help

[silk elm]

hello im new to this i never used it before can someone guide me step by step how to use the ai voice models ?

[idle stratus]

Hello!
Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do?:
  • AI Covers
  • TTS
  • E girl trolling/catfishing
  • Roleplay in VC
  • Roleplay in Games
• what tutorial link are you using (if you used any)
• a screenshot of the program (if you used any)

[silk elm]

nvidia rtx 3050 ti
windows 11
cpu : intel core i7 11800h
i want to do a research im ai / ml engineer i want to train a model and test it on this program but i need first to know how does it fully work a whole picture of it and i might consider buying RTX 5070 ti if it can help me because i own a laptop

[idle stratus]

you got an rtx 3050 ti laptop, it got 4gb vram right?
if so, that's an issue, you can train an RVC (Speech To Speech) voice model, but you won't be able to use batch sizes higher than 4, which might be needed depending on your dataset lenght

[silk elm]

i got imac with 16gb amd radeon x

vram : 16gb

will it help ?

[idle stratus]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• Cloud (remote good pc, easier and faster than ur PC but limited time):
  • Applio UI Kaggle: great for training because of 30 hours weekly for free
  • Weights.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio UI Colab: max 4 hours daily, not granted, of GPU

[idle stratus]

nope, that's even worse

[silk elm]

what is the minimum requirements and high end requirements

[idle stratus]

also rvc doesn't have an official scientific paper, but you can check a complex fan blog https://gudgud96.github.io/2024/09/26/annotated-rvc/

[silk elm]

5070 ti good ?

[idle stratus]

yeah ofc

[silk elm]

im actually specified in large language models i usually use online platforms like amazon sage maker

does amazon sage maker's gpus help in this purpose

?

i can write the code and it uses a strong gpu / cpu for trainning

[idle stratus]

you can use any cloud service you want, we just don't usually help with those because we don't use those services, so you'd have to train the code yourself

also cpu training takes A LOT of time, don't bother with it

[silk elm]

yeah as long as it does the processing part which is the issue here for the gpu it does the job and it helps with testing samples quickly

it does have
8 NVIDIA A10G Tensor Core GPUs and second generation AMD EPYC processors

192 vCPUs, up to 100 Gbps of network bandwidth, and up to 7.6 TB of local NVMe SSD storage

[idle stratus]

Yeah it's good, you just need to write the notebook code to adapt to the cloud service yourself

[silk elm]

okay now to the dataset part is there any available datasets that if i want to use to try this out i don't want to waste time because they can cost me by usage

it cost 1$ per hour

so i want to write my code first and use the training and testing on this gpu

[idle stratus]

we don't share pre-made datasets, you can check how to make your own in https://docs.aihub.gg/rvc/resources/dataset-isolation/

[silk elm]

this is the most important part 🙂

okay how do i test my models with a software ? i can try many samples but its going to be hard to test different parameters using code snippets

[idle stratus]

there isn't really a "test dummy" dataset

[silk elm]

nvm i went through a guide here

but i have a question

those are the parameters that i meant

what are those

is it the quality by HZ ? for audio ? like my microphone is 48khz

and those are the embedders based on the category of the audio ?

[frigid flicker]

These settings look similar to a typical W-Okada program, another distinct program that uses RVC voice model, which both chunk and extra present; the GUI itself doesn't seem to look like a typical RVC fork. Which program are you using at this moment?

[thorny bronze]

stop using the sketchy voice changer application

[thorny bronze]

also crepe_full is too slow compared to the recommended rmvpe

[silk elm]

i don't know which program to use you guys tell me which is the best

i just searched it up since no one answered at that time

[frigid flicker]

Google Search won't gonna be your reliable result on this one. However, you have told here before that you own different PCs (a laptop/PC with GeForce RTX 3050 Ti and an Apple iMac), so I'm not sure which one is your main one. For realtime voice changer, try W-Okada; this one has versions for both Windows and Mac, but Windows is always preferred over Mac. https://docs.aihub.gg/realtime-voice-changer/local/deiteris-w-okada-fork