Hi @K3, this works for us! For the moment audio is not cutting while subtitles continue. Thanks for the suggestion. @omar.venturi try it!
@juliana, just to double check, we’re talking about an integration in Unreal Engine, right?
Also, have you tried removing FaceSync? If you disable or remove FaceSync, does the avatar’s face still move accordingly, or does it stop responding?
Hi @omar.venturi, yes, it is about integration in Unreal Engine. We didn’t remove FaceSync, we only did what K3 suggested, adding in project settings
1 Like
Thanks for confirming, @juliana.
It looks like both you and @heivoll followed the same suggestion from K3 (updating the project settings in Unreal Engine), but you’re getting different results: for you, everything is working correctly, while for heivoll the issue still persists.
@heivoll, could you please share some more details about your setup? For example, which version of Unreal Engine are you using, which version of the Convai plugin, and whether there are any other differences in your configuration that might help us understand why the fix worked for Juliana but not for you?
Maybe we can pinpoint what’s causing the difference.
@omar.venturi Our setup is using UE5.5 and plugin version 3.5.4, running on Linux servers. I can’t think of anything else particular to our config.
For us it seemed to work at first, but after a while the issue came back. It goes up and down a bit on how often the errors occur anyway, so sometimes it’s difficult to test how changes affect this.
@K3 I’ve removed the face sync component (and still having the extra parameter) now. As far as I can tell, we’re not getting the issues with audio stopping completely now, though there have been some minor glitches in the audio. I can’t say 100% sure that the issues are gone, as it always seems to go up and down how prevalent they are, but we haven’t experienced it in the last 10-15 play sessions.
I’m assuming this is meant as a test to give you more info? We can’t run the production environment without facial animation. Just let me know if you need any extra background info. from our end.
1 Like
Alright, could you please try testing after removing the extra parameter as well and keeping the FaceSync component removed? This will help us isolate the cause more clearly.
If the issue returns, it’s likely a buffering issue. If it doesn’t, then the problem may be related to lipsync not fully engaging. Let us know what you observe!
@heivoll Did you check?
@K3 We’re in the process of testing it, needed some more testing first to completely rule out issues when both the changes are present. We’ll have the results from testing the last setup tomorrow.
In general this takes quite a long time to test, both since it’s not always triggering, and because we have limited resources.
@K3 From what we can tell, the issue does not occur when FaceSync is removed, even when the buffer setting is also removed. So looks like it could be related to FaceSync.
1 Like
Great, thanks.
Hi @K3 and team,
unfortunately I can confirm that the audio-cut issue is still present.
Here are my latest tests (UE 5.5.4 on Windows 11, Convai SDK 3.6.0):
| Test | MinBuffer | FaceSync | Voice | Family | Error |
|---|---|---|---|---|---|
| #01 |  |
 |
Arnold | ElevenLabs voices | Yes |
| #02 | |
|
Custom | ElevenLabs private | Yes |
| #03 | |
|
Custom | ElevenLabs private | Yes |
| #04 | |
|
Custom | ElevenLabs private | Yes |
| #05 | |
|
Italian Male 3 | GCP voices | No |
Key findings
- The issue appears to be tied to ElevenLabs voices (both public and private).
- It’s intermittent: sometimes every reply, sometimes 1 in 4, with noticeable variations across the day (less frequent during night / very early morning, CET).
- Tested on different networks (wired & mobile) – the cut still occurs.
- Disabling FaceSync or setting
MinBufferDuration = 0made no difference: cuts keep happening. - With GCP voices I couldn’t reproduce the problem after ~10 minutes of long interactions.
IDs for reference
| Test | Character ID | Session ID |
|---|---|---|
| #01 | caeb32a0-cd03-11ef-a528-42010a7be016 | d2b1ef15384e158d272b633a58496091 |
| #02 | 09f85624-eea8-11ef-85b7-42010a7be016 | 3349b83a8e97f16e5818b3ba93cc3ccd |
| #03 | 09f85624-eea8-11ef-85b7-42010a7be016 | dea7d0b256f689c70f5c8b73931efb33 |
| #04 | 09f85624-eea8-11ef-85b7-42010a7be016 | d18e93a8c3908b6371f326403accdc0f |
| #05 | caeb32a0-cd03-11ef-a528-42010a7be016 | cf08495ec0dd7a2944eea5991f100c5f |
This problem has been active for almost a month now and it’s becoming critical for production use.
I noticed that Convai SDK 3.6.1 was released today (02 July 2025); I’ll install it as soon as possible and report back.
Thanks
Omar
1 Like
Thanks for the detailed reports. We’ve attempted to reproduce the issue under the same conditions but haven’t been able to encounter the same behavior.
We’ll continue investigating and, if we identify the root cause, we’ll work on a fix. And could you please share the logs when you encounter the issue?
Just to clarify, the 3.6.1 update does not include a fix for this issue. Please refer to the update notes for more information.
Quick update – first test with Convai SDK 3.6.1
I just installed 3.6.1 and ran a quick test.
| Parameter | Value |
|---|---|
| Character ID | 09f85624-eea8-11ef-85b7-42010a7be016 |
| Session ID | 4a2fbbfb2ed6c00d54e429d3a5b292be |
| MinBufferDuration | disabled |
| FaceSync | enabled |
| Voice | Custom (ElevenLabs private) |
Result
- 10 consecutive replies
- Audio cut occurred only on the 10th reply.
- Mid-session the On Failure event of
ConvaiChatbotfired once (this never happens when the issue is the audio cut), so I assume it was a brief network glitch on my side.
Takeaways
- That’s better (9/10 replies intact), but time of day may still affect the outcome.
- K3 just confirmed that 3.6.1 does not contain a specific fix for this bug, so I’m cautious.
I’ll run more tests tomorrow in different time slots to see if the improvement holds.
At this link on my Google Drive you can find 4 full logs (API keys redacted). Hope this helps isolate the root cause.
1 Like
We’re running additional tests, but so far we haven’t been able to reproduce the issue, the audio is not getting stuck at the end of the conversation.
If you could share a screen recording and the relevant log files from when the issue occurs, that would be extremely helpful.
@K3 I’ve sent you a Message with required info.
Hope this can help to solve quickly
1 Like
As you may have seen, I just ran some tests with SDK 3.6.1 and the audio-cut issue is still showing up.
How are things looking on your side?
- Are you still experiencing the cuts?
- Have you found any stable workaround?
@K3 — do you have any updates for us?
Thanks a lot for any feedback! 
@omar.venturi We haven’t tested with 3.6.1, but as I understand it there are no changes there to suggest it would help.
We experience the problems (audio cuts, glitches, etc) as long as the face sync component is present, but have not (in 30+ sessions) experienced the same issues when we’ve removed it. Granted, our sessions are usually short, and there are sometimes smaller glitches, but it does seem like our problem is tied to face sync.
We do not have a workaround, so this is very much affecting our production environment and clients.
I’ve been testing the new MetaHuman in Unreal 5.6 with the new plugin version as well and today caught this error finally. This has been a thorn on my side since UE 5.3
Log attached herewith:
VivienceDH_20250706_181246.log (83.4 KB)
ConvaiGRPCLog: EmotionResponse: neutral 1
ConvaiAudioStreamerLog: State transition: WaitingOnAudio → Playing
ConvaiGRPCLog: Received Text If there’s a particular facet of her story you’re interested in, I’d be glad to explore it further.: | Character ID : 60e087de-76af-11ef-80ce-42010a7be011 | Session ID : 937ac7d376dc776bac37b72bdcaea629 | ReceivedFinalResponse : False
ConvaiAudioStreamerLog: State transition: Playing → WaitingOnAudio
ConvaiGRPCLog: Chatbot Total Received Lipsync Responses: 3404 Responses
ConvaiChatbotComponentLog: Chatbot Total Received Audio: 34.009956 seconds
ConvaiGRPCLog: Received Text : | Character ID : 60e087de-76af-11ef-80ce-42010a7be011 | Session ID : 937ac7d376dc776bac37b72bdcaea629 | ReceivedFinalResponse : True
ConvaiGRPCLog: GetResponse SequenceString: none
ConvaiGRPCLog: Action: None
ConvaiGRPCLog: No more data to read after 3 attempts. Calling Finish…
ConvaiGRPCLog: Calling Stream Finish | Character ID : 60e087de-76af-11ef-80ce-42010a7be011 | Session ID : 937ac7d376dc776bac37b72bdcaea629
ConvaiGRPCLog: Warning: OnStreamFinish: Status.ok():Not Ok | Debug Log: | Error message:enter idle | Error Details: | Error Code:14 | Character ID:60e087de-76af-11ef-80ce-42010a7be011 | Session ID:937ac7d376dc776bac37b72bdcaea629
ConvaiChatbotComponentLog: Warning: UConvaiChatbotComponent Get Response Failed! | Character ID : 60e087de-76af-11ef-80ce-42010a7be011 | Session ID : 937ac7d376dc776bac37b72bdcaea629
ConvaiChatbotComponentLog: UConvaiChatbotComponent Request Finished! | Character ID : 60e087de-76af-11ef-80ce-42010a7be011 | Session ID : 937ac7d376dc776bac37b72bdcaea629
ConvaiAudioStreamerLog: onAudioFinished
ConvaiAudioStreamerLog: State transition: WaitingOnAudio → Playing
ConvaiAudioStreamerLog: onAudioFinished
Hello @Jeasy_Sehgal,
You already shared this error in a different post. This is something completely different.