_tau_ It seems to be most noticeable on Index mics. my audio settings hadn't changed at all. Unfortunately since it's now in the live build, i can't collect empirical evidence, but is it possible a rollback branch can be arranged?

Catboy Slim: We can't do a full rollback branch for security reasons, but if you are willing to test I can upload a branch with some vaguely relevant changes reverted.

_tau_ sure!

Catboy Slim: I have set up a special beta branch on Steam named , you can opt-in using the password .

Please report if this branch sounds any different on your system! It should otherwise match live.

_tau_ it does sound better! :D

Here is a recording of me switching between branches: https://files.catbox.moe/wb4goh.mp3

Here is a short recording of the live branch: https://files.catbox.moe/2sbwix.mp3

Here is a short recording of the a-test branch: https://files.catbox.moe/mn3qv2.mp3

_tau_ this branch sounds significantly better

_tau_ I've been performing some tests with it too and I also can conclude that there's noticibly less clipping in the a-test build.

I've done the tests using an alt account (sowwy) and went to the same world each time and stood on the same spot, which was roughly 2m apart. I loaded a clean speech sample to play through the mic using virtual audio cable and recorded through OBS.

The waveform shows the peaks are being chopped off, though it's not as obvious as listening to the audio. I've put the audio samples into a shared folder. It's definitely the most obvious when the word "market" is being spoken.

Hopefully this is helpful, for the time being I will continue using the test build, it's just easier on the ears.

Thank you all, this is super helpful! I'll check the diff again and see if I can form a minimal fix to include in the current 2025.3.4 open-beta cycle. I will message here when I have an update.

_tau_out of curiosity, was the change just due to a number being changed somewhere? and if so, could that number be exposed in a config file?

Catboy Slim: No, and no :) It seems to have been an order-of-operations change that came in as a byproduct of unrelated Steam Audio work. In general, we try to not make audio that configurable, since some things affect worlds. Making worlds sound good is hard enough without adding client-specific configs into the mix.

_tau_ gotcha! i do remember before this change, the steam audio beta was a lot quieter than live and the extra headroom was lovely as voices weren't fighting for dominance anymore. if i had my way, the voice volume slider would affect the signal before the compressor, rather than after, but i suppose that's out of scope for this canny. still, i really appreciate your help on this issue!

Catboy Slim: We have a complete rework of the compressor and gain architecture of Steam Audio in the works right now. It will still apply after the compressor, but with different headrooms and a bit softer overall. Excited to see what people think of that, but that's unrelated to here :)

The branch (build 1724) should now have a proper fix. Please let me know if this sounds good again so I can close !

_tau_ sounds much better, there's the odd case of there still being some clipping but that is most likely because:

Eitherway, it's sooooooo much better than the live build right now.

What surprised me the most is how vastly different the steam audio branch sounds like, but in a good way. I will be testing that branch much more now that the clipping issues have been resolved.

I have uploaded the recordings of build 1724 in case those are going to be needed at some point. https://drive.google.com/open?id=1vq5c49irvFaT775oHaaKiX9JU_hGe71E&usp=drive_fs

Will have to wait until friday to see how these changes perform in larger instances, but until then

Thanks for the fixes!

Plaster: Data is clipped in [0,1] on entering and exiting the OPUS compressor step over the network. This has always been the case however. It usually does not present any issues, as all voice gain is applied on the receiver, meaning for this to present an issue, the microphone on the sender needs to produce samples outside the range [0,1], which usually indicates a misconfiguration more than anything.

Plaster i don't like the steam audio beta compared to the open beta (and i much preferred when it was quieter), there's no bass in people's voices and they're louder than the regular beta, the compressor is working harder and voices are crunchier. and voice falloff doesn't apply until you're like 10 feet away. people sound about twice as close as they actually are, and i am definitely getting overstimulated faster.

related ticket: https://feedback.vrchat.com/steam-audio-open-beta/p/1693-voices-are-too-loud-and-over-compressed-after-this-update

StormRel i am not in any beta.

StormRel i am not referring to any beta. opting out does not fix the issue, as the issue is on live.