

Check it out!

[native granite]

FaceFusion Lightning.ai YouTube Tutorial!

Is the lightning.ai port too hard for you to setup & use?

Check out the YT video for an easier explaination!

[native granite]

~~ # GOOGLE COLABS UI HAVE PROBLEMS, GRADIO IS CURRENTLY DOWN! ~~
~~ Gradio which is used to share the public url in most google colab uis, is currently down, ~~
~~ You can: ~~
~~ - Wait ~~
~~ - Use a NO UI Google Colab ~~
~~ - Use another Tunnel if the google colab gives you the option ~~
~~ Check the status at: https://status.gradio.app/ ~~
All solved!

[native granite]

THE AI HUB DOCS ARE TEMPORARELY DOWN

Our https://docs.aihub.wtf is TEMPORARELY down,

FOR THE MEANWHILE USE: https://docs.ai-hub.wtf/

[native granite]

Minecraft AI skin generator YT Tut!

As most people do things on cloud, and a new user could be prefer a video rather than a written guide, I made a youtube tutorial!

Check it out!

[fiery fable]

Automaze Command Updates

Hey there, I hope you're doing well! These are the latest changes to the @boreal prairie commands.

• /faq: New command including common questions about RVC
• /close: Replaces the old -close
• /help: Major command overhaul including comprehensive explanation on slash (/), context and prefix (-) commands. Replaces the old -help
• /cloud: Combines -colab, -kaggle, -hf and -lighning prefix commands in one place. You can still use the prefix ones if you’d like.
• /guides: Removed outdated links and fixed errors when selecting a language.

That's it, until next time!

[native granite]

Our Termux guides aren't currently not working.

Our Termux guides (RVCs & FaceFusion) will temporarely not work since they updated a newer ubuntu which can't use python 3.10.

[native granite]

Termux Guides are back, with a new one too!

FaceFusion, RVC (EasyGUI & Applio) are back!!

With a new one!
Check out how to run Stable Diffusion locally on your CPU: here

[native granite]

Lightning.ai is currently having issues

If you open any lightning.ai notebooks link or by people's profile, they redirect you to the search featured notebooks
And if you search for any notebooks on the featured page, it just goes blank

I already contacted them about the issue, meanwhile we can either use other ports or wait.
Fixed

[pure bloom]

A big update for UVR5 UI is here! 🚀

What's new? ✨

• New models added: 🎵
  • MelBand Roformer Kim | Inst V1 by Unwa
  • MelBand Roformer Kim | Inst V2 by Unwa
  • MelBand Roformer Kim | InstVoc Duality V1 by Unwa
  • MelBand Roformer Kim | InstVoc Duality V2 by Unwa
  • UVR-MDX-NET-Inst_HQ_5
    Check all available models here
• Inference speed improved by 80% ⚡
  This is possible thanks to the use of Pytorch Autocast and the new code improvements, check the comparison table here. Tested on Tesla T4 btw.
• New supported audio formats: 🎶
  • ogg
  • opus
  • m4a
  • aiff
  • ac3
• Progress bar added to interface 📶
• Stop renaming input audio files (The name of the original audio is now preserved) 📝
• Stop using CLI (use python function instead) 💡
• Code Improvements 🖥️
• New features have been added to the interface (Batch Size, Amplification Threshold, Normalization Threshold and more!) 🛠️
• 14 available languages 🔤
  If you want to help me to translate UVR5 UI to other languages contact me 🙏

Try it now! 🚀

Cloud: ☁️

• Colab
• Kaggle
• Lightning.ai

Local: 🖥️

• Windows
• Linux

Any questions or problems feel free to ping me. Currently working on HF update.

[shadow wigeon]

Applio App — First public alpha is now available!

We’re excited to share some great news: our new desktop application is ready for early testing on Windows! After a lot of work behind the scenes, we’re about to take this next step with you.

If you’d like to be one of the first to try it out, you can join the waitlist here:

https://applio.org/products/app

In the next few days, the public alpha will be launched for selected users on the waiting list. If you are one of them, be sure to join our Discord server to receive access to the application. We are waiting for you!

Supporters have priority access.

To show our appreciation to Supporter subscribers (both new and existing), we are offering them priority access to the testing phase. You can try Applio App now!
If you are interested in supporting our work and enjoy some additional benefits, you can subscribe via Ko-fi here:

https://ko-fi.com/iahispano/tiers

[pure bloom]

A new update for UVR5 UI is here! 🚀

What's new? ✨

• A lot of amazing new models were added ! 🎵
  • Mel Band Roformer Vocal by Kimberley Jensen
  • MelBand Roformer Kim | FT by unwa
  • MelBand Roformer Kim | Inst V1 (E) by Unwa
  • MelBand Roformer | De-Reverb by anvuew
  • MelBand Roformer | De-Reverb Less Aggressive by anvuew
  • MelBand Roformer | De-Reverb-Echo by Sucial
  • MelBand Roformer | De-Reverb-Echo V2 by Sucial
  • MelBand Roformer Kim | SYHFT by SYH99999
  • MelBand Roformer Kim | SYHFT V2 by SYH99999
  • MelBand Roformer Kim | SYHFT V2.5 by SYH99999
  • MelBand Roformer Kim | SYHFT V3 by SYH99999
  • MelBand Roformer Kim | Big SYHFT V1 by SYH99999
  • MelBand Roformer Kim | Big Beta 4 FT by unwa
  • MelBand Roformer Kim | Big Beta 5e FT by unwa
  • BS Roformer | Chorus Male-Female by Sucial
  • MelBand Roformer | Aspiration by Sucial
  • MelBand Roformer | Aspiration Less Aggressive by Sucial
  • MelBand Roformer | Bleed Suppressor V1 by unwa & 97chris
  • MDX23C De-Reverb by aufr33 & jarredou
  • MDX23C DrumSep by aufr33 & jarredou (Only 2 stems atm)
  • UVR-De-Reverb by aufr33 & jarredou
    Check all UVR5 UI available models here
• Added better description for stems. e.g. stem_1/stem_2 to Vocals/Other 💡
• UVR5 UI now automatically opens in the default browser
• Added a notebook for Docker-based paid cloud services (thanks to iroak) ☁️
• Code Improvements 🖥️

Try it now! 🚀

Cloud: ☁️

• Colab
• Kaggle
• Lightning.ai
• Docker-based notebook

Local: 🖥️

• Windows
• Linux

Any questions or problems feel free to ping me. Currently working on HF update.

[native granite]

RVC DISCONNECTED IS HAVING ISSUES, BUT THERE'S A FIX FOR NOW!

The RVC Disconnect Colab is having an issue right now:
AttributeError: 'FigureCanvasAgg' object has no attribute 'tostring_rgb

Untill @fierce storm fixes it, there's 3 temporary fixes:

• Use @celest pagoda's modified version
  -# (which uses the same fix below, just already modified so you don't have to manually)
• Add pip install matplotlib==3.7.0 in the "Install Dependencies" cell before the "Packages up to date." print statement (shown in the picture)
• Use alternatives which can be found in our Docs

[fierce storm]

Update: The above hotfix has been pushed to the main RVC Disconnected notebook. Further remediation of this error will be looked into at a later time.

[pure bloom]

UVR5 UI on Hugging Face has been updated! 🚀

New features include:

• 25 incredible new models ✨
• Inference speed increased by 80% ⚡
• New supported audio formats (ogg, opus, m4a, aiff, ac3) 🎵
• Progress bar added to the GUI 📶
• Stop renaming input audio files (The name of the original audio is now preserved) 📝
• New features have been added to the interface (Batch Size, Amplification Threshold, Normalization Threshold and more!) 🛠️
• Added Leaderboard sorted by stems and SDR ⬆️

With this update you can separate up to 20 minutes of audio in just 1 minute, with no queue!

Try it now with A100 GPU! 🚀

Hugging Face Link 🤗

Thanks to @ruby plinth for hosting this space

[ionic glade]

Docs are down for maintenance, will be up again shortly
Docs are back up again, there could be some issues with SSL (you get a https warning)
Docs are fully up and operational again

[native granite]

GOOGLE COLABS ARE HAVING ISSUES!

Warning: This message will get updated frequently
Google updated Colab's Python version from 3.10.12 to 3.11.11, colabs are having issues.

You can:

• Wait
• Use other cloud services like Kaggle & Lightning.ai,guides in our docs
• Use Weights
• Use fixed colabs:
  • Mainline RVC
  • MST (Inference)
  • Applio UI & Applio NO UI
  • Hina's Mod OG W-Okada's
  • Hina's Mod AICoverGen WebUI, Eddy's AICoverGen NoWebUI
  • W-Okada's Deiteris Fork
  • FaceFusion UI & FaceFusion NO UI (disconnect issues on Free Tier)
  • UVR5 NO UI & UVR5 UI
  • AI Cover Maker UI

[ruby plinth]

Hello everyone, recently you had problems with duplicating Ilaria RVC Zero and the space of Ilaria Upscaler was broken.

Thanks for the wait, its now fixed also thanks to a random stranger (reddit gold medal moment) on HuggingFace.

Ilaria RVC Zero: https://huggingface.co/spaces/TheStinger/Ilaria_RVC
Ilaria Upscaler: https://huggingface.co/spaces/TheStinger/Ilaria_Upscaler

Stay cool and fresh!

[ruby plinth]

WE ARE ON PINOKIO NOW!

List of avaliable and official ports in the Community Scripts about to be verified:

• Ilaria RVC Zero
• Ilaria Upscaler Zero
• UVR5-UI (by @pure bloom)

You can now enjoy your favourite AI apps locally, without limitations (like HuggingFace or Colab ones) on your PC!
An Nvidia card is strongly reccomended.

[fierce storm]

RVC Disconnected Update

I have pushed the Torch installation fix to the main version of the notebook. This should hopefully ensure that model loading will work.
With this, the RVC Disconnected notebook is officially end of life. It will not be updated further.
Please use a different method of training RVC models, such as Mainline RVC or Applio No-UI.
Thanks for all the love,
-KitLemonfoot

[native granite]

RVC AI Cover Maker UI Update

RVC AI Cover Maker UI was originally broken and the original dev ( @ivory goblet ) couldn't keep maintaining it #✨│ai-help message,
But @pure bloom is now going to maintain it and the ownership changed, it's fixed now!

• Github (Local)
• Cloud Colab
• Cloud Kaggle

[native granite]

GOOGLE COLABS ARE HAVING ISSUES!

Warning: This message will get updated frequently
Google updated Colab's Python version from 3.10.12 to 3.11.11, and at first people found a bypass via using UV environments, but now colabs are having issues.

You can:

• Wait
• Use other cloud services like Kaggle & Lightning.ai,guides in our docs
• Use Weights/
• Use it Locally (runs on your pc) if you got a good one that meets the requirements
• Use fixed/non affected colabs:
  • Eddy's UVR UI & NO UI
  • MST (Inference)
  • Applio UI & Applio NO UI ONLY WITH THE WORKAROUND
  • Wokada's Deiteris' Fork WARNING: This colab works, but even with the recently added encryption, it gets free tier users flagged for using a web ui
  • AICoverGen No WebUI: English & Spanish
  • AI Cover Maker UI
  • FaceFusion UI & FaceFusion NO UI (disconnect issues on Free Tier)

[pure bloom]

A new update for UVR5 UI is here! 🚀

What's new? ✨

• 33 new models have been added ! 🎵
  • MelBand Roformer | Karaoke by Gabox
  • MelBand Roformer Kim | FT 2 Bleedless by unwa
  • MelBand Roformer | Vocals by becruily
  • MelBand Roformer | Instrumental by becruily
  • BS Roformer | Vocals by Gabox
  • MelBand Roformer | Vocals by Gabox
  • MelBand Roformer | Instrumental by Gabox
  • MelBand Roformer | De-Reverb Mono by anvuew
  • MelBand Roformer | Big Beta 6 by unwa
  • And more!!
    Check all UVR5 UI available models here
• UVR5 UI now has Discord Rich Presence! 🎮
• More languages ​​have been added 🌐
• UVR5 UI now has docs. Check it out here 📝
• Code improvements 🖥️

Try it now! 🚀

Cloud: ☁️

• Colab
• Hugging Face thanks to @ruby plinth
• Kaggle
• Lightning.ai
• Docker-based notebook

Local: 🖥️

• Windows
• Linux
• Pinokio

UVR5 NO UI has also been updated on Colab / Kaggle / Lightning.ai

[native granite]

~~ # CLOUD UIs HAVE PROBLEMS, GRADIO IS HAVING ISSUES! ~~
~~ Gradio which is used to share the public url in most Google Colab/Kaggle/Lightening uis, is having issues with files around larger than 1MB or images larger than 600x600, ~~
~~ You can: ~~
~~ - Wait ~~
~~ - Use NO UI Cloud Options ~~
~~ - Use another Tunnel if the google colab gives you the option ~~
~~ - Check the issue https://github.com/gradio-app/gradio/issues/10738 ~~

Gradio is fixed!

[celest pagoda]

MSST Colab Inference (Custom tweak by me) ✨

This is a tweaked version of jarredou's colab notebook that clones and runs ZFTurbo's MSST repository.

What's New?

• New separation models:
  • unwa ft2 normal & bleedless
  • unwa big beta 6 (less fullness than beta 5 and similar to ft2)
  • gabox vocal Fv3 and Fv4 (the latter has more fullness)
  • gabox inst v7 (the last and best known)
  • gabox karaoke (experimental finetune, more fullness but still not as effective as the original aufr33 karaoke)
  • gabox debleed (personally still not as effective as unwa's bleed suppressor)
  • Amane's BS roformer (4 stems, but I still recommend SCNet XL)
  • anvuew mono dereverb (the normal one is for stereo reverb)
• The installation is done at the colab's default main environment with Python 3.11. No fixes other than ml_collections and loralib package applied, and it shouldn't suffer the issues like other RVC colabs are having.

Note

• If there are models you want but not included in the colab, you can leave me suggestion on it or edit the code in Separation cell. Any other feedbacks and issue report are welcome.

Try it now!

[native granite]

AICoverGen will be REPLACED!

AICoverGen is a known tool that automates the process of creating AI Covers, but the code is basically abandoned and no update has been done to fix the issues with the colabs and main code. Initially, @midnight dagger fixed the web ui colab with Hina Mod AI CoverGen, and Eddy fixed the NO UI ones for Spanish & English, but both of them that it's filled with bugs and not worth maintaining anymore as it uses older models.

RVC AI Cover Maker UI, originally made by @ivory goblet , now maintained by @pure bloom , will be a good replacement for it, including better separation models and improved code.

• English UI Colab
• UI Kaggle
• Guide for Kaggle & Colab

[pure bloom]

UVR5 UI now has a precompiled version 🚀

You just need to extract the .zip file and you already have all the features of UVR5 UI !

Try it now!
Windows
Linux
-# To update UVR5 UI you still need Git.

[native granite]

RVC V3 is here.

Embrace TTS + STS running on CPU

We're happy to announce that RVC v3, the latest, greatest, and last, updated is OUT NOW.
We thought this was going to be just a dream, but.. RVC is never as good as it is TODAY

Wanted TTS with Emotions? Because it's the perfect time for that, as it can run even on your laptop CPU

This project took alot of efforts from our best engineers, we had to assemble the Applio team, the AI Hub team along with Weights , to be able to forge such a majestic project

Check it out NOW!

-# this was an april fools joke btw

[ionic glade]

<@&1159344375290138665> Make sure to also offer RVCv3 model creation. If you want to produce models with even higher quality that most modern phones can run with medium speed, RVCv3 is a must!! Make sure to check out the introduction video above!!

[ionic glade]

Rest in Electronics @merry ridge,
you were a good Bot

[native granite]

REMEMBER TO STILL CHECK THIS ^^^

[pure bloom]

A new update for UVR5 UI is here! 🚀

What's new? ✨

• 10 new models have been added !
  • MelBand Roformer | Karaoke by becruily
  • MelBand Roformer Kim | FT 3 by unwa
  • MelBand Roformer Kim | Inst V1 Plus by Unwa
  • MelBand Roformer Kim | Inst V1 (E) Plus by Unwa
  • MelBand Roformer | Instrumental Bleedless V3 by Gabox
  • MelBand Roformer | INSTV7N by Gabox
  • MelBand Roformer | INSTV8 by Gabox
  • MelBand Roformer | INSTV8N by Gabox
  • MelBand Roformer | FVX by Gabox
  • MelBand Roformer | Big Beta 6X by unwa
    Check all UVR5 UI available models here
• Added feature to save your current settings on gradio and load it next session 💾
• Added feature to download models and config files using wget or curl
• Added feature to manual select your language 🌐
• Added support for NVIDIA 50 series
• New best models for have been updated on docs
• Installer and Updater script have been improved
• Pydantic bug finally fixed
• Colab/Kaggle notebooks update
• Code improvements

Try it now! 🚀

Cloud: ☁️

• Colab
• Hugging Face
• Kaggle
• Lightning.ai

Local: 🖥️

• Windows
• Linux
• Pinokio
• Precompiled Versions
  -# UVR5 NO UI and Hugging Face have been updated!

[native granite]

~~ # GOOGLE COLABS UI HAVE PROBLEMS, GRADIO IS CURRENTLY DOWN! ~~
~~ Gradio which is used to share the public url in most google colab uis, is currently down, ~~
~~ You can: ~~
~~ - Wait ~~
~~ - Use a NO UI Google Colab ~~
~~ - Use another Tunnel if the google colab gives you the option ~~
~~ Check the status at: https://status.gradio.app/ ~~
All solved!

[native granite]

MAINLINE COLAB/KAGGLE & ORIGINAL W-OKADA COLAB/KAGGLE FINAL UPDATE

@midnight dagger 's:

• Mainline Hina Modified Colab
• Mainline Hina Modified Kaggle
• Original Hina Modified W-Okada Colab
• Original Hina Modified W-Okada Kaggle

were broken for various months, as she's busy, she recently pushed their last update to fix them
Now they are officially end of life (EOL), they won't be supported anymore.

You can either:

• Keep using them till they are broken, even tho I wouldn't suggest it, especially for Original W-Okada since Deiteris Fork is better
• Use better alternatives in the AI HUB Docs such as:
  • Applio Colab
  • Applio Kaggle
  • Wokada Deiteris Fork Colab (PAID TIER ONLY)
  • Wokada Deiteris Fork Kaggle

[native granite]

ILARIA RVC ZERO & (ILARIA's HF SPACE) EDDY's UVR5 UI ARE BROKEN!

The HuggingFace ZeroGPU Spaces Ilaria RVC Zero & UVR5 UI are broken for unknown rare ZeroGPU issues, we can't do anything about this unless hoping they might work in the future restarting and contacting HuggingFace.

You can either:

- It's highly suggested to use BETTER ALTERNATIVES in the AI HUB Docs: https://docs.aihub.gg/

• Alternatives like:
  • Applio Kaggle
  • Appllio Colab
  • AICoverMaker Colab/Kaggle
  • Other Isolation Programs
• Wait till they are fixed

[native granite]

~~ # CLOUD OPTIONS UIs HAVE PROBLEMS, GRADIO IS CURRENTLY DOWN! ~~
~~ Gradio which is used to share the public url in most Cloud Options (google colab, Kaggle, Lightning.AI) UIs, is currently down, ~~
~~ You can:~~
~~ - Wait ~~
~~ - Use a NO UI Cloud Option ~~
~~ - Use another Tunnel if the Cloud Option lets you do that ~~
~~ Check the status at: https://status.gradio.app/ ~~
All solved!

[native granite]

Wokada Deiteris Fork New Cloud Port: Lightning.AI! ⚡

For users with poor hardware, you can now run Wokada Deiteris Fork, Realtime Voice Changer for Calls/Games, on Lightning.AI, a cloud (remote good pc) platform to run Jupyter Notebooks that uses a credits system.

This breaks down to the following free monthly hours, depending on the GPU you choose:

• T4 (16GB VRAM): 75 hours monthly
• L4 (24GB VRAM): 31 hours monthly
• L40S (48GB VRAM): 15 hours monthly

Note: This has been calculated by the price per hour of the gpus, and meant if u use only one gpu for all 15 credits, you can switch between GPUs. Pricing might change in the future.

• Persistent Storage: Your files and models are saved, so you don't need to reinstall everything every session. Unless you need to update the program.
• No Ban Risk: Lightning.AI allows Web User Interfaces in the Free Tier, unlike Google Colab or Kaggle.

Check out the guide: https://docs.aihub.gg/realtime-voice-changer/cloud/deiteris-w-okada-fork-lightning-ai/

-# Note: this isn't a partnership, it's just a port made by me.

[native granite]

<@&1403035771270729758>

We will ping this role to notify users about dev stuff! (such as downtimes, updates, etc)
You can get it Channels & Roles!

[native granite]

Applio New Cloud Port: Lightning.AI! ⚡

For users with poor hardware, you can now run Applio RVC Fork for inferencing & training, on Lightning.AI, a cloud (remote good pc) platform to run Jupyter Notebooks that uses a credits system.

This breaks down to the following free monthly hours, depending on the GPU you choose:

• T4 (16GB VRAM): 75 hours monthly
• L4 (24GB VRAM): 31 hours monthly
• L40S (48GB VRAM): 15 hours monthly

Note: This has been calculated by the price per hour of the gpus, and meant if u use only one gpu for all 15 credits, you can switch between GPUs. Pricing might change in the future.

• Great GPUs: Offers the best GPUs that you can even try out for free, helpful for training!
• Persistent Storage: Your files and models are saved, so you don't need to reinstall everything every session. Unless you need to update the program.
• No Ban Risk: Lightning.AI allows Web User Interfaces in the Free Tier, unlike Google Colab or Kaggle.

Check out the guide: https://docs.aihub.gg/rvc/cloud/applio-lightning-ai/

-# Note: this isn't a partnership, it's just a port made by me.

<@&1403035771270729758>

[native granite]

ILARIA RVC ZERO & (ILARIA's HF SPACE) EDDY's UVR5 UI ARE BACK AND FIXED!

<@&1403035771270729758>

[celest pagoda]

MSST INFERENCE COLAB (my custom tweak) HAS BEEN FIXED!

Changelog:

• Applied fix on the installation cell, adopting the jarredou's colab
• The custom forked MSST repo has been adapted to the newest version of the main repo branch
• New models added: Gabox Fv6 and BS roformer Revive Inst

Should you encounter further issues or have model suggestions to add, feel free to ping or contact me.

[turbid mica]

few minor fixes https://github.com/IAHispano/Applio/releases/tag/3.3.1

[turbid mica]

https://github.com/IAHispano/Applio/releases/tag/3.4.0

[turbid mica]

Chatterbox multilingual has just been released! 23 languages. One model. MIT Licensed!

The following languages are supported:
Arabic (ar) •Danish (da) • German (de) • Greek (el) • English (en) • Spanish (es) • Finnish (fi) • French (fr) • Hebrew (he) • Hindi (hi) • Italian (it) • Japanese (ja) • Korean (ko) • Malay (ms) • Dutch (nl) • Norwegian (no) • Polish (pl) • Portuguese (pt) • Russian (ru) • Swedish (sv) • Swahili (sw) • Turkish (tr) • Chinese (zh)

The latest code is up on the official github repo: https://github.com/resemble-ai/chatterbox
Audio Samples here: https://www.resemble.ai/introducing-chatterbox-multilingual-open-source-tts-for-23-languages/
Gradio Demo here: https://huggingface.co/spaces/ResembleAI/Chatterbox-Multilingual-TTS
Weights here: https://huggingface.co/ResembleAI/chatterbox/tree/main

[native granite]

📣 Announcing Wokada Tg-Develop's Fork! 📣

A new Open Source fork based on Deiteris' Fork of W-Okada AI Realtime Voice Changer, developed by Tg-Develop, is now available and in the AI Hub Docs.

Note: Keep in mind that this is just a small public personal project fork with just some QOL updates, don't expect too much nor contact the developer

Key Features:

• Audio Effects: The fork includes a variety of audio effects to alter your voice in real-time, such as Bitcrushing and other 20+ effects for FREE.
• Spin Embedder Type: Models now trained with the Spin Embedder Model Type, are supported too!
• Unified Support for NVIDIA RTX 5000 Series: This version includes a unified update for the NVIDIA RTX 5000 series, instead of the 2 separate links like in the Deiteris' Fork.
• Updated Dependencies & Web User Interface!

Guides:

• Local
• Cloud Kaggle
• Cloud Google Colab

<@&1403035771270729758>

[native granite]

ILARIA RVC ZERO IS BROKEN!

The HuggingFace ZeroGPU Space Ilaria RVC Zero is broken for unknown rare ZeroGPU issues, we can't do anything about this unless it might work in the future restarting and contacting HuggingFace.

You can either:

- It's suggested to use Better Alternatives in the AI HUB Docs: https://docs.aihub.gg/

• Alternatives like:
  • Applio Kaggle
  • Appllio Colab
  • AICoverMaker Colab/Kaggle
• Wait till it's fixed are fixed

<@&1403035771270729758>

[turbid mica]

Now with realtime voice changing https://github.com/IAHispano/Applio/releases/tag/3.5.0

[native granite]

Applio Kaggle and other Cloud Notebooks with Ngrok could have ISSUES!

Cloud Jupyter Notebooks that use the Ngrok Tunnel have issues, because Ngrok lowered the Free Tier requests per minute from 4k to 120.
Applio Kaggle and other notebooks aren't the issue, this isn't a bug that can be fixed, it's just that the Ngrok Tunnel used for Public URLs got a stricter Free Tier now which might affect some programs.
This includes the Applio Kaggle.

You can either:

• Use alternatives in the AI HUB Docs: https://docs.aihub.gg/
• Alternatives like:
  • Applio Lightning.AI
  • Weights.com
  • Appllio Colab
  • AICoverMaker Colab/Kaggle
• Try a workaround for the Ngrok Tunnel:
  1. open the link and click something to get the error
  2. refresh instantly after getting the error
  3. wait atleast 1 minute, then hardrefresh again. Now the Applio Interface should work via Ngrok
• Pay for Ngrok
• Wait till other Tunnels get added (which might be later on via This Pull Request )

<@&1403035771270729758>

EDIT: The issue seems solved!

[native granite]

AI HUB Docs Important March Updates

• Added Wokada TgDevelop LightningAI
• Wokada Tg-Develop Fork Client Audio Processing fix
• Use Vonovox with a Dual PC Setup, using a PC only for Vonovox and another PC will be able to game using the AI Voice.
• fix for "Pipeline not initialized" error in Wokada Tg Develop
• You can now use Applio for Realtime both for Local & Cloud!
• Big Revamp Training Info, Pretrains, "Overtraining", Tensorboard and Vocoders
• Added RVC Zero Cloud
• Blend Models in Applio
• Improved the Cloud Providers in Glossary
• Added RVC Termux Android
• Added ApplioX HuggingFace Space
• Revamped Troubleshooting and HAGS.
• Added Multi Speakers Dataset Applio

The AI Hub Docs are Open Source, if you want to help us, check How To Contribute to the AI Hub Docs and our To-Do List

Note: This message might be edited frequently based on recent important updates for users

<@&1403035771270729758>