I am writing to urgently report a critical issue I am facing with the Convai integration in a Mixed Reality application for Meta Quest 3, developed in Unreal Engine 5.4.
Although the integration seems correct, the audio consistently cuts off before the character finishes their explanation. The speech stops abruptly in the middle of the dialogue, which disrupts the experience.
This issue is critical as I need to deliver this project to my client very soon. Could you please advise on a fix or troubleshooting steps specific to this MR/Quest 3 environment?
Thank you for the response and for offering to investigate.
At the moment, the Unreal Engine logs do not show any errors or warnings related to audio or TTS. The application is still under development, so unfortunately I’m not able to share screenshots or video recordings yet due to project confidentiality.
What is consistently happening is the following:
the AI starts answering correctly, but the audio fades out and stops before the sentence or explanation is finished. The text response itself is complete, but the TTS audio is cut mid-speech.
This is happening on Meta Quest 3 in a VR project using Unreal Engine 5.4, and it seems related to how spatial audio / attenuation or voice handling behaves in VR, rather than a runtime crash or log-level error.
Once the project reaches a stage where I can share visual material, I’ll be happy to provide a video. For now, I wanted to clarify that:
There are no UE log errors
The issue is audio cutting before TTS completion
The problem is reproducible and consistent
Please let me know if there are any specific Convai audio or TTS settings you would like me to test in the meantime.
I really need the community’s help. I have a critical client delivery in less than 3 days, and I am currently blocked by two major issues in my Mixed Reality medical project (Unreal Engine 5.4 targeting Meta Quest 3):
Premature Audio Cut-off: The AI starts speaking, but the audio fades out or cuts off before finishing the sentence (as if an interruption was triggered), even in a silent environment.
High Latency: The response time is too slow for a fluid interaction.
I have checked the logs and there are no errors. I’ve tried adjusting Blueprint logic for the “Talk” button, but the issue persists.
Given the tight deadline, is there any quick workaround, config tweak, or specific VAD setting in Unreal that I might be missing to force the audio to play until the end?
If you plan to use GCP voices, pick ones whose names or descriptions clearly specify the style and language, for example: “Despina (Smooth, Gentle Japanese Female Voice).” You can also try using one of the Azure voices.
For the LLM, I recommend trying 2.5 Flash Lite or 4.5 Haiku.
I have already tested this setup, and unfortunately it is simply not working. In its current state, I cannot implement this solution for my client.
I would really like to use your solution with my customers, but the issue persists on my end: the AI speech continues to be cut off before finishing, even though there are no errors shown in the logs. I also noticed that this same issue has already been discussed on the forum, which makes it seem that the problem may not be on my side.
Since I have subscribed to the monthly plan, I will continue testing and working together with you until the end of the subscription period. However, if we are unable to resolve this issue, I will unfortunately not be able to move forward with your solution for my clients.
I truly hope we can work together to identify and fix this problem.
Thank you very much for your support and attention.
I have an important update that might help isolate the root cause.
This issue is NOT limited to Unreal Engine 5.4 or the Meta Quest 3. I have tested the character directly on the Convai Web Dashboard (Playground), and the behavior is broken there as well:
Microphone Failure: When I use the microphone on the website to speak to the avatar, it often fails to respond entirely.
Text vs. Voice: It seems to respond only when I type the text manually.
Audio Cut-off: When it does respond, the audio cut-off/fade-out issue also happens on the Web interface.
Since this is occurring natively on your platform, it suggests the problem is likely on the backend services or my specific character configuration, rather than my local Unreal integration.
Could you please check my account/character status on the server side?
Yes, but it doesn’t work. I would really like to continue using the platform, but it’s been quite difficult. I’m also experiencing significant delays in the AI’s responses.
I am writing to report that I have followed your previous advice and created a brand new character ID to test the system. Unfortunately, the issue persists: the audio still cuts off/truncates before the NPC finishes the explanation.
I am running this on Meta Quest 3 using Unreal Engine. I have already tried the following troubleshooting steps to rule out local configuration errors:
Created a new Character ID as suggested.
Verified that the Project Sample Rate is set to 48000 Hz.
Increased the Callback Buffer Size to 4096 to prevent underruns.
Checked permissions (RECORD_AUDIO and MODIFY_AUDIO_SETTINGS are enabled).
I am heavily invested in using Convai for my upcoming medical XR projects, but this audio truncation issue is a critical blocker for us right now. I urgently need a solution to get the full audio response working on the standalone headset.
Could you please investigate if there is an issue with the Android/Quest plugin specifically regarding audio stream termination?
I would like to ask if you could help me resolve an audio interruption issue I am experiencing on Convai. I really like the application and I am planning to use it in many upcoming projects. However, I need to solve this audio problem before I can confidently continue using the platform.
I would really appreciate your support on this matter and look forward to your guidance.
Thank you in advance, and I look forward to your response.
Best regards,
Will
Thanks for testing with a new character and for the update.
Just to clarify: the suggestion to try a new character was mainly to check response time, not to directly fix the audio cut off. How is the response time behaving now with the new character ID?
On our side, we still have not been able to reliably reproduce the audio cutting off in our internal tests. Because of that, we unfortunately don’t have a short term fix specifically for the issue yet.
In the meantime, my best recommendation is to try switching to a different voice provider and see if the audio cut off still occurs.
