How it works

The technology behind

a smarter alert.

SoundSense is four technologies working together: a low-power listener, a classifier that learns your specific sounds, a context engine that understands where you are, and a privacy model that keeps everything on your phone.

Pillar 01

Always listening,

without killing your battery.

SoundSense uses a layered detection pipeline. A lightweight energy and frequency monitor runs continuously, using a fraction of the power a full classifier would. When the monitor notices something potentially meaningful, like a sharp onset, a characteristic frequency band, or a pattern that doesn't match ambient noise, it wakes the full classifier for a closer look.

The heavy neural network only runs when it needs to. That's the difference between a battery drain of 3 to 7 percent per hour of active listening and the 20 to 30 percent you'd get from naive always-on classification.

iOS handles the background audio session, and the app respects every system power hint. If your phone is in low-power mode, SoundSense adapts without silently degrading its detection.

Layered detection
1. Energy monitor  ● always on
2. Frequency match  ● on trigger
3. Full classifier  ● on match
4. Context engine  ● on classify
5. Alert delivery  ● on decision
Pillar 02

Learns your specific sounds,

in three samples.

Your doorbell doesn't sound like everyone's doorbell. Your smoke detector has a specific chirp. Your microwave beeps in a specific cadence. Generic sound detection treats them all as approximations of a category. SoundSense treats each as its own fingerprint.

When you add a custom sound, the app records three short samples, extracts a compact acoustic fingerprint, and stores it locally. The next time your phone hears that fingerprint, it knows. No retraining, no cloud dependency, no delay.

The base app ships with 11 preset sounds that work out of the box. Custom training is where the accuracy comes from, because your apartment isn't a lab.

Add a sound
Sample 1 of 3
Recording your doorbell…
Sample 2 of 3
Tap and play sound again
Sample 3 of 3
Final confirmation
Pillar 03

Alerts that match

the moment, not the category.

Home · 3:00 AM · Bedroom

Smoke alarm

Full-screen, strongest haptic, watch vibration, flashlight strobe
Home · 2:00 PM · Kitchen

Microwave beep

Gentle haptic, logged to timeline
Office · 10:00 AM

Fire alarm test

Critical, because fire alarm always matters

Every sound gets an urgency score. The score is a function of what the sound is, what time it is, where you are, and what you've told the app matters to you. The more you use it, the better the scoring gets.

Pillar 04

Private by design,

not by policy.

Most privacy promises are policies. Ours is architecture. SoundSense doesn't send your audio to the cloud because the app isn't capable of sending your audio to the cloud. There is no server. There is no recording. There is no upload.

The machine learning model ships bundled with the app. Your phone does the listening, the classification, and the discarding. In a single operation. The raw audio is never written to disk. Not even in crash reports.

If you turn on cloud sync, we sync the names of your sounds (like "Front doorbell" or "Upstairs smoke detector") and their settings, so you can restore on a new phone. We do not sync the fingerprints or the audio. Ever.

Read the full security page →

What your phone does
✓ Capture audio buffer
✓ Run classifier locally
✓ Match against your fingerprints
✓ Calculate urgency score
✓ Deliver alert
✓ Discard buffer
No recording to disk
No cloud upload
No crash-report audio
See it in action

Hear what matters.

Starting today.

Free to download. No account required.