Stream raw audio as S16LE with an explicit sample rate, using the content_type query parameter. Set the language explicitly too, since async jobs default to English.

Context: Third-party provider research (cmdspace STT report, citing Rev AI official docs): streaming requests set the content_type query parameter to audio/x-raw or audio/x-flac, and raw audio must be S16LE with a sample rate included. Send anything else and Rev AI cant parse your frames. Also there is no sync one-shot transcription endpoint, and async jobs default to English if you omit the language, so set it explicitly.