HearMate
Bone-conduction companion system
For people with hearing loss, the hard part of a conversation is rarely a single word — it is missing information, having to ask again, and the quiet pressure of slowing everyone down. HearMate pairs an open-ear bone-conduction headset with an AI watch that captions speech in real time, saves what matters on request, and lets the wearer tune their sound simply by saying how it feels. A companion app keeps the record; a doctor portal turns daily feedback into data an audiologist can act on.
Missed words, repeated questions, quiet pressure
Hearing loss is usually framed as a volume problem. In everyday conversation it shows up as three frictions — missing information, asking again, and the pressure of holding others up. Each one became a design requirement.
Sound is blocked before the cochlea while the inner ear still works. Bone conduction goes around the blockage — the clearest fit.
Helps with the conductive part; the benefit depends on how much the inner ear can still do.
Still relies on the cochlea and auditory nerve, so gain is limited. Here captions carry more of the load.
Design rule. An audiologist confirms suitability before fitting. HearMate is a communication and companion aid — a consumer bone-conduction headset is not a medical hearing aid.
- Follow family talk without asking people to repeat
- Catch prices, names and appointment times
- Nothing that looks like a “hearing aid”
- Noisy places blur every voice
- Forgets details given at the clinic
- Can’t explain which sounds are hard
- Keep up in fast group discussions
- Capture homework and deadlines
- Ears open in class and on the street
- Asking twice in front of classmates
- Canteen and corridor noise
- In-ear devices feel blocked and itchy
- Objective data between visits
- Know which situations are hard
- Set safe limits, not exact values
- Patients forget or soften feedback
- Settings drift with no record
- No channel for small questions
One conversation, four touchpoints
HearMate is designed as a service, not a gadget. The headset and watch handle the moment, the app keeps the record, and the doctor portal closes the loop between visits — with the wearer deciding what leaves the wrist.
People describe how sound feels, not which band to boost. Mate maps “muddy” to clarity and “harsh” to comfort — then asks.
→ voice tuningCaptions back up the ear instead of replacing it: only the current line is large, and the last line is one swipe away.
→ live captionsEvery prompt is also shown, and every action works by voice, touch or button — no step depends on hearing a beep.
→ multimodal inputNothing is stored unless the wearer asks. Summaries stay linked to the original words, so they can always be checked.
→ save & reviewOutput ceiling, small steps, one-tap mute. After a dropout, volume never returns above the last confirmed level.
→ safety railsAI drafts the trends; the audiologist decides. Chats and full transcripts reach the clinic only if the wearer shares them.
→ doctor portalDesigning for 1.83 inches
The watch is where the conversation happens. Everything on it must read at a glance, at arm’s length, mid-sentence — and nothing may depend on hearing a prompt. One idea per screen, two choices at most, and every change can be undone.
Live state, how many voices, time and battery — 11 px, quiet grey.
16 / 21 px at 42 % white: context that never competes.
Nunito ExtraBold 25 / 30 px, pure white, four lines at most.
Dotted orange underline when confidence is low — never silently guessed.
Names, numbers, times and places are boxed; one tap confirms and saves.
Gestures spelled out on screen — no hidden moves.
Swipe down for the line just missed. Names, numbers and times ask for one tap before they are trusted. When too many people talk at once, the watch says so instead of guessing.
3:30 pm?
The wearer describes the problem in their own words. Mate picks one of four intents, moves one 2 dB step inside the doctor’s range, then asks two questions before anything is kept.
clearer?
comfortable.
logged for Dr. Wang
The face stays calm. Mate checks in at quiet moments, and when something fails the screen explains it — sound never returns louder than the last confirmed level.
disconnected
Press the lower button
to unmute.
| Action | Say | Touch | Button |
|---|---|---|---|
| Last line | “What was that?” | swipe ↓ | — |
| Save | “Note this down.” | hold 1 s | — |
| Confirm info | “Yes.” | tap the box | — |
| Tune | “Too quiet.” “Muddy.” | — | upper + speak |
| Talk to Mate | “Mate, …” | tap Mate | upper |
| Mute | “Mute.” | — | lower |
Feedback is visual first; a short tone through the headset only confirms. Both side buttons are the board’s programmable PWR / BOOT keys under printed orange caps.
From everyday moments to a clinical record
The app is the wearer’s own record — notes, summaries and sound history, each traceable to its source. The portal shows the clinician only what care needs; anything personal sits behind a switch the wearer controls.
The watch is for the moment; the phone is for later. Every summary keeps a link to the words it came from, every sound change is logged with its before and after, and sharing is decided per data type.
Good evening,
Zhou
Notes
Clinic visit
• Next visit Thu 29 Oct, 3:30 2
• Call the clinic if dizzy 3
My sound
What Dr. Wang
can see
Zhou Jianguo
Summaries, transcripts, Mate chats: not shared
Audiogram 12 Oct · both ears
Fitting ranges gain dB · ceiling locked
Wear & comfort last 30 days
Where it is hard requests + swipes
Tuning requests 30 days · kept / reverted
Draft note AI · you decide
The wearer tunes daily, inside safe limits. The doctor moves the limits every few weeks — now with evidence instead of memory. AI only drafts; nothing changes in the fitting until the clinician accepts it.
From CAD model to wrist
A working watch built around the Waveshare ESP32-S3 1.83″ touch board: a 3D-printed case and back cover modelled around the board, a 300 mAh cell, two printed button caps and a woven loop band — paired with an off-the-shelf bone-conduction headset.
1.83″ capacitive touch LCD, part of the board module
ESP32-S3 board on four brass standoffs
300 mAh Li-Po, 502035, 3.7 V
3D-printed · orange, over PWR / BOOT
3D-printed · textured, lugs built in
3D-printed · held by four screws
01
02
03
04
05
06
07
08| Display | 1.83″ capacitive touch LCD, 240 × 284, 65K colour |
| Processor | ESP32-S3R8, dual-core 240 MHz, 8 MB PSRAM, 16 MB flash |
| Audio | 2 MEMS mics → ES7210 with echo cancellation; ES8311 codec |
| Sensors | QMI8658 6-axis IMU, PCF85063 RTC |
| Radio | Wi-Fi 2.4 GHz, Bluetooth 5 LE |
| Power | 300 mAh Li-Po (502035), AXP2101 PMU, USB-C |
| Structure | 3D-printed case, back cover and button caps; 4 screws |
| Headset | Off-the-shelf bone-conduction, Bluetooth |