search

Anjok07 / ultimatevocalremovergui

GUI for a Vocal Remover that uses Deep Neural Networks.
MIT License
17.35k stars 1.3k forks source link

(Linux-Bug) - Very high pitched results #926

Open Husky110 opened 10 months ago

Husky110 commented 10 months ago

Hi - I hope this belongs here...
I've installed UVR on my system (Ubuntu 22.04) and everytime I try to run anything through the DeReverb-DeEcho-thingy (sorry, new to this) the result changes the pitch to drastically highs. I mean were talking like "+200" (it fells like it).
Is this a known issue?
Right now I try to sepparate this video: https://www.youtube.com/watch?v=IxcxyGDD38E
I'm trying to use the vocals-track provided by vocalremover.
Model is UVR-DeEcho-DeReverb Example-output provided here: https://vocaroo.com/1eIcT4HvNWBk (listen at your own risk! I highly suggest to turn down the volume...)

DinaMoon commented 10 months ago

I have about the same problem with all models. Also on Ubuntu 22.04.3 LTS. It works on Windows 10 with the same hardware configuration without problems.

Husky110 commented 10 months ago

@DinaMoon - Ouch... A Linux-Only-Bug...

rexendevar commented 10 months ago

I'm getting the same thing. It looks like UVR is getting confused somewhere with regards to sample rate or buffer size because little sections of the waveform seem to get compressed and duplicated over screenshot2023-11-03_21-47-36

rexendevar commented 10 months ago

regardless of model or method btw. i even tried running on the CPU because i know how bad nvidia drivers can be but nothing I did fixed it. i looked through separate.py, and while i did find a place where the sample rate was listed with an extra 0 on the end, changing it did not seem to affect the output at all.