Could you please help us investigate an issue? Our bot configuration has "firstMessage" set up. While it says the message correctly once during the call, the transcription shows it duplicated. The call ID is 74b158fc-9bdc-44e9-80bf-9252c5fa5e3a. Sorry for submitting the issue several times, I dunno why, but the bot marked my previous posts resolved 🙂 @User

message has been deleted

@User could you please help me?

@tsyrya. Please ask support for more info on this.

@tsyrya. I'm glad to hear that you've found a workaround by setting the

modelOutputInMessagesEnabled

parameter to false. This parameter typically controls whether the bot's speech recognition model outputs are included in the message transcripts. When set to true, it might cause the transcription service to include additional recognition results, which could lead to duplicates if the system misinterprets the first message as being spoken twice. By setting

modelOutputInMessagesEnabled

to false, you're likely preventing the speech recognition model from adding its output to the messages, which can help avoid the duplication issue in the transcript. It's important to note that this setting should be used carefully, as it may also filter out other model outputs that could be useful for debugging or understanding user interactions. If you need a more detailed explanation or further assistance, please let me know, and I can look into it or direct you to additional support resources.

Could you please look at the call with id da8c398a-10ad-47f3-ab5b-948e53570a26. The first phrase in the transcription is

"Hi! Hello! How can I help you? Hi. Hello. How can I help you?".

you can see that it goes twice. But the bot actually says it once, why does it work it this way if modelOutputInMessagesEnabled is true?

Hi there @tsyrya. when I took a look into that specific call ID i see that the following in the call event:

"firstMessage": "Hi! Hello! How can I help you?",

As you can see if you set the parameter

modelOutputInMessagesEnabled

to

true

then it will use the model's output in the conversation history instead of the raw transcription. If you set that to

false

the assistant should behave in the way you are expecting, where the first line of transcription will be spoken verbatim. Here is a link to the docs that might help clarify the

modelOutputInMessagesEnabled

parameter: https://docs.vapi.ai/api-reference/assistants/create

@ram_vapi thanks for reply. According to the docs: "This determines whether the model’s output is used in conversation history rather than the transcription of assistant’s speech." If I understand correctly, the behavior depends on the value—whether it uses the provided firstMessage from the model or the raw transcription. Either way, it seems like duplication shouldn't occur. Am I misunderstanding something here?

@tsyrya. just checking if resolved for you?

@Shubham Bajaj nope, I still don’t understand why there was duplication in the transcription. Could you please clarify it based on call id I provided earlier? From my point of view, now it looks like a bug or misconfiguration. So I am trying to understand what the problem is.

@tsyrya. modelOutputInMessages works only for 11labs and the double transcription issue is on our side and we are working on fixing it. cc: @User

@Shubham Bajaj Thank you for looking into this. We are waiting for the fix then. Could you please share an ETA for it?

11labs is a TTS provider

@bryant yeah, but the duplication was in the transcription. There were no problems with the verbal part; the bot said exactly what we wanted. Sorry, but I'm confused about how the TTS service is involved here.

@tsyrya. can you try again now?