in the example prompt vapi creates with the default "Riley" assistant there are explicit instructions about pronunciation. Does this prompt somehow affect the STT provider? or did someone just use chatgpt to write a prompt for the example lol. I thought the system prompt really only affects the llm model generating the response. and is not seen by the STT model that actually produces the pronunciation. @Shubham Bajaj can you confirm which is accurate? @Vapi https://cdn.discordapp.com/attachments/1211484481848873020/1421344116402290698/Screenshot_2025-09-26_at_11.49.31_PM.png?ex=68d8b15c&is=68d75fdc&hm=cb3b7153dbbb422a986ba69c15caabfaa2b7f7c84e2b4864a4d9f3e10d2b98ce&