Classifying noise from signal: the patent logic behind DREM sleep intelligence

Masking is not the same as deciding

Most sleep audio products mask. They play rain, they seal the canal, they hope. DREM is built on a different claim: classify the soundscape, then apply a user-defined priority so noise is cancelled while desired sound — an alarm, a partner’s voice, a baby monitor — can pass. That dual-path idea is what the granted method patent is about.

US 11,509,995 B1 (filed 2021, granted 2022) covers AI-generated silence in earbuds with classification, dual signal paths, standalone operation and user priority. Nine issued claims. Expected term into the 2040s. Continuations are planned for self-training and sleep-state work as the prototype path advances toward a 2027 arrival.

Four steps in the overnight loop

1. Sense: on-device mics classify snore, traffic, speech, alarms and other classes in real time.
2. Prioritise: the wearer’s hierarchy decides what becomes silence and what remains sound.
3. Cancel: anti-phase paths act on noise while pass-through keeps the signals that matter.
4. Protect: the loop holds overnight; phone optional after setup.

Silence on purpose is a product. Silence by accident is a foam tip.

Why patent-first product development

We filed before we had a mass-market SKU because the moat is the method, not a colourway. Hardware targets — under 4.2 g side-sleeper geometry, ~10 h runtime, wing-fit shells — exist to practise the patent, not the other way around. That is the same ownership logic as Mifo and Wildon: own the idea, then tool the object.

Licensing enquiries are welcome. Commodity earbud clones of a granted method are not. The article is a map of the claim set for partners, investors and engineers who want to know what “acoustic intelligence” means when it is written into the USPTO record.