https://vapi.ai/ logo
Join Discord
Powered by
Stitched conversations with 11labs scribe v2
# support
f
Hi, I’m seeing what looks like a Vapi + ElevenLabs Scribe v2 issue on phone calls. When I use: - transcriber: 
11labs
- model: 
scribe_v2_realtime
the 
conversation-update
payloads and final 
end-of-call-report.artifact.messages
collapse into just 2 cumulative messages: - one growing 
user
message - one growing 
assistant
message But when I switch the same assistant back to Deepgram, the conversation history is correct and each turn is appended normally. Important detail: - in the Scribe call logs, per-turn transcript events are still being emitted - but the conversation / artifact messages are merged by role instead of stored as separate turns So this seems specific to the Scribe integration path, not my backend. Broken Scribe call: 
019d4d73-8570-7000-85cb-a34b0672019b
Working Deepgram call: 
019d4d83-ec0a-7445-9441-abe25fdacb28
Expected: 
artifact.messages
/ 
conversation-update.conversation
should contain separate turn-based messages. Observed with Scribe: they get concatenated into 2 cumulative blocks. Can you please check whether this is a known issue with 
11labs
+ 
scribe_v2_realtime
?
v
To address the cumulative message issue with 
11labs
+ 
scribe_v2_realtime
, please use the Vapi dashboard to compare the call logs for both the broken and working calls. Then, report the issue with both call IDs and your configuration details via Vapi support or Discord for further investigation. Example steps: 1. Go to 
Observe > Call Logs
in the [Vapi dashboard](https://dashboard.vapi.ai/). 2. Review and export the logs for both calls. 3. Share these details with support for targeted troubleshooting. This approach ensures the Vapi team can analyze the integration path and help resolve the formatting issue. Source: - [Debugging voice agents](https://docs.vapi.ai/debugging)
f
I added both call log ids in my message. I inspected the call logs.
s
Hi, thanks for reaching out.
We're investigating this and will soon get back to you.
Hi, Thanks for the clear report and for sharing both call IDs.
We checked your Scribe v2 call (019d4d73-8570-7000-85cb-a34b0672019b), and this behavior is confirmed on our side. The Scribe v2 transcript events are coming through, but the timing metadata is not in the format Vapi uses to separate turn-by-turn messages. As a result, all user speech is merged into a single user message, and all assistant speech into a single assistant message in both conversation-update payloads and the final end-of-call-report. Your Deepgram call (019d4d83-ec0a-7445-9441-abe25fdacb28) shows correct behavior because Deepgram provides the timestamp data Vapi expects, so messages are split properly per turn. This is a known issue with the ElevenLabs Scribe v2 integration path, and our team is working on a fix.
Temporary workaround: Please switch your transcriber to Deepgram for now. It supports the same flow and will restore proper turn-based message structure. We’ll update you as soon as the Scribe v2 fix is released. Best, Vapi Support
f
Thank you!
s
No worries!
If you face any further issues, feel free to open a new ticket.
j
Hi @Shaunak do we have a ETA on the fix?
I have had problems with Scribe for the last month
@Shaunak
s
Hi Jarod, Apologies for the delay! Unfortunately we don't have an ETA on the fix for this just yet.
In the meantime, we'd recommend switching to Deepgram or another supported transcriber to restore proper turn-based message structure. We'll share an update as soon as there's progress on the Scribe v2 fix.
PreviousNext
Powered by