Transcriber Settings Save/Refresh Failure
# support
m
Hi, I hope this message finds you well. I'm writing to report a critical issue with the transcriber settings that requires immediate attention. I am unable to save modifications to the transcriber settings. The changes are not properly refreshing after being made. As a result, our production agent cannot understand voice inputs correctly. This malfunction has significantly disrupted our operations and is causing considerable problems for our team. Due to the production impact, I kindly request that this issue be addressed with the highest priority. I'm available to provide any additional information needed to troubleshoot this problem. Thank you for your prompt attention to this matter. Best regards, Matteo
v
Hi Matteo! 👋 I understand you're facing a critical issue with the transcriber settings not saving or refreshing properly, which is causing your production agent to misunderstand voice inputs. This is definitely a high-priority issue that needs immediate attention. Based on the information I have, here are some potential reasons and troubleshooting steps for your issue: ## Possible Reasons for Save/Refresh Failures ### Model and Language Compatibility Issues The
codeSwitchingEnabled
setting is only compatible with
nova-2
and
nova-2-general
models. If you're trying to use it with
nova-3
, you'll encounter errors. Similarly, the
nova-3
model only supports
'en'
and
'en-US'
languages - using other languages will result in errors. ### Technical Glitches According to [information from Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1336674769055252594), system updates or glitches can cause settings to disappear or not be saved correctly. There might also be UI issues where the interface indicates settings are saved, but the changes aren't actually applied in the backend. ### Synchronization Delays After making changes, it may take a short period for the changes to propagate and for the assistant to synchronize with the new settings. This delay could make it appear as if the settings aren't being saved or refreshed. ## Troubleshooting Steps ### 1. Check Configuration Compatibility Verify that your transcriber settings are compatible with your selected model. For example: - If using
nova-3
, ensure you're only using
'en'
or
'en-US'
languages - If using code switching, ensure you're using
nova-2
or
nova-2-general
models ### 2. Update Via API Try updating the transcriber settings using a PATCH request to the Vapi API. Here's an example: ``` curl -X PATCH https://api.vapi.ai/assistant/your-assistant-id-here \ -H "Authorization: Bearer token" \ -H "Content-Type: application/json" \
-d '{ "transcriber": { "provider": "deepgram", "endpointing": 300, "model": "nova-2", "language": "en" } }' ``` ### 3. Adjust Endpointing Settings Adjusting the transcription endpointing settings can help the transcriber process voice inputs more accurately. For example, you could: - Increase
onNumberSeconds
in the
transcriptionEndpointingPlan
- Set Deepgram's endpointing to 300ms - Adjust
waitSeconds
to 0.8 in the
startSpeakingPlan
### 4. Consider Alternative Transcribers If the default Vapi transcriber isn't working correctly, consider using a custom transcriber or switching providers (e.g., from Deepgram to Whisper or Azure). Would you be able to share which specific transcriber settings you're trying to modify? This would help me provide more targeted troubleshooting steps for your situation. Sources: - [Discord discussion about system glitches](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1336674769055252594) - [Discord discussion about model compatibility](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1341463939246198784) - [Discord discussion about API updates](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1331035264751112202) - [Discord discussion about endpointing settings](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1342583376707850250) - [Discord discussion about alternative transcribers](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1343612796717371393)
s
I have the same problem...
@Matteo I reported an outage... if you would report it too within an hour there will be an emergency triggered and we all can get help faster
p
I think it was already fixed just now
Or not, not it wasnt
You can try setting values, refreshing with ctrl+f5 few times, then trying setting and saving again. Or you can do it thorugh API
s
Hi Pigeon, thanks for your response. BTDT but nothing seems to help. The Assistants are all messed up, not responding, no adherence to the prompt etc. Any change made will not be saved since there was an update to vapi UI
It worked until i was changing the prompt ... since then its carnage
p
Yes, the frontend is completely broken, it sends old values when you press
Publish
You can try updating through api, either in your code or through https://api.vapi.ai/api#/Assistants/
s
Thanks Pigeon for your valuable support. It doesnt work via API either... another OP posted the same negative result via API
p
I mean I'm able to change transcriber settings, with few refreshes and few publish click, but Im staying on
en
, front never sends language in its payload with transcriber settings
s
just like my case
according to other posts it seems like its not only an issue with transcriber but any changes to assistants in general ... could you #1293690685781839933 too maybe? or do you think they are on it by now anyways?
p
If they are in US its very early morning still, I will report outage
The threshold of 3 unique reporters has been reached. We have triggered the emergency alert to our dev team.
s
great ... thanks Pigeon
s
We are aware about this issue and we are working on it.
s
Thanks @Sahil
i
Facing the Same issue
@Arvind @Sahil Still not fixed unfortunately... Are you on it? Any update? Thanks in advance
s
@Arvind your button is changing the icon too ... does it work with deepgram on your end?
@Arvind @Sahil if i change the provider to openai or google first it kinda saves the changes
if i then change again to deepgram it worked
i
im working with Deepgram..if i save, go to a different agent and then come back..the save is gone
yes what Sven said works
s
try change to google or openai ... save and then change back to deepgram... that worked for me ... kinda ... the agents responses are whacky still but the language remains changed
i
thanks Buddy
something definately broke in their code
s
yap ... agree
i
and I think it impacted all the agents that were created using duplicate agent
I remember seeing a toast message yesterday which asked me permission to sync with all agents..i ignored
so now whatever language is set on master agent is reflected on child agents
similarly one of the other agents which I created by duplicating the child agent..it's language remained same as that of the child agent
s
i recognized that the agents that were not changed since the update of the UI were not effected at all
i
maybe some clues dor @Arvind and @Sahil to fix things
s
hopefully
The responses are still unreliable
i
i havent tested mine..will see
s
good luck buddy
i
thanks..hopefully they'd continue debugging and not close the issue as resolved
s
thats what i'm thinking but lets hope for the best
a
Appreciate the extra info, taking a deeper look into this. Some transcriber configs were updated.
s
@Arvind Agents working reliably again ... thanks for your effort
p
Thank you for reaching out to us. We've received your message and will get back to you as soon as possible.
Hi, checking if this is resolved/solved for you?
i
yeah its working..but one thing i noticed yesterday, I'm not able to change the text for Voice mail and End call message..it reverts to default everytime I go to another agent and come back
p
Hey Matteo, could you record a Loom video where even after making the new changes, your existing changes are not getting published or reflected?
i
sure will do
p
a gentle reminder to continue this thread.
2 Views