The new Vapi integration with Voiceflow looks promising but I ran into some issues: I have been trying out the new Vapi integration with Voiceflow to check and see if I can use my Voiceflow agent as a backend and the Vapi voice platform as a frontend. Basically, in my Voiceflow assistant I am using an API call to an AI assistant that I am running on a server to process the user’s input and return a valid response. However, with this new integration, somehow when I ask the assistant to mention all the products that a store has, I can see in the console in my server that the assistant outputs a response of about 10 products (which is correct) but somehow when the voice assistant is turning this text response into voice output, it only outputs the first and latter part of the response. So, for example, it will say something like: “Sure, we have a whole list of products. …… Product number 10 is a beach bag. Would you like to know more about any of these products?” So it will skip the 9 other products while still for some reason mentioning the last product. Has any one of you have had issues with this? Because when I use the exact same test in the Voiceflow chat window, then I do get a full response. Can this be because I am using markdown formatting and letting the assistant output responses in a JSON format? Hope someone can help me out with this 🤙

@Yannick. It sounds like you're encountering an issue where the voice output is truncating the list of products. This could be due to a variety of reasons, such as limitations on the length of speech output or issues with the way the response is being parsed and handled by Vapi. To address this, let's first ensure that the response from your AI assistant is being received in full by Vapi. If the response is in JSON format with markdown, it's possible that the parsing process might be affecting the output. We should verify that the response is correctly formatted and that Vapi is configured to handle markdown if that's required. If the response is correctly formatted and received in full, then we might need to look into the settings of the Vapi voice platform to see if there are any limitations or configurations that could be causing the truncation of the speech output. Since you mentioned that the full response is received correctly in the Voiceflow chat window, it's likely an issue specific to the voice output processing in Vapi. If you can provide more details about how the response is formatted and how you have configured Vapi to handle the response, I can assist you further. If you're unsure about the configurations or need more detailed support, please reach out to Vapi support for more information on this. They would be able to provide specific guidance based on your integration setup.

@Vapi well, what kind of format response would I need to send for the Vapi assistant to correctly process it into a voice text?

@Yannick. For Vapi to correctly process and convert text into voice, you should send a plain text response without any markdown or special formatting. JSON is typically used to structure data, but the actual text that you want Vapi to speak should be a plain string within the JSON payload. Here's an example of a JSON response that Vapi would process into voice:

json
{
  "response": "Sure, we have a whole list of products. Product number 1 is a sports shoe. Product number 2 is a jacket. Product number 3 is a hat. Product number 4 is a backpack. Product number 5 is a water bottle. Product number 6 is a pair of sunglasses. Product number 7 is a tent. Product number 8 is a sleeping bag. Product number 9 is a portable stove. Product number 10 is a beach bag. Would you like to know more about any of these products?"
}

In this example, the

response

key contains a plain text string that lists all the products. Vapi would then take this text and convert it into spoken words. If you are using markdown or other formatting within your response, it's important to strip that out and provide only the text you want spoken. If you need to send structured data alongside the text (for example, product IDs or details), you can include that in the JSON as well, but make sure the text meant for speech is plain and easily readable. If you continue to experience issues, it might be helpful to check the documentation for any character or length limitations in the speech synthesis and adjust your responses accordingly. If the problem persists, please reach out to Vapi support for further assistance.

@Yannick. share the call id with timestamp.