How Sound Masking Works: Principles and Applications

A Conversation Three Desks Away

Two people talk quietly at the far end of a silent open-plan office, and every word carries. Switch on the ventilation and the same conversation becomes a murmur — still audible, no longer followable. Nothing was removed from the room. Something was added, and the words stopped being words.

That is sound masking, and the distinction from noise cancellation matters because the mechanism is the opposite one. Cancellation removes sound energy through destructive interference. Masking adds a low-level, continuous, broadband sound that reduces how far speech stands above the background, which costs the listener the detail that makes speech intelligible. The goal is not silence and not loudness. It is a floor that is high enough to cover speech cues and featureless enough to be forgotten.

Intelligibility Is a Ratio, Not a Level

Speech is understood or not understood according to its level relative to the background, not its level in absolute terms. This is the signal-to-noise ratio, and the reason masking works at modest volumes is that the ratio can be attacked from either side. Raising the background by ten decibels does the same arithmetic as making the talker ten decibels quieter.

What makes the ratio tractable is that intelligibility does not decline gradually across its whole range. The relationship between the audible speech in each frequency band and the proportion of speech a listener can understand was established in the 1940s by French and Steinberg, whose band-importance weighting underlies the measures still used today. It is non-linear: over part of the range, a few decibels of change in the ratio produce a disproportionate change in how much gets through. That is the region a masking design aims for, and it is why a masking system does not need to approach the level of the speech it is masking.

How Much Floor Is Enough

The target is set by a metric rather than by preference. The Speech Intelligibility Index, standardised as ANSI/ASA S3.5, predicts the fraction of speech information available to a listener from band-by-band speech and noise levels; it is the modern successor to the older Articulation Index. For speech privacy, the usual criterion is an index at or below 0.20, which corresponds to catching the occasional isolated word and not the sense of a sentence.

Subjective studies at the National Research Council of Canada put the practical compromise for conventional open-plan offices at a background level of about 45 dBA together with that 0.20 criterion — high enough to deliver privacy, low enough that occupants rate the space as acceptably quiet. Those two constraints pull in opposite directions, which is the whole difficulty of the design: a floor loud enough to guarantee privacy everywhere is a floor people complain about.

Line chart with octave band centre frequencies from 125 Hz to 8 kHz on the horizontal axis and band level in decibels on the vertical axis. An upper curve shows speech at the talker, peaking near 500 hertz. Below it, two curves show the room background: a faint lower curve for a quiet room, and above it the masking floor, labelled in the legend as bands summing to 45 A-weighted decibels overall. A vertical arrow at 1 kilohertz runs from the quiet floor up to the masking floor, labelled as a rise of ten decibels in that band. 75 60 45 30 125 250 500 1k 2k 4k 8k Band level (dB) Octave band centre frequency (Hz) +10 dB in this band Speech at talker Masked floor bands summing to 45 dBA Quiet floor
The masking floor raised beneath a speech spectrum. The floor is drawn twice in one hue because it is one quantity at two settings. The axis is band level, so the sourced 45 dBA target — an overall level rather than a band level — is drawn as a set of band levels that A-weight and sum to it; the quiet floor below and the band shapes are illustrative. Speech is shown at the talker — a listener several desks away receives it much attenuated, which is why a 45 dBA floor suffices.

The figure shows why 45 dBA is sufficient despite sitting well below the level of a talker. The speech curve is the level near the person speaking; a listener several workstations away receives it considerably attenuated by distance, screens, and absorption, so the gap that masking has to close at the receiving desk is far smaller than the gap drawn here. Masking lifts the floor to around 45 dBA, and distance has already done the rest of the work.

Shaping the Spectrum

Flat white noise is a poor masker, and not because it fails to mask. It masks well and is uncomfortable. Equal energy per hertz puts substantial power above 5 kHz, where speech has little content for it to cover, so that energy contributes hiss rather than privacy. Pink noise is better balanced but places significant energy below 200 Hz, which adds rumble without covering speech either.

An effective masking spectrum concentrates energy where speech lives — roughly 200 Hz to 5 kHz — and rolls off gently outside it. A practical shaping chain is three bands: a low shelf cutting below 200 Hz, a broad lift of two to three decibels around 1 kHz, and a high shelf rolling off above 5 kHz. The result is quieter in total than flat noise at the same masking effectiveness, because none of the level is being spent where it cannot help. Commercial systems use the same principle with finer control, often per zone, since different areas of a building present different acoustic problems.

Why Absorption Alone Does Not Solve It

Open-plan offices concentrate several problems at once. Hard surfaces reflect efficiently. Low screens let sound pass over them. The absence of enclosed rooms means nothing contains a conversation at its source.

Acoustic treatment — ceiling tiles, wall panels, carpet — reduces reverberation and absorbs reflected energy, and it is worth doing. What it cannot touch is the direct path from talker to listener. A listener twenty feet away receives enough direct sound to follow a conversation in a room with excellent absorption throughout, because absorption acts on reflections and the direct sound never meets a surface. Masking addresses what absorption structurally cannot, which is why the two are complements rather than alternatives.

Where the Loudspeakers Go

Deployment follows from that. Masking loudspeakers are commonly installed above the ceiling tiles facing upward, so the sound reaches the room indirectly after reflecting off the structural deck. The diffusion is deliberate: a listener who can localise the masking source will notice it, and a noticeable masking system is a failed one. Commissioning means measuring the existing ambient floor at many positions, setting the masking above it by enough to take control of the background, then walking the space to find and flatten the hot spots and dead zones. Uniformity matters as much as level, because a gradient in the floor is audible as the occupant walks through it.

Other Rooms, Same Arithmetic

Libraries are the instructive case, because the intuition runs backwards. The traditional expectation is near-silence, but near-silence is what makes a dropped bag or a scraped chair startling. A floor in the high thirties to low forties dBA reduces the contrast between those events and the background, and the room reads as calmer than it did when it was quieter.

Financial and medical settings use masking for confidentiality where architecture cannot provide it, raising the floor locally around a desk so that account or patient details do not carry to the next person waiting. The level needs care in that application: it has to reduce intelligibility for the bystander without intruding on the conversation it is protecting.

The Same Arithmetic at Home

In a home the problem is usually a specific intrusion — a neighbour, a road, a compressor cycling on — rather than general privacy. The same mechanism applies at a smaller scale, with the advantage that one listener can adjust the level and spectrum directly instead of commissioning a system for an average occupant.

Measuring Whether It Worked

Privacy is measurable, which means a masking installation can be verified rather than argued about. The Speech Intelligibility Index is computed from one-third-octave band levels: measure the speech and the background in each band, take the ratio band by band, weight each band by its contribution to intelligibility, and sum. The result is a single number between 0 and 1.

Measuring before and after is what makes the number meaningful, since the index depends on the room as much as on the system. A before-and-after pair also distinguishes the two ways a masking system can disappoint — a floor that is too low to reach the criterion, and a floor that reaches it but is uneven enough that some desks sit well above the average while others are noticeably loud.

References

Frequently Asked Questions

Is sound masking the same as noise canceling?

No. Noise canceling uses destructive interference to reduce specific sounds, typically in headphones. Sound masking adds a low-level broadband sound to reduce the intelligibility of unwanted sounds without canceling them.

How loud is a typical sound masking system?

Most systems operate between 40 and 48 dBA, which is comparable to a quiet air conditioner. After a brief adjustment period, most occupants are not consciously aware of the masking sound.

Can sound masking make an office completely silent?

No. Sound masking does not eliminate sounds; it reduces their perceptibility by raising the background noise floor. Conversations may still be audible, but they become unintelligible at a distance.

What type of noise spectrum works best for masking speech?

A shaped spectrum that concentrates energy in the speech frequency range of 200 Hz to 5 kHz with gentle rolloff outside that range is more effective and comfortable than flat white or pink noise.

WhiteNoise.top

WhiteNoise.top builds free browser-based noise generators, mixers and breathing timers. Our articles explain the acoustics behind those tools and cite published research where it exists.