← digitallabo.be
luistervink
Idle
NL EN

Direction

position 0, 0 — say up / down / left / right

Number lock

say four digits to unlock

0
1
2
3
4
5
6
7
8
9

Switches

yes / no yes no
stop / go stop go

AI listens in.

Luistervink recognises a limited set of spoken English words through your microphone, live, and uses them to drive the panel next to it: numbers, directions, and switches. No typing, no camera — pure speech.

This is a small, specialised model: it only knows 18 words, in English, and nothing outside that. That's a deliberate choice — a small model that does one thing very fast and locally, rather than a large general-purpose language model that understands everything but is slow and heavy. Say one of these words, clearly and in English:

yesno updown leftright stopgo zero … nine

Runs entirely on this device. Nothing is recorded or sent anywhere.

not listening
model: speech-commands (18 words)
75%
Nothing recognised yet
Luistervink only recognises these 18 English commands and digits — no free speech, no Dutch. A large language model (like the one behind a chatbot) understands almost anything you say, but needs far more compute to do it. This model is the opposite: tiny and blazing fast, precisely because it only knows one narrow task.