There are VAPI delays between user speech stopped and Model request. Below are logs from a recent call 08:50:23:170 (+01:12:176) [CHECKPOINT] User speech stopped 08:50:24:152 (+01:13:158) [CHECKPOINT] Model request started The one second gap between speech stopped and model request makes the overall gap to be over 2 seconds which makes the whole experience unnatural. This happens multiple times in a single call. Is there something that can be done to improve this experience? Example Call ID 0d960b14-aa3d-428f-b02c-bb554c390de9

Sorry for the delays as we are currently experiencing an extremely high volume of support requests. The team will get back to you.

Hi, could you please let me know how long would it take to get to this.

Hey! I checked the logs for your call. The short delay you noticed between finishing your speech and the assistant’s response is due to endpointing latency. This is an intentional pause to make sure you’ve finished your turn before the model request is sent. - For longer utterances, the timeout can be \~1.5s. - For shorter ones, it’s usually just a few hundred ms. This behavior is expected and depends on the configured endpointing timeout.

Thanks Aditya. Isn't [CHECKPOINT] User speech stopped indicative of the endpoint detection. This is the timestamp when configured endpointing detected user speech as already stopped. The gap between endpoint checkpoint and model call should be 50-100ms as it is in most cases. It is occasionally during the same call that agent takes long pauses - that too at end of sentence endpoints. Issue is 100% reproducible in every call - happens at least 2-3 times

Thank you for providing this information. We’ll be looking into it further to better understand the situation and determine the next steps. Our team will update you as soon as we have more details, and we truly appreciate your patience and cooperation in the meantime.

Hi, any updates?

Thank you so much for your patience. We truly understand how important this matter is to you, and we want to assure you that we’ll be with you to provide an update as soon as possible. We appreciate you bearing with us while we continue looking into this.

It has been nearly a month since I reported the original issue. I understand this is discord based support but I am hoping you have some SLAs which are not in months.

We completely understand how frustrating it must be to have waited this long, and we’re truly sorry for the delay. Please know that our team is actively checking every nook and cranny to identify why this issue is happening. We deeply appreciate your patience, and we’ll share an update with you as soon as we have more details.

Thank you Kyle. I will try these out and report back the results in a couple of days.

Sounds good. We will standby in the meantime

Hi, apologies for delay in responding. I am grateful for your assistance. There is certainly an improvement in the response time. I am not entirely sure why. From logs I cited above, the gap appeared to be between endpoint confirmation and LLM call while configs were targeted at shortening endpoint identificaiton. I would love to study this in more detail and get a handle on it. Can you please point me to some resources that would be helpful here? Also, now I have a problem where assistant speaks too soon. This is more prominent in cases where user is spelling something, e.g. their name or address. What is the best way to handle that?

could you please provide call ids where it speaks too soon or gets the spelling incorrect?