#my mic sounds super chopy when i record or on discord but when i record a file it sounds fine? help

1 messages · Page 1 of 1 (latest)

junior onyx
#

i know its super delayed in the clip but idc if its delayed

#

plsss its so laggy plssssssssss help

fiery badger
# junior onyx

What is your PC GPU? You seem to using the older original W-Okada DirectML version.

verbal ridge
#
  • The echo is likely caused because you play the sound through your speakers and your microphone hears that.
  • You can turn on Noise Suppression 2 to relief some of that. There are other tools that can also assist with sound issues.
  • Try putting the audiodg.exe process priority to high, then set the affinity to one processor thread. (disable the others). This way you can do other stuff that use up resources while those will not be affected all that much by this setting. I recommend picking the 3rd last processor in the list which tends to be used the least, so if you have 8 processors listed, put it on 4. (since processor 7 is going to be the last one in that case).
    When you restart your PC these settings will be reset to normal, so do not worry on that part.
  • When you change the chunk size or extra value, a bug happens... you need to swap the processor unit before that clears:
    Set the processor unit to CPU once and then back to GPU0. This will also get rid of some of the issues.
  • Considering you're not using the index file, you can lower your chunk size by a ton...
    this is especially the case if you're using a fork (such as deiteris's fork or tg-develop's fork) for this voice changer.
    Actually I'd recommend using one of these when you use AMD (which I assume you do considering you're on a DML release), because they are less laggy (even though these two forks have their own bugs).
    Let me link tg-develop's fork of this same voice changer:
    https://github.com/tg-develop/voice-changer
    (Note: ONNX support on this one is completely broken, but regular tensorflow (pth) will work just fine. Just do not turn on 'convert to onnx' or use onnx models.)
    Note 2: this one doesn't support beatrice v2 models either, which are a lot faster (though the default models are of low quality). for beatrice I'd recommend wok's beatrice client, rather than this combined one anyway.
#

To explain recording: no audio leaves your computer till you press play, so there is no echo.
Discord has its own noise suppression and echo cancellation system. The echo cancellation and volume autotuning system discord uses are by default choppy. I recommend disabling auto volume adjusting in discord at the very least. For echo cancellation... it depends. You need to test out voice chat with someone and see where the problems lie with just default audio; and then after that see if there are less or more problems with the voice changer.)

#

To route audio, I think it's better to use Elgato's Wave Link (version 3, the beta software) which is free to use instead of the VAC or VB-cable.
It serves the same purpose, but is easier to use essentially. It lets you choose what audio goes where.
VAC is on the guide due to the ability to set the max delay to 2ms, which makes it miliseconds faster than VB-Cable (theoratically at least). It depends on your setup, since it's really the max delay and not the delay.