PRODUCTION SCENARIO
A kiosk records a short WAV question and sends it straight to a deployed audio-enabled chat model, with no separate transcription step.
How should the audio be included in the request?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: As an input_audio content part with base64 data and format
The audio quickstart base64-encodes the WAV and sends it as a content part whose type is input_audio, with the encoded data and the format inside. The audio field of the request configures the spoken output alongside the modalities list, so putting the recording there sets the wrong thing. There is no audio content part taking a mime type, and file IDs belong to the files and vector store flow that file search uses.