#anyone knows why suddenly the voice begins sounding glitchy then back to normal?

1 messages · Page 1 of 1 (latest)

bleak rune
#

Problem Details:

Full GPU Name:
NVIDIA GeForce RTX 4070 SUPER

Operating System:
Windows 11

Detailed Description of the Problem:
I’m getting brief, random audio corruption while using voice features. The voice suddenly sounds glitchy for about 1second, then returns to normal without errors. This repeats sometimes intermittently.

Where it happens: [okada webui]

When it happens: [randomly when talking]

What it sounds like: metallic/bitcrushed/robotic, sometimes time-stretched or stuttery

Frequency: [pretty much all the time]

No explicit error messages. Network looks normal.

Expected: continuous clean audio output

Actual: short bursts of corrupted audio that self-resolve

humble meteor
#

There can be many things that are causing this to happen.
You should check to see if your input audio is also changing along with the output. You can do this by using the serverIO analyzer.
Also check if anything happens to spiking your usage of your GPU or CPU. There should be a small window showing your GPU usage, and you can check CPU usage through task manager. (I doubt this is the problem though, because I am assuming your specs are enough)
Another common thing could be something with the configuration of any audio routing that you may doing on your pc. If you are using any kind of virtual cable or voicemeter, try setting your output directly to your headphones and double check to see if you still have an issue.
Outside of this, you may have to try a clean reinstall, ensuring that you have the latest version.

#

I must say though, that many models "glitch" often. This is due to training on unproperly cleaned datasets and bad input audio. If you are consistently getting artifacting in your audio regardless of model, it may be a sign that your input audio has background noise or large spikes in volume. Ensure that you have clean audio when using the voice changer for the best results.
You can test the model itself by using "file" as an input and inputting a clean audio file of talking to test. If it still glitches, it is likely there is some software or hardware issue rather than audio. If it doesn't, I'd think it was an audio input issue.