Speech Streaming API

stream_format=sseで音声イベント、audioでバイナリストリームを受信します。

POST /v1/audio/speech

KITSCH_OPENAI_BASE_URLには末尾が/v1のAudio API URLを設定します。

curl "$KITSCH_OPENAI_BASE_URL/audio/speech" \
  -H "Authorization: Bearer $KITSCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kitsch-voice","voice":"<voice_id>","input":"Hello.","response_format":"wav","stream_format":"sse"}'

応答

Content-Type: text/event-stream.

event: speech.audio.delta
data: {"type":"speech.audio.delta","audio":"<BASE64_AUDIO>"}

event: speech.audio.done
data: {"type":"speech.audio.done"}

speech.audio.deltaのbase64 audioを順番にデコードして連結します。speech.audio.doneが転送終了を示します。