Skip to content
Zumu
Sign inBook a demo
Book a demo

Answers when they’re done, not when they pause.

A dedicated audio model hears the end of a turn in the sound of the caller’s voice, not a silence timer.

Book a demoTalk to Zumu
Kestrel Auto · Front deskExample
  1. Caller“My reference is B-7…”
  2. PauseIt keeps listening; a silence timer would answer here.
  3. Caller“4-9.”
  4. Turn overHeard in the voice.
  5. Agent“Thanks. B-7-4-9: your car is—”
  6. Caller“Actually, can I get it tomorrow?”
  7. Gives wayThe caller cuts in; the agent stops.
  8. Agent“Tomorrow at four. You’re booked.”
BookedPickup · tomorrow, 4:00 PM

Response time

No dead air after the caller finishes.

<500 ms

From the caller’s last word to the agent’s first.

~300 ms

For the audio model to know the caller is done, from the sound itself.

Your dials

Set how patient each agent is.

Each agent keeps its own settings. Try a change in the playground, then test it on a share of your calls before every caller hears it.

Conversation flow

When the agent replies, and whether callers can cut in.

Front desk

Eagerness

Needs more certainty before it replies. Interrupts least, and feels a little slower.

Replies soon after the caller stops, and waits longer when it’s unsure. The balanced default.

Gives up waiting sooner when it’s unsure. The quickest to reply.

The agent stops talking as soon as the caller speaks.

The agent finishes its sentence, even if the caller talks over it.

Interruption sensitivity

Stops only for sustained speech. Ignores coughs and background noise.

The balanced default.

Stops at the first sound of the caller’s voice.

Applies when callers can interrupt.

Every caller gets to finish their sentence.

Talk to the live agent now and interrupt it, then set eagerness and interruptions for your own agent on your demo call.

Caller cuts in · agent gives way

Book a demoTalk to Zumu