# EOS\_RTCAudio\_AddNotifyAudioBeforeRender is overflowing our audio buffers. Do we need bigger buffers? Is this expected behavior, or have we done something wrong? 

**URL:** <https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436>\
**Category:** Multiplayer\
**Tags:** epic-online-services, question\
**Created:** [January 21, 2026, 2:01am UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436 "2026-01-21T02:01:11Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![anonymous-edc](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@anonymous-edc](https://forums.unrealengine.com/u/anonymous-edc)\
**Post date:** [January 21, 2026, 2:01am UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436/1 "2026-01-21T02:01:11Z")

</div>

We’ve integrated, for the most part, EOS Crossplay into our upcoming game. And RTCAudio is working “most” of the time for our player lobbies.

However, sometimes when a player is talking it seems like the callback we get from EOS\_RTCAudio\_AddNotifyAudioBeforeRender will unexpectedly spam our audio system and overflow the buffers we have to receive the audio data. These buffers are sized to hold 100ms of audio, 10x the 10ms of audio that is supposed to come in from each callback of AudioBeforeRender.

Is this expected behavior? Does EOS RTCAudio play “catch up” with a players communication if, for some reason, their is a connection issue? How big should we make our audio buffer size if this is intended behavior? 1 full second? More? Are we expected to “fast forward” the audio a little if our buffer is starting to fill faster than its draining?

Or is this a sign we’re doing something wrong? We’re currently letting the EOS SDK handle capturing the microphone. We’re only plugging into the AudioBeforeRenderer as we’d like to, eventually, do more complex per-player volume adjustments like proximity chat. Also, our implementation is pretty by-the-book/by-the-sample.

And to reiterate, it is working normally most of the time. The issue seems more likely to occur the longer a specific player has been talking.

Happy to provide any further details if needed, but any insight into this problem would be greatly appreciated!

~~~~

**UPDATE 1/26/2026**

** ~~~~ **

We were able to get a playtest where I was able to monitor our audio buffers in realtime as we made use of EOS Voice Chat.

One user, on a mac, to my Window’s machine suddenly sent a huge glut of sound data, eating up the rather large audio buffer we have reserved for EOS coms. The buffer should be able to handle a glut of sound send in delayed fashion as long as there some silence behind it, but then the problem came that EOS never sent actual, total silence.

From what I could tell, EOS Voice Chat began sending a mostly silent buffer with a single frame at the front with a non-zero value. Does EOS, if it doesn’t receive new data, just resend the last frame? Is this a bug where EOS is not properly clearing that buffer to send silence? Or is this a bug with the MAC implementation where it’s not properly silencing itself and it’s actually resending some audio over and over again.

I really don’t think I’m accessing the audio buffer pointer wrong. It’s just an int\_16 pointer and I’m accessing it with just a basic 0-indexed for loop.

Again, any help would be greatly appreciated!

---

<div class="post-metadata">

**Author:** ![jamesherd](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@jamesherd](https://forums.unrealengine.com/u/jamesherd)\
**Post date:** [February 19, 2026, 3:29pm UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436/2 "2026-02-19T15:29:33Z")

</div>

Hi, apologies for the delay in responding to this question. What you’re describing sounds like a known bug in the SDK. Do you notice any voice distortion when encountering the issue, such as voices becoming robotic-sounding?

---

<div class="post-metadata">

**Author:** ![anonymous-edc](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@anonymous-edc](https://forums.unrealengine.com/u/anonymous-edc)\
**Post date:** [February 25, 2026, 6:12pm UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436/3 "2026-02-25T18:12:11Z")

</div>

Yes, we are noticing distortions when this begins happening. From my understanding, it’s because of the fact we run out of buffer, a little bit of the buffer clears, and then sounds that don’t naturally follow what was in the buffer get played.

Is there a known, preferred workaround for the bug? Right now, for our “push to talk” configuration, we’re forcefully silencing the buffer in OnRTCLobbyAudioBeforeSend, but that doesn’t fix the issue for our “always on” mode.

---

<div class="post-metadata">

**Author:** ![jamesherd](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@jamesherd](https://forums.unrealengine.com/u/jamesherd)\
**Post date:** [March 9, 2026, 4:48pm UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436/4 "2026-03-09T16:48:35Z")

</div>

Hi, it sounds like you have indeed encountered the known bug then. This is caused by latency in the audio thread, which results in all missed frames being sent at once in an attempt to catch up. This leads to unintended behavior in the WebRTC service, which sends duplicate packets filled with silence, causing the robotic voice effect.

Other customers have successfully mitigated this issue using a hash check to identify and discard the silent audio packets. These will all share the same hash.

---

<div class="post-metadata">

**Author:** ![anonymous-edc](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@anonymous-edc](https://forums.unrealengine.com/u/anonymous-edc)\
**Post date:** [March 9, 2026, 5:34pm UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436/5 "2026-03-09T17:34:54Z")

</div>

So, to confirm the way to mitigate the issue:

1) Each time we receive a call from the EOS\_RTCAudio\_AddNotifyAudioBeforeRender callback, we should hash the EOS\_RTCAudio\_AudioBuffer::Frames buffer.

2) If we receive a buffer that generates the same hash as the previous, we should discard it.

Is that correct?

---

<div class="post-metadata">

**Author:** ![jamesherd](https://d3kjluh73b9h9o.cloudfront.net/original/4X/2/8/5/28502927f1d97ba1051b3b47ccfe17233bf41ef0.png) [@jamesherd](https://forums.unrealengine.com/u/jamesherd)\
**Post date:** [March 13, 2026, 4:38pm UTC](https://forums.unrealengine.com/t/eos-rtcaudio-addnotifyaudiobeforerender-is-overflowing-our-audio-buffers-do-we-need-bigger-buffers-is-this-expected-behavior-or-have-we-done-something-wrong/2858436/6 "2026-03-13T16:38:25Z")

</div>

It’s a little more complicated than that, as the silent packets are interwoven between the genuine packets. The silent packets are effectively duplicates (just empty), and so they will be received in quick succession following a genuine packet. A confirmed mitigation is to check for the time between received packets, and if any are received in rapid succession (for example, with a delay of less than 10ms), confirm the hash on the second in those pairs specifically. Those packets will likely all contain silence with an identical hash and can be discarded.
