Release blocker: Unreal 5.7 sessions connect without SONIOX STT
What happened?
In Unreal Engine 5.7, the Convai Court Voice Router connects successfully and the microphone streams audio. However, some sessions never receive the session-config packet that provisions SONIOX STT. No user-transcription packets are then returned, so push-to-talk voice commands time out even though Convai reports the session as Connected. Restarting or reconnecting does not reliably fix it.
What did you expect?
We expect every successfully connected Convai session to provision SONIOX STT and return user-transcription packets when microphone audio is streamed. If STT cannot initialize, the plugin should report a clear failure or STT-not-ready state instead of reporting the session as fully Connected.
Steps to reproduce
Open the Unreal Engine 5.7 project.
Start Play In Editor in the courtroom level.
Wait for the Convai Court Voice Router to report Connected.
Hold Left Alt to activate push-to-talk.
Say: “Call the defendant forward.”
Release Left Alt.
Observe that the microphone streams audio, but no transcription is returned and the command times out.
Check the Unreal log and confirm that the session did not receive session-config with “stt”:“SONIOX” and produced no user-transcription packets.
Unreal logs or error messages
[Convai] Court Voice Router: Connected
[Convai] Attendee connected
[PTTDIAG] PTT allowed → reset and unmute
[PTTDIAG] JudgeMic UNMUTED
[PTTDIAG] Microphone audio streaming started
[PTTDIAG] Left Alt released
[PTTDIAG] RETRY timeout - no final transcription dispatched
Expected but missing:
[Convai] session-config: {“stt”:“SONIOX”,“type”:“session-config”}
[Convai] user-transcription
[PTTDIAG] OnJudgeTranscription
Please share the full Unreal Engine logs so we can review this further.
This does not appear to be specifically related to SONIOX. Please also make sure that the correct microphone device is selected from the F10 settings panel and that microphone permissions are enabled for your application.
Thank you. I confirmed microphone permissions are enabled.
Using the F10 settings panel, I explicitly selected:
Microphone Array (Intel Smart Sound Technology for Digital Microphones)
The F10 Record/Stop microphone test succeeds and plays the recorded audio clearly. The in-game microphone test also passes.
I have attached two redacted full Unreal Engine logs from the same project and plugin version:
A working session where transcription packets were received.
A failing session where the microphone streamed but no user-transcription packets were returned.
Please compare the two sessions. The downstream game command path also works when transcript text is injected directly, so the failure occurs before the game receives transcription.
Microphone Array (Intel Smart Sound Technology for Digital Microphones)
Other microphone devices do not work correctly on this PC. The Intel Microphone Array passes the Convai F10 Record/Stop playback test and the game’s microphone test. Windows microphone permissions are enabled.
Despite that successful microphone capture, some Convai sessions return no user-transcription packets. I have attached one working and one failing full Unreal log for comparison.
my apologies, I reviewed the Audio and Microphone Issues guide and verified the required checks.
The correct Intel Microphone Array is selected, opens as 2-channel 48 kHz, passes the F10 recording/playback test, and Windows permissions are enabled.
Both working and failing logs show the Convai Player and Court Voice Router session initializing. The failing session also detects the user as the active speaker.
The key difference is:
Working session: 16 successful stt-toggle server responses.
Failing session: zero stt-toggle server responses.
The Blueprint reports that PTT unmuted locally, but the Convai session never acknowledges STT unmute/mute in the failing session. Please investigate why the realtime session intermittently does not receive or acknowledge the stt-toggle request.
Result:
Holding Push-to-Talk for about 1 second after finishing speech does not fix the issue.
Tested both cases:
Release immediately after speaking.
Keep holding Push-to-Talk for about 1 second after finishing, then release.
In both cases:
F10 microphone test still works.
The selected device is still Microphone Array (Intel Smart Sound Technology for Digital Microphones).
Windows microphone permissions are enabled.
The game’s own microphone test passes.
No usable user-transcription reaches the game.
The downstream game command path still works when transcript text is injected directly.
This makes the issue look unrelated to early PTT release timing.
Can you please confirm whether this could be related to the Convai Unreal Engine plugin version we are using with Unreal Engine 5.7?
Specifically:
Is there a known issue in this plugin version where sessions connect and local microphone capture works, but no user-transcription packets are returned?
Is there a newer Unreal plugin version that fixes STT toggle/session transcription reliability?
Should we upgrade the Convai Unreal plugin for Unreal 5.7?
If yes, which exact plugin version should we use?
Are there any required Blueprint or C++ changes after upgrading?
The key difference remains:
Working session: stt-toggle server responses and user-transcription packets are received.
Failing session: local PTT/mic capture appears active, but no user-transcription packets are returned.
We tested the newer Convai Unreal SDK-V4 plugin 4.0.0-beta.24 in the production Unreal 5.7.4 project.
Result: the issue still occurs.
Evidence from the log:
PTT press is received.
Convai starts streaming audio.
Mic device opens correctly:
Microphone Array (Intel Smart Sound Technology for Digital Microphones), 2 channels, 48000 Hz.
Convai detects the user as active speaker.
PTT release stops streaming audio.
No final transcription is returned.
The game times out with:
[PTTDIAG] RETRY timeout - no final dispatched
This means beta.24 did not resolve the issue. The microphone, PTT input, and local audio capture are working. The missing part is still user-transcription output from the Convai realtime session.
Please escalate this as a realtime STT/session issue, not a microphone-device issue.
When text is injected directly into our game command parser, the command parser executes as expected. The Q radial / deterministic command path also works. The failure is isolated to Convai realtime user transcription.
What works:
PTT press/release is received.
Convai starts/stops audio streaming.
The microphone opens correctly.
Convai detects the user as active speaker.
Text-injected commands work in-game.
What does not work:
No final user transcription is returned from the Convai realtime session.
The game times out with [PTTDIAG] RETRY timeout - no final dispatched.
So this does not appear to be a microphone hardware, PTT binding, or downstream game parser issue. It appears to be the realtime STT/session path not producing final transcription output in Unreal 5.7.4 with SDK-V4 4.0.0-beta.24.