Tinker Voice agents

by Kondasamy Jayaraman · drawn with Colorfigures by April Zhu

...

Ravi’s call, ready to run.

Chapter 3 of 7

Whose turn is it?

Streaming helped. Then Ravi from Marlow Builders calls, a man who pauses while he thinks, and the assistant starts answering before he has finished.

This chapter: How does the line decide the caller has finished?

You will learn

  • How a silence setting ends a turn, and why there is no perfect value
  • What a cut-off and dead air are, and how one trades against the other
  • Why a noisy line needs an energy threshold too
Words you’ll meet (4)
Turn
One person’s stretch of talking, before the other replies.
Voice activity detection (VAD)
Deciding, from the audio, when someone is speaking and when they have stopped.
Cut-off
Answering before the caller has finished: the line took a pause for the end of their turn.
Dead air
A silence that runs on too long after the caller has finished, before anything is said.

The picture on the left is live. Each step below asks you to do one thing in it; the story continues here as you go.

Step 1 of 4

A turn is one person’s stretch of talking. The line has to decide when the caller’s turn is over; deciding from the audio when someone is speaking is voice activity detection (VAD). Larkspur’s detector ends a turn after a set stretch of quiet: the silence setting. Ravi wants to change a delivery address, and he pauses while he thinks. Before you run it:

A new delivery address recorded OK

  1. Ravi: “I need to change the delivery address on my order. It's order ten fifty-one. Can it go to our new site on Mill Road?”

    recorded Speech-to-text wrote: “I need to change the delivery address on my order. It’s order 1051. Can it go to our new site on Mill Road?”

  2. Larkspur’s assistant: “Order ten fifty-one has already left us. I've asked the carrier to take it to Mill Road instead.”

Ravi pauses between thoughts. The line ends a turn after 300 ms of quiet. What happens?