other-emerald•17mo ago
Stumbling over words and saying the same sentence twice
Call ID: 9d7b70ae-4333-487d-b4b9-ec026ec02e0b
Timestamp: 00:32
The agent keeps stumbling on its words and/or saying the same sentence twice. I am using GPT 4o as the LLM
29 Replies
Hi @Saien | Exousia AI, I'm an AI assistant for Vapi. I'll analyze available sources and get back to you shortly.
-----------------------------
While I generate an answer, here are some sources that might be helpful:
* Launch Screen Assets
* Orchestration Models
* Glossary
* Web Calling
-----------------------------
Answer
The issue you're experiencing with the agent stumbling over words or repeating sentences may be related to the
responseDelaySeconds and llmRequestDelaySeconds settings in your assistant's configuration. These settings determine the timing of how quickly Vapi responds after user input and how long it waits after detecting a possible end of user speech before making a request to the LLM.
You might consider adjusting these settings to see if it helps resolve the issue. Here's how the parameters work:
- responseDelaySeconds: This sets the minimum number of seconds to wait after user speech before the assistant begins speaking.
- llmRequestDelaySeconds: This sets the minimum number of seconds to wait after punctuation before sending a request to the LLM.
If these delays are set too low, it might cause the assistant to respond prematurely or interpret quick, consecutive user inputs as separate commands, leading to repetition or overlapping speech.
Here's an example of how you might adjust these settings:
This adjustment can be made via the API or the Vapi dashboard under your assistant's configuration settings (1) .
If you continue to experience issues, it could be beneficial to look into the transcriber settings and ensure that it is correctly interpreting user speech, as errors here could also lead to repeated or garbled outputs. Make sure your transcriber is accurately configured with an appropriate model and settings for your use case.Relevant threads
These are relevant threads you might take a look into:
<#1247531250323816459>
<#1239972817345446028>
<#1236883124907151370>
<#1245005815014756392>
@Saien | Exousia AI
I'm sorry to hear that you're experiencing issues with the agent's speech. To address this, we might need to look into the specific configurations or settings you're using with GPT-4o. Could you please provide more details about how you have integrated GPT-4o with VAPI? This information will help me assist you more effectively.
You can continue this conversation by mentioning me in the message.
other-emeraldOP•17mo ago
I tried to increase the LLM request delay but that didn't help. Also I had to do it via the PATCH request since trying to update the setting on the dashboard is really buggy - it toggles straight to 0 then I can only update it in increments of 1 second.
@Vapi the 'ask for help' button isn't working
@Saien | Exousia AI
I'm sorry to hear that you're experiencing further issues with the dashboard and the 'Ask for Help' button. It seems like there might be a technical glitch affecting the dashboard's functionality. I would recommend reporting this issue to the technical support team so they can investigate and resolve it.
In the meantime, if you have any other questions or need further assistance with the VAPI, please let me know, and I'll do my best to help you out. If your issue with the agent's speech persists, we can explore other potential solutions or workarounds.
other-emeraldOP•17mo ago
@Ask Inkeep this 'ask for help' button didn't work so I'm tagging like this.
Thanks for the follow-up, @Saien | Exousia AI. I'll get back to you soon.
-----------------------------
While I generate an answer, here are some sources that might be helpful:
* Prompting Guide
* Inbound Support Example
* Voice Widget Example
* Call Forwarding
-----------------------------
Answer
It seems like there might be an issue with the UI component on the platform you're using. Unfortunately, without specific details about how Vapi's UI elements like the 'ask for help' button are integrated and function within your environment, I can't provide a tailored solution.
If you're encountering UI issues, it could be related to how the widget or assistant is embedded or how event listeners are set up for button interactions within your system. You might need to review the integration code and ensure that all elements are correctly configured and that the JavaScript managing these interactions is functioning as expected (1) .
For further assistance, you might want to inspect the browser's console for any error messages or warnings that could indicate what might be going wrong when the button is clicked. If the issue persists, checking the implementation against the Vapi documentation could help identify any discrepancies (2) .
@Saien | Exousia AI
I understand that you're experiencing difficulties with the 'Ask for Help' button and updating settings on the dashboard. It's important that these issues are addressed promptly. I will report this to our technical support team for you.
For the issue with the agent's speech, since adjusting the
llmRequestDelaySeconds did not resolve the problem, we may need to investigate further. It could be beneficial to check the logs for any errors or unusual activity that might indicate what's causing the agent to stumble or repeat sentences.
In the meantime, if there's anything else I can assist you with or if you have any more details that could help us troubleshoot the issue, please let me know.Got it. Tagging @Vapi Team for additional help.
deep-jade•17mo ago
We are working on it.
Glad I could be helpful. Feel free to create a new thread with any new questions you may have.
other-emeraldOP•17mo ago
Call ID: ff41f239-036d-471c-b115-aabb39adb733
Timestamp: 01:06
Definitely a major issue here
deep-jade•17mo ago
Please provide some time. We are looking into it.
other-emeraldOP•17mo ago
Just wanted to provide another example, didn’t mean to come across as rushing you guys in any way. I appreciate the help a lot
I noticed that this ticket says 'solved' but I'm still experiencing this issue
deep-jade•17mo ago
Yeah. I just marked it as solved because this issue was being faced by multiple users and we will notify in discord one this issue is fixed
other-emeraldOP•17mo ago
got it, thanks
Hi, which channel will you make the post regarding this?
deep-jade•17mo ago
I made in general that it will be fixed by monday. We identified the issue but we didn’t push the change yet to production cause this is a crucial part
other-emeraldOP•16mo ago
0aeb3126-bad9-4c3f-a08e-37cc68059033
00:43
I see that this issue hasn't been resolved yet. Is the update still yet to be pushed?
stormy-gold•16mo ago
hey happnes to me as well
other-emeraldOP•16mo ago
Please let me know when this issue will be resolved
deep-jade•16mo ago
Sure.
other-emeraldOP•16mo ago
a294b44c-2139-41c2-8588-4c637b747e96
01:37
I didn't experience this issue all day until now. The assistant tried to say two different sentences at once.
frail-apricot•16mo ago
sometimes this happen due to ass internet connection
@Saien | Exousia AI check your interenet
deep-jade•16mo ago
It is resolved can you try again? Also, if you face the similar issue then use different voice provider and give it a try
other-emeraldOP•16mo ago
It seems like it has been fixed but there’s only one situation where it usually stumbles over its words. And that’s during the times when a function is being called
I’ll send you an example in a bit
Here's the example:
d4ea1560-4469-4dd3-b122-df23b79ba448
01:55
Okay this time it just completely stumbled:
a535fab7-d804-4e43-b57f-fbf3288ce2fc
00:10
I think for this one, I should just remove the message the AI says during the function call via API (seems like it's only available for tools).
update: I think I got it to be fine with some updated prompting but I'll confirm this later
But for this one, I don't think there's much I can do
deep-jade•16mo ago
Will check that during our call.
So, for now marking it as complete.
other-emeraldOP•16mo ago
cool
Since we're only speaking on Tuesday, is there anything I can do in the meantime to resolve this?