Apply for community grant: Community Project

#1
by Kangarroar - opened

We want to do DIFF-SVC Inference on Cloud, Diff-SVC is a “Singing/Speaking Voice Conversion via diffusion model”
The Idea of this is that users can upload their own models or use preloaded models or load models from HuggingFace for DIFF-SVC, input a wav file and the program will render out the wav file using the user model (.ckpt file)
And implement the API on DIFF-SVC SYNTH being developed by me so users can do the same.

This is DIFF-SVC Inference working locally taking a WAV file.
First audio player is the input file
Second audio player is the output file that used the model (In this case Namine Ritsu DIFF-SVC Model)

This is the SYNTH that is being developed:
photo_2022-12-30_07-18-51.jpg

we need one gpu cuda compatible to run it without problems or without rendering times higher than 10 minutes per audio

Kangarroar changed discussion title from Apply for community grant: Personal project to Apply for community grant: Community Project

Hey there! We'll provide a free temporary GPU Grant. They are usually left for a couple of weeks, and if there is significant usage we can leave it there. Would that work for you?

Hey there! We'll provide a free temporary GPU Grant. They are usually left for a couple of weeks, and if there is significant usage we can leave it there. Would that work for you?

Yes! It would be nice and would work! Whatever GPU that goes with PyTorch works, I tried doing CPU inference but I can't modify serializer.py on root folder so... And we have tested CPU Inference only and it has taken more than 10 minutes to finish a 4 minute input audio

Awesome! The GPU is now granted. You might need to do some changes in your code to ensure you're using the GPU https://huggingface.co/docs/hub/spaces-gpus#frameworks for more info.

Have fun!

Thanks!!!

osanseviero changed discussion status to closed

Sign up or log in to comment