/v1/realtime is served over WebSocket only. The upstream API also serves WebRTC and SIP
through POST /v1/realtime/calls — create a call with an SDP offer, then accept, reject,
hang up or refer it.
That matters twice over: WebRTC is the transport OpenAI recommends for browser clients,
because it handles jitter, packet loss and echo control that a WebSocket carrying raw PCM
does not; and SIP is the integration point for telephony, which is the largest commercial
use of a speech-to-speech API.
Feature: POST /v1/realtime/calls with the SDP offer/answer exchange and the call
control verbs, bridged onto the existing session machinery. The WebSocket transport stays
as it is.
Until then the limitation is documented, and browser clients are told how to connect over
WebSocket with the Sec-WebSocket-Protocol credential.
/v1/realtimeis served over WebSocket only. The upstream API also serves WebRTC and SIPthrough
POST /v1/realtime/calls— create a call with an SDP offer, then accept, reject,hang up or refer it.
That matters twice over: WebRTC is the transport OpenAI recommends for browser clients,
because it handles jitter, packet loss and echo control that a WebSocket carrying raw PCM
does not; and SIP is the integration point for telephony, which is the largest commercial
use of a speech-to-speech API.
Feature:
POST /v1/realtime/callswith the SDP offer/answer exchange and the callcontrol verbs, bridged onto the existing session machinery. The WebSocket transport stays
as it is.
Until then the limitation is documented, and browser clients are told how to connect over
WebSocket with the
Sec-WebSocket-Protocolcredential.