Tap, let them vocalize, tap again. Up to 3 seconds.
Hands-free: Hum picks out each sound by itself and watches for several upset sounds in a row.
Or play a real recording
From sessions the model never heard. The caregiver's label stays hidden until Hum answers.
Every sound means something.
Hum learns one person's sounds from the people who know them best, then helps everyone else understand. Pick a voice, then listen.
Today
Sounds just heard
What did it mean?
One tap teaches Hum. It learns on this device straight away.
Sound is analysed on this device and then discarded. Nothing is uploaded. After your first visit Hum also works with no internet connection.
One person's sounds, laid out by how alike they are
Each dot is one vocalization. Dots that sit together sound alike to Hum. Colour is the meaning the caregiver gave. Where colours separate, Hum can tell meanings apart; where they mix, it will say it is not sure.
Communication passport
A page to hand to a new teacher, nurse, or sitter. Add what helps in your own words, then print it.
Hand this voice to another device
Save the voice as a small file and send it to the new caregiver. They open Hum, load the file, and it understands this person straight away. The file holds numbers that describe the sounds, not recordings.
How well it works, measured honestly
Scores below come from recording sessions the model never heard. Clips from one session share a room, a microphone, and a mood, so mixing them across train and test flatters any model. We show that inflated number too, so you can see the difference.