using rime-ai Mist model Tried adding pauses using and and neither of them worked to add time in my prompt. So I tried instructions on this video: but then I got an error: { "message": [ "voice.property enableSsmlParsing should not exist" ], "error": "Bad Request", "statusCode": 400 } chatgpt said: Why this happens: Some older tutorials reference enableSsmlParsing which used to be a valid property, but Vapi removed support for it in a more recent version of their API. Vapi now handles most pause/intonation logic internally or via your assistant’s system prompt + voice provider features, not with SSML flags directly. 1. So, i can't figure out how to add pauses. can you tell me? and is that the same for all llm's and for all voices providers? if not, how do i figure out how to add pauses on those? 2. Why does Vapi make this so hard to figure out?

Hi there! 👋 I see you're having trouble adding pauses in Vapi with Rime.ai and the Mist model. I understand how frustrating this can be! Unfortunately, I don't have complete information about whether pause methods are exactly the same across all LLMs and voice providers. However, I can definitely help with how to add pauses in Vapi and explain why it might seem difficult. ## How to Add Pauses in Vapi There are several methods you can use: ### 1. Using Ellipses in System Prompt According to the [Vapi prompting guide](https://docs.vapi.ai/prompting-guide), you can use ellipses ("…") or a series of periods to indicate a pause in your system prompt:

"I… I don't know how to say this"

### 2. Adjusting Timing Parameters You can control pauses by adjusting these parameters: - `numWordsToInterruptAssistant`: Set this higher (3-5 words) to give more time before the AI continues - `maxDurationSeconds`: Define how long the AI waits for a response (5-10 seconds is common) ### 3. Using

assistantOverrides.maxDurationSeconds

Setting this parameter prevents the assistant from interrupting for a specific duration:

"assistantOverrides.maxDurationSeconds": 120

### 4. Custom Endpointing Rules These can instruct the pipeline to pause for user input for a specified duration. There's a helpful gist available at: https://gist.github.com/bajajcodes/cd664c81344ef9f27a4d304de43bd313 ### 5. Using

idleTimeoutSeconds

This parameter controls the delay before sending a message. ## Why Vapi Makes This Difficult There are several reasons why adding pauses in Vapi might seem challenging: 1. **API Changes**: Older tutorials reference

enableSsmlParsing

which is no longer supported, causing errors and confusion. 2. **Internal Handling**: Vapi now handles pause/intonation logic internally or via the assistant's system prompt and voice provider features, not with SSML flags directly.

So your messages should be in quotes "message yada yada" "message poopity poop" In code its:

"\"message yada yada\" <break> "\message poopity poop"\"

thanks for the tip. But can we input the pause length, such as ```` rather than just ```` ? I'm doing a meditation app and the voices are all going wayyy too fast

yes, of course, just escape quotes as well

do you know which voice providers allow for this? so far I've noticed that Vapi voices don't work with these breaks, while Azure voice do. Maybe some others have it enabled? I'm looking to use the cheapest ones. Also if you know which provider allows for the ``enableSsmlParsing`` property please that would be super useful. Thank you

This is 11labs thing I believe

ok thanks

But maybe more providers support it, Im not sure

checking if this is resolved/solved for you?

No, I’m unable to use the when I use Vapi. But when I test them within elevenlabs control panels the pauses work fine. Vapi always gives me an error: "voice.property enableSsmlParsing should not exist". I’m actually thinking to skip Vapi entirely now unfortunately, been trying to use pauses and delays for 3-4 days now

Hey Idris, could you share the call ID with me? I could take a look at what might have gone wrong for you. Most likely, you might have used some incorrect property name. Just wanted to be sure what you're using. Don't want to make any assumptions by the description, so please share the call ID.

Thank you for checking, I just made another test (unsuccessful with the enableSsmlParsing property) here is the call ID: ``cc28f87a-f85c-49a0-a6e0-0cdf33f2c9c6`` Now I manage to not have any error in the console when I test, but the voice assistant completely ignores the ```` even though when I test the assistant within the ElevenLabs app it does work well with the breaks.

Hey Jeff, you have enabled the SSML parsing but you haven't mentioned anywhere inside your prompt for your assistant to generate the SSML tags when generating its content or response and if it never generates those tags they will never get parsed to the 11Labs voice provider so it won't work. So you have to explicitly mention how to generate those tags and how to make use of it.

I've decided to just use an elevenlabs voice to make things easier since none of teh things were working on the other voice provider. So i'm using code like this in my assistant prompt. Hey there. This is Sarah from Radiant Glow Med Spa. We just received your info from Facebook, about our Botox special. Is now a good time to give you some more info? But chatgpt said i might need to go to postman and do all of this type of stuff. is that necessary? are they correct instructions, or do you have better ones? https://cdn.discordapp.com/attachments/1359667494146801694/1361510599888080946/image.png?ex=67ff050e&is=67fdb38e&hm=9d2926969f738cc25fa054ad7e0642c6c63939f7464ab74748fc8c8e97dfd530&

Watch this video

@Shubham Bajaj have you been able to find what was wrong for my call ID please?

sahil, if you look at that video, on the comments, i already have 2 comments from 6 days ago saying how that video didn't help. after that though, i actually did get the postman thing to work, but the AI still didn't pause even after doing all of those steps correcttly

no don't think you can with vapi, in the end I'm using 11labs voices without vapi and now the pauses work fine

well, if you can make LLM return responses with 11labs syntax - https://elevenlabs.io/docs/best-practices/prompting/controls

you could try enabling SSML processing to use <speak> or <break time>. if you try this, just make sure its enabled on the agent side too. let us know how ti goes