----------------------------- **While I generate an answer, here are some sources that might be helpful**: * [Get Assistant]() * [List Assistants]() * [Create Assistant]() * [Update Assistant]() -----------------------------

### Answer It sounds like you might be encountering issues with overlapping voice responses due to the LLM being triggered multiple times. One way to mitigate this problem could be to adjust the settings for interruptibility and response delays in your assistant configuration. To handle interruptions more smoothly, you can set the

numWordsToInterruptAssistant

parameter, which controls how many words the user must say before the assistant stops speaking. Additionally, adjusting the

responseDelaySeconds

parameter, which controls the delay before the assistant responds after the user has finished speaking, might help prevent overlapping speeches. Here are the relevant settings you might consider adjusting: -

numWordsToInterruptAssistant

to define how many words trigger an interruption. -

responseDelaySeconds

to set a delay before the assistant responds, giving the user more time to finish speaking. These configurations can be adjusted in the assistant's settings through API calls or dashboard configurations. For more detailed guidance on managing these settings, you could refer to the sections on managing assistant behavior in the Vapi documentation [(1)]() , [(2)]() , [(3)]() , [(4)]() . If issues persist, reviewing the specific implementation details and the sequence in which the LLM is triggered might provide further insights into preventing simultaneous voice responses.

sus call id 54ade894-5a55-4786-88db-468c3bcff401

the first bit is fine. But that it has like three different audio responses on top on each other

i think its when it switches to Model request started (gpt-4o-2024-05-13, openai)

Actually no i dont think it is. i removed all fallback models and without switching to gpt-4o-2024-05-13 it still overlapped the speach. 300a611e-0938-4fed-bd79-04f89b9cb7e8

i have been playing around with the responseDelaySeconds llmRequestDelaySeconds as i think this is the issue? For this call 6c7e41d0-d03a-4919-8508-c1ab5d9d70cd I put the responseDelaySeconds to 0.05 llmRequestDelaySeconds": 0.15 It work working fine until the transciber output 'yes' like twice really fast. meaning the model request started twice. generting two outputs? making the speach overlapped?? https://cdn.discordapp.com/attachments/1242805073860624484/1243015824034369596/Screenshot_2024-05-23_at_11.38.45_AM.png?ex=664ff032&is=664e9eb2&hm=d4c677114bb4a0a345b65064f5a868f28abb64e69f25c6af25fa432fee55829c&

i removed the "assistantOverrides": { "variableValues": and it seems to be happening way less. I also increaesd llm and reponse delays

still an issue

hope that helps thos

i have removed knowledge base

and it happens even more less

as well

but defintely seen a change

25cc329e-7703-4763-bdee-15b494461cf2

45 seconds

in

and it just happened again

365c122c-a298-4fb4-a486-14cad2251c6b

please man

I think I have discovered the issue. Instead of using "numFastTurns": 1, always use "numFastTurns": 0 or do not include this parameter.

I don't see any issue with your call_id can you please provide me the timestamp as well describe little more about what is the issue that you are getting?