Cameras raise alarm bells. Wearables get left on the nightstand by the second week. Pressure mats and door contacts tell you where someone is, but not what they're doing. Sound, it turns out, tells you almost everything — and nobody minds it being there.
Ask an occupational therapist what they actually need to know about a patient's week at home, and the list is short: did they get up and move around, did they make it to the bathroom on their own, did they cook, did they take longer than usual to do any of it. None of that requires a picture. Almost all of it leaves an acoustic trace.
The privacy problem nobody solved
Every serious attempt at continuous home monitoring runs into the same wall: older adults, quite reasonably, do not want a camera watching their bathroom. Family members feel the same way on their behalf. Wearables fare a little better on privacy but badly on adherence — devices get taken off, forgotten on a dresser, or simply never charged. The result is a monitoring gap precisely where the richest information about daily function actually lives: the home.
Sound sensing sidesteps both problems. A small microphone-based sensor mounted in a room can distinguish a running tap from a flushing toilet, footsteps from a fall, a cabinet door from a front door, without ever recording or transmitting anything that resembles a conversation or an image. Nothing to wear, nothing to charge, nothing pointed at anyone.
What a sound actually tells you
The acoustic signature of daily life is more structured than it first appears. A shower running for four minutes followed by a hairdryer is a bathing routine. A tap running for eight seconds, six times in twenty minutes, with cabinet doors in between, is meal prep. Footsteps that stop abruptly, followed by silence, followed by a chair scrape, is very different from footsteps that stop abruptly followed by nothing at all for ninety minutes.
None of these patterns require understanding speech. They require pattern recognition on ambient sound — the same category of problem as detecting a smoke alarm or a doorbell, just tuned to the acoustic vocabulary of a home. That tuning is the hard engineering problem, and it's also exactly why sound sensing has lagged behind flashier approaches: it looks unglamorous next to a computer-vision demo, even though it solves the actual deployment problem.
The best sensor is the one that's still switched on six months later. For home rehab monitoring, that has turned out to be the one nobody notices.
Where it falls short, and why that's fine
Sound alone can't tell you if someone is limping, or whether a transfer from bed to wheelchair was steady or shaky. That's why acoustic sensing works best as the backbone of a layered system — paired with a pressure sensor to confirm sitting and standing, a floor-vibration sensor to catch fall signatures, and contact sensors on the fixtures that matter most. Sound tells you that something happened and roughly what; the other layers add texture. Trying to make one sensor type do everything is how most home-monitoring pilots overreach and quietly get switched off. Trying to make sound do the everyday-life heavy lifting, and asking the other sensors to confirm the moments that matter clinically, is a much more honest division of labour.
What this means for a therapist's caseload
The payoff isn't a live feed to stare at. It's a daily activity timeline that flags deviations — bathing thirty minutes later than a patient's own baseline, three days running, or a drop in kitchen activity that lines up with a medication change. That's the kind of signal a therapist can act on in a five-minute check-in, instead of discovering it three weeks later at the next scheduled visit, if at all.
The technology isn't the interesting part. What's interesting is that the least invasive sensor turned out to be the most informative one — and that the privacy constraint, which looked like the obstacle, ended up shaping a better system.