Conversation · Oct 2026
The pause problem: turn-taking that waits for meaning
Silence is a poor signal of a finished thought. Cadence combines acoustics with semantics to decide when to speak.
Most voice agents decide it's their turn after a fixed silence threshold. Set it short and the agent interrupts people mid-thought; set it long and every exchange feels sluggish.
Cadence predicts turn completion from three streams at once: prosody (falling pitch, slowing rate), lexical completeness (does the sentence so far form a complete thought?) and conversational context (was a yes/no question just asked?).
The result is an agent that can answer quickly after 'Yes, Tuesday works' but waits patiently through 'I think the main reason was… um…'. For interviews and coaching, where people think out loud, this matters enormously.