Generate music at the end of a conversation

Prompting
,
1

Hi all,

I got a service that is running on gpt-realtime and I’ve just had a strange behaviour.
Basically, my service is a Q&A and , at the end, there’s a NPS and if the user want to give a suggest.

In last call, after greetings, I can hear correctly a music, even if there’s no prompt that could generate it.

Is there anyone that had this problem?
From my side, I’ve added a guardrail “at the end, say a goodbye and close the call” but, due to I had that problem just once, I cannot replicate it

thanks

1 Like
2

Hi andrea

I reproduced this in the very beginning. Seems it is part of the data with which the model was trained. I heard a media file played that could not have come from us.

Once we developed a tool to end the call, and call the tool, it never happened again.

Looks like the same thing is happening to you. I blame this on a hallucination