How it works
Most of what it tells you is not a sentence.
A conversation gives you no room for a paragraph. So it earns every word it spends.
The window
A conversation leaves about a fifth of a second.
Across ten languages, from major world languages to small indigenous communities, people begin their reply within about 200 milliseconds of the other person finishing. Stivers et al., “Universals and cultural variation in turn-taking in conversation”, PNAS, 2009
That is the whole space a device has to work in without talking over someone. So the fastest thing Blynq says is not a word at all.
Three tiers
A tone, a word, or a full answer.
- Tier 0
under 200ms - No words at all.
Two rising notes mean someone joined the room. Learned in ten minutes, then free forever. This is most of what you hear.
- Tier 1
in the gaps - One to four words.
“Someone on your left.” Spoken only into detected silence. It will not speak while a human is speaking, and it will drop the cue rather than interrupt.
- Tier 2
on request - A full answer, and only if you asked.
One press on the temple. Nothing is ever spoken at you unprompted.
Try it
Watch where it speaks.
A short meeting. Press play and follow the playhead: cues land only in the silences, one is dropped rather than interrupt, and the full answer waits for its turn. Turn sound on to hear the cues.
In the room
Four people around a table. Press play.
In your ear
Nothing yet.
Illustrative conversation, with time compressed.
Measured
Taken off the device.
Every figure here is measured on the glasses, not estimated. Each one sits under a tier above it.
- A frame across the link
- 868msFrom the camera on the glasses to the phone, over the radio.
- Press to spoken answer
- 8.9sOne press on the temple to a full tier 2 answer in your ear.
- A spoken command, decided on the glasses' phone
- 1msWhere a spoken command goes is decided on the phone itself.
Measured on the device. The radio used to be the slow part of this; it is now under a second of it.