Psychoacoustic Masking: The Quiet Sounds Your Ears Never Register

A loud sound can make a quiet one disappear

Play a loud 1000 Hz tone and a much quieter 1100 Hz tone at the same time, and the quiet tone can vanish from perception entirely, even though in isolation it would be perfectly audible. This is psychoacoustic masking: a strong signal at one frequency raises the hearing threshold for nearby frequencies, hiding anything weaker underneath a masking curve that spreads out from the loud tone, wider on the high-frequency side than the low-frequency side.

Where the masking happens: the cochlea

The mechanism starts in the basilar membrane inside the cochlea, which performs a rough mechanical frequency analysis of incoming sound — different positions along its length resonate most strongly to different frequencies, tuned high frequencies near the cochlea's entrance (the base) and low frequencies near the far end (the apex). A loud tone drives a broad region of this membrane hard, and a nearby weaker frequency, whose own resonant region overlaps that same driven patch, gets physically swamped there before the two signals are even separated into distinct neural channels. The basilar membrane's tuning is asymmetric — each resonant region responds more broadly to frequencies above it than below — which is exactly why masking curves lean toward the higher-frequency side of the masker.

Simultaneous, forward and backward masking

Simultaneous masking is the case above: masker and masked tone present at the same time. Forward masking is stranger: a loud tone can still raise the threshold for a second, quieter tone played up to roughly 100-200 milliseconds after it stops, because the auditory system needs time to recover from the strong stimulation. Backward masking is stranger still: a loud tone can mask an even quieter one played just a few milliseconds before it starts, attributed to the slower neural processing time of a weak signal versus a strong one — the weak signal's neural representation is still forming when the loud tone's stronger, faster signal arrives and overtakes it.

The engineering payoff: lossy audio compression

MP3, AAC and every other perceptual audio codec exploit masking directly: a psychoacoustic model estimates, frame by frame, which frequency components would be masked by louder nearby components and therefore inaudible to begin with, and the encoder allocates fewer bits (or none at all) to those components while spending its limited bit budget on the parts of the spectrum a listener could actually notice. This is precisely how these formats achieve roughly 10x compression over uncompressed audio while sounding nearly identical on typical playback — they are not compressing the waveform faithfully, they are discarding exactly the information the masking model predicts you cannot hear.

encoder per frame:
  1. estimate masking thresholds from the loud components present
  2. compare each frequency component's level against its local threshold
  3. quantize components below threshold coarsely or drop them
  4. spend the saved bits on components above threshold

Frequently asked questions

Try it live

Everything above runs in your browser — open Psychoacoustic Masking and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.