@Ben You have a good understanding of the process. In the case of VAPI, the transcriber is indeed an integral part of the speech-to-text functionality. However, if the real-time API you're referring to can provide a textual transcript along with the audio output, it might be possible to bypass the transcriber. Unfortunately, VAPI is designed to work with a transcriber to ensure the accuracy and consistency of the text output. Disabling the transcriber might not be supported as it is a core component of the service. The transcriber's role is to convert spoken words into written text, and it is optimized for the language it is set to transcribe. For dynamic language switching in transcriptions, this would typically require a transcriber that supports multiple languages or has the ability to detect and switch languages on the fly. If you're looking to reduce costs or have specific requirements for language switching, I would recommend reaching out to the support team for more detailed information on whether VAPI can accommodate your needs without a transcriber or if there's an alternative solution that would work for you. Please let me know if you would like me to connect you with support for more detailed assistance on this matter.

Ah this was my misunderstanding - the transcriber is optional but a seperate job https://platform.openai.com/docs/api-reference/realtime-server-events/conversation/item/input_audio_transcription So presumably utilising whisper in VAPI in transcriber would be equivelent to what you get via OpenAI API