Resume speech from confirmed utterance boundaries

When a voice agent has only whole-context TTS completion marks, it cannot infer the unheard word suffix of an interrupted context from planned text or audio chunk timing. For suffix resumption, split planned speech at short natural utterance boundaries, give each a distinct playback context, and advance only through marks acknowledged before transport clear. A late mark may acknowledge cleared audio. On an uncertain boundary, clarify or restart that short utterance; let a new caller question or correction supersede continuation. This trades extra synthesis and pause overhead for an auditable checkpoint. It does not prove handset audibility or enable exact word-level resumption.

Applies when: A Pipecat telephone agent needs conservative speech continuation without verified word alignment. Keywords: Pipecat Twilio TTS interruption resume suffix context mark clear utterance alignment