Character struggles to deliver speech items. Seems to be caching problem or network connection, but we have a killer connection -
User input
Hi
What happened?
Plays ok through playground, but we are using the API interface and are seeing this on different machine from multiple locations. we have gone through all options, changed LLM models etc.
What did you expect?
Should execute a standard communication. We are seeing this on multiple character IDs. they have different designed convai options.
Steps to reproduce
Use an unreal engine EXE with convai plugin that utilizes the API for speech.
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.
Unfortunately, I’m not fully sure I understand the issue from the description alone.
Could you please share a video recording showing the behavior, along with the relevant logs from the Unreal build?
Also, is there a specific reason why you are using the API directly instead of the Convai Unreal Engine Plugin for this setup? Understanding your integration approach will help us review the issue more accurately.
We do integrate via the Convai Unreal Engine plugin (gRPC streaming, stream.convai.com) and have a live training demonstration coming up, so streaming reliability is critical for us. We did extensive isolation testing and want to share what we found and lay out what we need from you.
What we measured (via a standalone gRPC emulator of the plugin — i.e., not our application code — across every character on our account):
gRPC streaming success dropped hard and has been recovering over the last ~day: roughly 17% → 67% → 92%.
Failures were initially DEADLINE_EXCEEDED (streams hanging ~60s with no audio), and are now UNKNOWN — “Unable to generate audio response” (~8% of calls, even at “recovered”).
The REST getResponse API (api.convai.com) was 100% reliable the entire time — same account, characters, network, and time window. So this is isolated to the gRPC streaming path, not our integration.
What we’re asking for:
Root cause / confirmation: Was there a known incident on the gRPC streaming service or the audio-generation stage over the past week? What was it?
Full resolution of the residual UNKNOWN / “Unable to generate audio response” errors — it’s still ~8% at best right now.
Stability assurance: Can you confirm the streaming path is stable, or tell us what to do to ensure it?
Recommended configuration to minimize exposure — voice provider (e.g., Azure/GCP/ElevenLabs vs. a Fish-Audio-backed voice), foundation model, and any gRPC deadline / retry guidance for the Unreal plugin.
How to monitor streaming-service health in real time (status-page component or an incident channel), so we can see degradation?