https://livekit.io logo
Join Slack
Powered by
i tried livekit inference: livekit/elevenlabs/scri...
# ask-ai
c
i tried livekit inference: livekit/elevenlabs/scribe_v2_realtime it's going in timeout.
t
Thanks for reaching out, Filippo. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Thanks — I’m sorry you’re running into timeouts with 
elevenlabs/scribe_v2_realtime
. I can help get this sorted, but I need a little more information to diagnose the cause. I’m responding to your message about timeouts when using the LiveKit inference model 
elevenlabs/scribe_v2_realtime
. To narrow this down, please answer the following: 1. How are you specifying the model ID in your code? ◦ The correct model ID is 
elevenlabs/scribe_v2_realtime
(do not include a leading 
livekit/
). 1. Where does the timeout happen? ◦ Is it when creating the AgentSession, during live audio streaming, on the STT request, or at another step? ◦ Share the exact error message or stack trace you see (copy/paste). 1. Environment details: ◦ Are you running locally or on a cloud VM? If cloud, which provider and instance type (e.g., AWS t3.large, c6i.xlarge)? ◦ If on AWS, are you using burstable instance types (t3/t4)? Burstable types can hit CPU credit limits and cause timeouts. 1. AgentSession config (redact any secrets): ◦ Share the relevant snippet showing how you construct 
AgentSession
(STT block and any 
vad
, 
timeout
, or network-related settings). 1. Repro steps and frequency: ◦ Does the timeout happen every run or intermittently? ◦ Can you reproduce it locally? 1. Additional runtime info: ◦ CPU, memory, and network usage when the timeout occurs (if available). ◦ Whether you’re using VAD (voice activity detection). Using a VAD like 
silero.VAD.load()
is recommended for reliable turn/interruption handling. What I can do next once you provide the above: • If it’s a configuration issue, I’ll point out the exact fix (model ID, session settings, VAD usage). • If it looks like an infrastructure/CPU issue, I’ll recommend suitable instance types to avoid throttling. • If the logs show a different root cause, I’ll provide targeted troubleshooting steps. Send the requested details and I’ll help you pinpoint and fix the timeout. Sources: ElevenLabs STT | LiveKit Documentation | Troubleshooting Latency and Timeout Errors with Turn Detection on AWS | LiveKit turn detector plugin | LiveKit Documentation | Deepgram STT | LiveKit Documentation
c
i confirm i used "`elevenlabs/scribe_v2_realtime"` other stt in livekit inference works
t
Thanks — that’s helpful to confirm. Since other STT models work and you’re using the correct model ID, we’ll focus on narrowing down what’s unique about 
elevenlabs/scribe_v2_realtime
in your environment. Please provide the following so I can pinpoint the cause and give a targeted fix: 1. Exact error/log output ◦ Copy/paste the timeout error or stack trace you see (redact secrets). This helps identify whether it’s an HTTP/connection timeout, internal inference timeout, or something else. 1. Where the timeout occurs in your flow ◦ Is it immediately when you start the model, after streaming audio for a bit, or when sending an STT request and waiting for a response? 1. Code snippet of your AgentSession/STT setup (redact keys) ◦ Share the 
stt
block and any VAD or timeout parameters you pass. Example lines like:
Copy code
stt=inference.STT(model="elevenlabs/scribe_v2_realtime", language="en", ...)
• Tell me if you use a VAD (for example 
silero.VAD.load()
). 1. Environment details ◦ Local or cloud? If cloud, which provider and exact instance type (e.g., AWS t3.large, c6i.xlarge)? ◦ If AWS: confirm whether you’re on burstable instance types (t3/t4). 1. Reproducibility and pattern ◦ Does the timeout happen every time or intermittently? ◦ Does it happen with short clips, long streams, or both? 1. Resource and network metrics when timeout happens (if available) ◦ CPU, memory, and network usage around the failure time. What I’ll do after you share these: • If logs show an instance-CPU throttling or similar, I’ll recommend non‑burstable instance types and configuration changes. • If it’s an STT-specific error, I’ll suggest configuration tweaks (VAD, timeouts) or next steps to reproduce reliably. • If the error is network/connection-related, I’ll walk you through network checks and retry/backoff strategies. Send the logs and the small config snippet and I’ll help you troubleshoot the timeout step-by-step.
2 Views
Open in Slack
PreviousNext
Powered by