How Adaptive Noise works

Adaptive Noise generates a selected color, then can make cut-only changes from a short local microphone capture. It is a masking heuristic, not cancellation, a sound-level meter, or a treatment.

Four spectra

Color names describe power change per octave. White is 0 dB per octave. Pink falls about 3 dB per octave. Brown falls about 6 dB per octave. Blue rises about 3 dB per octave. They are four different spectral balances, not a ranking of what a person should prefer.

Those slopes are defined over a stated bandwidth and level. They do not imply that brown is better for sleep, that blue is better for focus, or that any color is right for every room. The player starts with a shared target across colors, then leaves the volume decision with you.

Local capture and cut-only shaping

The browser listens briefly while its own playback is muted. It estimates energy across a ten-band bank centered at 63, 125, 250, 500, 1,000, 2,000, 4,000, 8,000, 12,000, and 16,000 Hz. It then reduces, never boosts, added noise in relatively loud ambient bands. That keeps the system from chasing a room tone by making total loudness creep upward.

Consider a compressor hum at 120 Hz with harmonics at 240 and 360 Hz. The 125 Hz band will respond most directly to the fundamental. The 250 Hz band can respond to the first harmonic, while the 360 Hz component sits between the 250 and 500 Hz regions. Shaping cuts added energy in those areas. It does not notch the compressor out of the room. It leaves more of the selected noise elsewhere, so the hum has more company in nearby auditory filters. Fletcher's masking work and Fastl and Zwicker's account of critical bands describe that mechanism: nearby sound can raise audibility thresholds. They do not promise a result for every listener or source.

What Automatic mode does and does not do

Automatic mode rechecks on a cadence you choose: one, five, or fifteen minutes. At each check, it mutes playback to listen, then adjusts conservatively if a stable new profile is available. It is not continuous analysis, and it does not keep the microphone open between checks. A refrigerator cycle, a passing truck, or a changing fan can still make an earlier capture a poor fit.

What we measured

The shipped generators were checked at 44.1 and 48 kHz. Measured spectral slopes were within 0.05 dB per octave of target for the four colors, and their common level was matched within 0.01 dB across colors. True-peak headroom was retained before playback. These are digital-source measurements, not calibrated room SPL measurements and not proof of subjective masking.

Why steady hums can be hard to ignore

Low-frequency and tonal environmental sound can be annoying, and dB(A) alone does not describe that experience. Persson Waye and Rylander reported an association between long-term low-frequency noise exposure and annoyance or disturbed rest and concentration. WHO treats environmental noise as a public-health issue. Masking can change perception in the moment; it does not reduce exposure, repair a source, or establish a health outcome.

Limits

• The microphone is uncalibrated, and the speaker response is unknown.
• Browser and phone background audio behavior varies. iOS screen-lock behavior can interrupt playback.
• The courtesy capture quota is client-side and bypassable. It is not copy protection.
• Small speakers may not reproduce low-frequency vibration well.

Sleep, focus, and safety

Continuous-noise sleep findings are mixed, with very low certainty in a systematic review. No color is a treatment for ADHD, tinnitus, or misophonia. Keep playback at the lowest comfortable level, preferably through room speakers rather than in-ear playback overnight. Occupational and safe-listening guidance is not a target for this app because it cannot measure sound level at your ear.

Provenance

Yogev Wallach is a physicist and product executive with audio and signal-processing experience. The product is deliberately bounded by what a browser microphone and generated audio can honestly do.

Last reviewed: 2026-09-11.