Urgent - Assistant Speech Cut-off Issue
# support
s
Hi team @Vapi @Vapi Support Bot , I am facing an issue with the VAPI Web SDK where the assistant starts speaking a tool call response but abruptly stops mid-sentence. The full response is received, but only part of it is spoken. This is affecting user experience and trust in the assistant. You can refer to this specific case using Call ID:
7951cdd1-760e-4be9-bac7-77c1a868bb5c
, where the assistant stopped speaking without completing the sentence.
v
Try adjusting your API settings to prevent speech cut-offs during tool call responses:
Copy code
javascript
// Update these parameters in your VAPI configuration
const vapiConfig = {
  numWordsToInterruptAssistant: 5,  // Increase this value
  maxDurationSeconds: 10            // Increase this value
};
You can update these settings in your VAPI dashboard under call flows/scripts. This should prevent the assistant from being interrupted prematurely during tool call responses. Source: - [VAPI Web SDK assistant speech cut-off issue troubleshooting](https://discord.com/channels/1211482211119796234/1334027136305991750)
v
Message marked as helpful by @! AetherMind! 🎉
s
clearly that's not the issue because for this call
e0d8d002-de59-45fb-a815-5635d98aa24c
it somehow spoke the full sentence. This is very annoying from an user's perspective as this issues are repeatedly occurring, again saying such repeated issues are making us loose trust on vapi. Please look into this urgently
c
Hey! To help track down this issue, could you share: - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!
s
Call ID:
7951cdd1-760e-4be9-bac7-77c1a868bb5c
At
1:14
you can hear the assistant only speaks "Before I can proceed with your request further, I need some more clarification. Please provide the following details to proceed with your password reset:. 1. What is the reason for your password reset request?" and then stops. If you check the logs you can see the complete sentence 17:49:35:094 [LOG]
Voice input: Before I can proceed with your request further, I need some more clarification. Please provide the following details to proceed with your password reset:. 1. What is the reason for your password reset request? (Select one: 1) Password expired or expiry approaching, 2) Account locked, 3) Forgot password). 2. Which Active Directory platform would you like to use for the reset? (Select one: a) On-Prem Primary AD (IP: forty three point two zero four.one hundred twenty eight point one eight one), b) Azure AD) Could you please provide these details?
c
What is your expected result if the assistant is able to speak the full sentence? Is the user saying a string value like "Password expired" or are they using dial tones to choose an option? Also, please provide a screenshot or code snippet of the tool "process_user_request" so I can get a better understanding of how the tool is being used.
s
the user will say the reason for pass reset and the ad platfom For example : Password Reset and On-Prem Primary AD thats it
tools: [ { type: "function", messages: [ { type: "request-start", content: "Processing your request. Please hold on..." }, { type: "request-failed", content: "Something went wrong while processing your request. Please try again shortly." } ], function: { name: "process_user_request", description: "Handles the complete user request lifecycle including validation, planning, authorization, and execution. Based on the current state, it will guide the next step.", parameters: { type: "object", properties: {} } }, async: false, server: { url:
${baseURL}/${tenantBotId}/${user._id}/processRequest
, timeoutSeconds: 120 } } ]
c
I see there is a couple of issues: 1. If you filled out the message for request-complete, it will override what is spoken in the toolCallResult. I suggest clearing the message from request-complete. 2. The toolCallResult needs to be one line in order to get consistent results. Multi-line results will be unstable. Please follow this documentation and let us know the results of the changes.
s
In the doc its mentioned
result (Y): This field holds the actual output or result of your tool’s execution. The format and content of “result” will vary depending on the specific function of your tool. It could be a string, a number, an object, an array, or any other data structure that is relevant to the tool’s purpose.
in my case this will be object and i don't want that object to be spoken by the agent that's why i am sending a formatted message to be spoken by assistant on request-complete. So are you saying this approach would not work and i should remove the request-complete message completely ?
c
Yes remove it and you will still be able to format it elsewhere
3 Views