Glasses-only sound awareness across eight curated concepts.
Hard-of-hearing users, headphone wearers, cyclists, and parents miss sounds that matter: alarms, sirens, baby cries, doorbells. Existing solutions tether to a phone or send audio to the cloud, and many cover the wearer’s central field of view, which is dangerous on the street or in the kitchen.
Solo product and interaction lead. UX research, alert anatomy, iconography system, on-device ML pipeline tuning.
Single-APK on Rokid Glasses. YAMNet runs on-device. Eight curated sound concepts in two severity tiers, encoded by surround shape: triangle for hazard, rounded-square for awareness. Primary alert top-left at 24/24 dp; persistent 6 dp dot at bottom-right; Tier-1 halo breathe at 1.2 Hz.
Sound awareness for the AR glasses, on their own. The glasses listen continuously, recognize eight categories of meaningful sound, and surface a small alert in the upper corner when one fires. No phone companion. Nothing leaves the device. Designed not to block the wearer’s view of the world.
Hard-of-hearing users, headphone wearers, parents, and cyclists miss sounds that matter. Existing solutions tether to a phone or send audio to the cloud. Many also cover the wearer’s central field of view, which is dangerous for a device worn while crossing streets or cooking.
Broad sound-awareness audience: Hard of Hearing, culturally Deaf, headphones-on, parents listening for a baby, cyclists listening for sirens. Not life-safety hardware. It is an awareness aid that gets the wearer attending to the right sense (smell, vision) faster than no glasses at all.
Eight curated concepts. Two severity tiers, encoded by surround shape: triangle for hazard, rounded-square for awareness. Primary alert anchored top-left at 24/24 dp. Persistent service-up indicator lives as a small 6 dp dot in the bottom-right. Severity conveyed through shape and motion, never color.
Rokid 480 x 640 per eye Micro-LED, monochrome green. Primary alerts anchor top-left at 24/24 dp with a 32 dp icon and 14 sp bold label. Tier-1 hazards add a 1.2 Hz halo breathe (48→56 dp); Tier-2 awareness slides in from above. The persistent service-up indicator is a 6 dp filled dot anchored bottom-right (12/32 dp inset).
On first launch, JustListen requires the wearer to read and acknowledge that the app is not life-safety or medical hardware. It is an awareness aid: faster than no glasses at all, not a guarantee. The wearer remains the one who decides what to do.
Almost the entire panel is empty. The only persistent element is a 6dp filled dot in the bottom-right corner (12dp from the right edge, 32dp from the bottom). Every other pixel of HUD chrome competes with the wearer's view of the actual world, and the actual world is what they're using their eyes for.
A doorbell fires. The icon takes the primary slot at top-left (24dp from the top, 24dp from the left) inside a rounded-square surround: the visual signature for awareness sounds the wearer probably wants to know about but that aren't safety-critical. The pill slides in from above and holds for 10 seconds before auto-dismissing.
A smoke alarm fires. Inside the primary slot at top-left, the icon sits inside a triangular outline: the universal visual signature for hazard. A halo ring expands from 48dp to 56dp at 1.2 Hz for three seconds to draw attention without strobing. The icon itself never blinks, so it stays legible throughout.
When a second concept fires while a higher-priority alert is already on screen, the higher-priority alert keeps the primary slot and the new one takes a compact secondary slot beside it at 160/32 dp: 16dp icon-only, dimmed to 70% opacity, no label, no halo. The visual asymmetry is deliberate; multi-icon equal-rank banners would crowd the field of view.
The full sound catalogue the system recognizes today. Hazard sounds (smoke, siren, gunshot, glass, alarm, baby crying) share a triangle outline. Awareness sounds (doorbell, dog) share a rounded-square outline. The icon inside the surround names which specific sound. Two shape rules and eight icons, designed to be readable through a monochrome AR display.