https://vapi.ai/ logo
Join Discord
Powered by
call message doesn't get sent to server url
# support
u
Call id: ddb18113-26a3-4994-87f5-b92107b36104 the last user message doesn't get sent to our server url, hence it's end up being silence-timeout Jul 18, 2024, 7:46 PM is it a bug to fix and how do we handle this in future
v
@8Ds It sounds like there might be an issue with the voice input not being captured or sent to your server URL, resulting in a silence timeout. To address this, we can take the following steps: 1. Verify if the 
voice-input
server message is enabled in your settings. This ensures that voice inputs are sent to your server. 2. Check the 
silenceTimeoutSeconds
setting to ensure it's configured correctly. If it's too short, it might be timing out before the user finishes speaking. 3. Review server logs to see if there's an issue with the message delivery or if the message is being received but not processed correctly. Would you like me to enable the 
voice-input
server message for you and check the 
silenceTimeoutSeconds
setting? If so, please provide the current 
silenceTimeoutSeconds
value or let me know if you'd like to adjust it.
You can continue this conversation by mentioning me in the message.
u
the voice-input won't help here, it's the custom-llm endpoint didn't get called when user said 
correct
at the end. You can see from the transcript, but this message was never sent to our backend for handle reply.
is this due to the fact that 
correct
is considered as an interruption and hence, not sent to the server url? If so, this is a common situation where user say confirmation words right at the end of the AI voice. Can you add a rule such that if the user interrupted close to the end of the AI speech, do send the words to the server url.
@User need more help, thanks
account: dingyu+vapi-prod@superpanel.io
s
🔵 23:50:25:231 Transcribed user speech but not confident enough to interrupt, message: Correct., confidence: 0.9572462, numWordsToInterruptAssistant: 4, words: 1
correct
wasn't consider as interruption because num of words to consider are 4 change it to 1 or 2.
u
Changing it to 1 would be too sensitive, since this user’s word was happening at the end of the assistant message, it’s a corner case. And it’s hard to handle 🤔
if you listen at the audio, the user said ‘correct’ right after the AI speech is done, from user’s perspective, it’s a new user message and should be pushed to our server. I wonder would it be better to disregard the interruption rule if the assistant message is about to end and send whatever user say at that moment to backend server to get the next assistant msg. Or alternatively, like humi.ai, would a websocket be more flexible here as we can actively push back messages as per diff conv status. 🤔
s
@8Ds great thought let's see what can be done.
can you please create a new issue for this for discussion.
3 Views
PreviousNext
Powered by