mark & clear

Audio you send is queued and paced out to the caller in real time, so “sent” and “heard” can be seconds apart. Two frames manage that gap.

mark — playback checkpoints

Direction: both — you set a checkpoint; Dial echoes it when playback reaches it.

Send mark after a batch of media frames; when everything queued before it has finished playing to the caller, Dial sends the same frame back. Use it to know when the agent finished saying something (e.g. to start a response timer, or to hang up after a goodbye).

1
{ "type": "mark", "name": "goodbye-done" }

A clear flushes pending audio and its unreached marks — flushed marks are never echoed.

clear — barge-in

Direction: your server → Dial. Immediately stops playback and discards everything queued. Send it the moment your stack detects the caller speaking over the agent, then start streaming the new response.

1
{ "type": "clear" }