Like what was the raw output of the LLM and what was passed to the transcription model?

@Chris The raw output of the LLM (Language Learning Model) would be the text that it generates based on the input it receives. This text is then passed to the transcription model, which in the context of voice APIs, typically refers to the Text-to-Speech (TTS) system. The TTS system, such as PlayHT with the voice of Jennifer, would take the LLM-generated text and convert it into spoken words, creating an audio representation of the text. If you need to know the exact text that was passed from the LLM to the TTS system, please provide the specific input or context, and I can assist you further.

How would I get this information for a specific call id, specifically 51b4b2d1-7803-4860-a607-e320e1a3736a

@Chris Please ask support for more info on this.

@Chris I will get back to you on this.

Posting something from our internal slack thread debugging this issue yes, it's because of this in the call instructions "After you respond to one of their responses pause before going on to the next question. so the issue is that if you tell the LLM to "pause", It doesn't know how to do it in reality (with respect to audio, since it just outputs text), therefore it wrote "pause as if you are processing the payment" to avoid this, I would put in the instructions that @Chris Alme linked for 11labs to introduce a voice audio pause, something like After you respond to one of their responses pause before going on to the next question, by using the special syntax "" exactly Curious if that approach would work, or if you're escaping stuff under the hood so 11 labs would get the instruction. I know we can also tune this with some of start speaking parameters, which I assume is doing something similar to the 11 labs tags under the hood

call id please related to this issue.