Intermittent gRPC timeouts, Missing UI Transcriptions, Premature OnAudioFinished & UE Crash

Core Platform Voice, Speech & Language
1

What do you need help with?

Bug report

Issue area

Other

Impact

Major - a core flow is broken

Character ID

88aa3dc6-6ed0-11f1-a4a3-42010a7be02e

Session ID

40dc31e2b321c146d7a1c85a1333dad3

Where does it happen?

Not sure

Language

English, Turkish

Voice provider

ElevenLabs, Azure

Voice name

Yunxiao - Azure (currently), Tested ElevenLabs voices

Environment

Uneal Engine 5.7, Playcanvas SDK

Short summary

Intermittently, sent audio fails to transcribe to UI. gRPC stream delays cause premature OnAudioFinished. Exiting causes a crash.

User input

Okuldaki topluluklardan bahset (Or any input tested with both voice and text)

What happened?

Voice & UI Desync: Audio capture is logged (NumberOfAudioBytesSent increases), but transcription intermittently fails to reach the UI chat and generates no AI response.

Missing Session Logs: Unanswered/failed queries never register in the Convai Dashboard session history.

Premature OnAudioFinished: Stream packet drops starve the audio buffer, falsely triggering OnAudioFinished mid-sentence and cutting off the character’s speech.

Crash on Exit: Closing PIE or UE5 after initiating a conversation almost every time causes an Access Violation crash.

Isolation Tests:

  • Network & Hardware: Reproducible across wired/wireless networks, multiple PCs, and blank UE projects.
  • API Key: Persists with newly generated API keys (rules out auth/quota limits).
  • PlayCanvas SDK: Shows severe text/audio stream desync (drops initial audio chunks and speaks more text than transcribed in the UI).

What did you expect?

Audio captured and verified by NumberOfAudioBytesSent in the logs should consistently transcribe to the UI chat and receive a response. The audio buffer should handle gRPC stream delays without triggering premature OnAudioFinished events. Finally, closing the engine should safely clean up the proxy without causing an Access Violation crash.

Steps to reproduce

  1. Open a UE 5.7 project with the Convai plugin.
  2. Send voice or text input. Voice data is sent (NumberOfAudioBytesSent increases), but UI transcription intermittently fails. Even when transcribed (or when using text), the AI frequently generates no response.
  3. Check Convai Dashboard Sessions (these failed queries are completely missing).
  4. When audio streams, buffer starvation triggers premature OnAudioFinished events, causing severe audio stuttering that worsens towards the end of the response.
  5. Close PIE/Editor after a conversation and observe an Access Violation crash almost every time.

Screenshots, recording, audio sample, or logs

Log.txt (33.5 KB)

I checked the relevant documentation, resources, and similar forum posts before posting.

on

I confirm that I did not include API keys, tokens, passwords, secrets, private character data, private conversation logs, or customer data.

on

2

Is there a specific reason why you are using an older SDK version?

3

We are currently using the Convai Unreal Engine SDK Version 3.6.9-hotfix2, which is the latest stable release available on GitHub.

We are aware of version 4.0.0 and we actually tried it first. But since it was still in beta, we found out that the walking feature wasn’t active. That feature is really important for our project, so we decided to switch to the 3.6.9 version.

4

V4 is currently more stable than V3, and many updates and fixes have been released since the initial beta versions.

The grpc related issues you are seeing in V3 are known issues with the older plugin infrastructure. Unfortunately, the recommended solution is to migrate to the V4 plugin.

You can download the latest V4 release here: