Speech Privacy in Open Offices Is Physically Impossible Without Sound Masking

Clients frequently ask for open office designs that achieve good speech privacy through acoustic treatment alone. They want high-NRC ceilings, full-height partitions between benches, and no sound masking — because sound masking feels like an add-on, an admission of defeat, something that can be cut from the value engineering list.

Every time this request comes in, the conversation needs to start in the same place: with the physics. Not because of professional obstinacy, but because the physics genuinely does not allow what the client is asking for. Speech privacy in an open office — Normal privacy, where conversations are not clearly intelligible to nearby workers — is physically impossible to achieve without raising the ambient noise floor. Acoustic treatment alone cannot do it. Here is why, with the calculations to prove it.

What Speech Privacy Actually Requires

Speech privacy is defined by the Articulation Index (AI), a number between 0 and 1 calculated per ANSI S3.5 (USA) or equivalent ISO 9921 procedures. The AI represents the fraction of speech information that reaches an unintended listener. ASTM E1130 establishes four privacy classes:

The AI at a given distance from a talker is determined by the sound pressure level of the speech signal minus the background noise level at each of 15 one-third octave bands from 200 Hz to 5000 Hz (the "speech frequencies"). Where the signal-to-noise ratio (SNR) is below −12 dB, that band contributes nothing to intelligibility. Where SNR is above +18 dB, the band contributes fully. The bands are weighted by their relative contribution to speech intelligibility.

The critical insight is this: AI is driven by the SNR, not by the absolute level of either signal or noise. A room where speech arrives at 45 dBA and background noise is 35 dBA has the same AI as a room where speech arrives at 55 dBA and background noise is 45 dBA — a 10 dB SNR in both cases. What distinguishes the second room is that the background noise has been raised by 10 dB, which reduces the AI to the same level as the first room while maintaining the same speech level.

This is the only lever available to achieve speech privacy: reduce the speech level at the listener, raise the background noise level, or both. Acoustic treatment addresses the first lever. Sound masking addresses the second. You cannot achieve Normal privacy with the first lever alone — the physics do not permit it.

The Calculation: Why Treatment Alone Fails

Let me work through a specific open office scenario. The room:

• Plan: 20 m × 15 m = 300 m²
• Ceiling height: 3.0 m
• Volume: 900 m³
• Occupancy: 30 workers at 10 m² per workstation
• Workstation layout: 60 cm partitions (bench-top screens)
• Activity: typical collaborative work, frequent phone calls

Current Condition: No Treatment

A talker at one workstation speaking at normal conversation level (60 dBA at 1 m). How loud does that voice arrive at a listener 4 m away?

Direct path (4 m):

SPL_direct = 60 − 20 × log₁₀(4) − 11 (hemispherical radiation)
           = 60 − 12 − 11
           = 37 dBA

Plus reverberant field. In a reflective room (painted concrete ceiling, glass partitions, carpet floor), the reverberant level can add significantly. Using the room equation:

SPL_reverb = PWL − 10 × log₁₀(R)

With bare room (ᾱ ≈ 0.08 at 1 kHz):

S = 2 × (300 + 300 + 900) = 900 m² [approximate]
R = 900 × 0.08 / 0.92 = 78 m²
SPL_reverb = 60 − 10 × log₁₀(78) = 60 − 19 = 41 dBA

Combined level at 4 m: approximately 43 dBA. Background noise: 32 dBA (untreated office with basic HVAC). SNR = +11 dB across the speech frequencies.

At SNR = +11 dB, the AI is approximately 0.45–0.55 in the mid-speech frequencies. This is firmly Poor privacy. Full sentences are clearly audible.

After Adding NRC 0.85 Acoustic Ceiling (100% coverage)

The acoustic ceiling dramatically changes the reverberant field. With ᾱ rising to 0.40 (from 0.08) at 1 kHz:

R = 900 × 0.40 / 0.60 = 600 m²
SPL_reverb = 60 − 10 × log₁₀(600) = 60 − 28 = 32 dBA

The direct field is unchanged at 37 dBA. Combined at 4 m: approximately 38 dBA. Background noise: still 32 dBA. SNR = +6 dB.

At SNR = +6 dB, AI is approximately 0.25–0.35 — Marginal privacy. Words are frequently understood. This is better, but it is not Normal privacy (AI < 0.20), and it is nowhere near Confidential.

After Adding 1.5 m High Partitions (NRC 0.65 surfaces)

Workstation partitions at 1.5 m height provide line-of-sight barrier attenuation for the direct path when the talker and listener are both seated (seated head height ≈ 1.2 m). The partition creates an acoustic shadow zone.

Barrier insertion loss (simplified Maekawa formula for a finite barrier):

IL = 10 × log₁₀(3 + 20N)

For a 1.5 m partition, seated talker and listener 4 m apart:

• δ ≈ 0.08 m at 1 kHz
• N = 2 × 0.08 / 0.34 = 0.47
• IL = 10 × log₁₀(3 + 20 × 0.47) = 10 × log₁₀(12.4) = 10.9 dB

This looks good. SNR = +1 dB gives AI ≈ 0.10–0.15 — Normal privacy territory. Problem solved?

No. Several conditions that were assumed in this calculation are false in practice:

The partition only provides attenuation in the direct path. A listener at an adjacent workstation who is 90° to the side rather than directly behind the partition sees no barrier attenuation at all — the sound travels over the top of the partition and reflects off the ceiling. In any open office, the number of "unshielded" positions (where the talker is not directly above the partition line) outnumber the "shielded" positions.

The calculation assumed 100% coverage of NRC 0.85 ceiling. In practice, lighting, sprinklers, ductwork diffusers, and HVAC equipment interrupt the ceiling treatment. Real coverage rates are typically 60–70% of the total ceiling area. At 65% coverage, the reverberant field contribution increases substantially.

The background noise assumption was generous. Modern open offices with well-attenuated HVAC systems and good acoustic design often measure 30–32 dBA ambient — lower than the 32 dBA used above. Many offices measured as part of the WELL v2 certification process measure 28–30 dBA. At 30 dBA background with speech arriving at 33 dBA, SNR = +3 dB and AI ≈ 0.20 — still only Marginal privacy, and at the very edge of Normal.

The calculation used 4 m. ASTM E1130 requires measurement at 2 m, 4 m, 6 m, and 8 m. At 2 m — which is the distance between adjacent face-to-face workstations in a 1400 mm bench layout — the direct path level is substantially higher and barrier attenuation is reduced because the angular geometry changes. At 2 m, even a well-treated room with partitions typically achieves only Marginal privacy.

The Maximum Achievable Without Sound Masking

The physical limits of acoustic treatment in an open office can be summarised as follows:

Background noise floor: Acoustic treatment cannot raise the background noise floor. It can only reduce the signal. If the background is 32 dBA and you treat the room to reduce reverberant speech by 8 dB (a substantial improvement), speech at 2 m distance still arrives at approximately 42 dBA — a 10 dB SNR. This gives AI ≈ 0.45 at 2 m. Poor privacy.

Reverberant field reduction has diminishing returns. The room constant R = S × ᾱ / (1 − ᾱ). When ᾱ goes from 0.08 to 0.40 (adding an excellent acoustic ceiling to a reflective room), R increases from 78 to 600 — a factor of 7.7, reducing the reverberant level by 8.9 dB. This is the majority of what acoustic treatment can deliver. Going from ᾱ = 0.40 to 0.70 (adding wall panels and furniture absorption) increases R from 600 to 2100 — a factor of 3.5, reducing the reverberant level by a further 5.5 dB. The second treatment investment delivers only 60% of the return of the first.

Distance decay. In a free field, sound level decreases by 6 dB per doubling of distance. In a reverberant field, it decreases by much less (or not at all beyond the critical distance). Acoustic treatment increases the critical distance — the point at which the reverberant field equals the direct field — but does not eliminate the reverberant field.

The critical distance formula: r_c = 0.14 × √R

With R = 600: r_c = 0.14 × √600 = 3.4 m. So beyond 3.4 m, the reverberant field dominates and additional distance provides no benefit. The adjacent workstation at 2 m is within the critical distance, meaning the reverberant field is already significant at that position.

The maximum AI achievable through acoustic treatment alone — without sound masking — in a typical open office is approximately 0.15–0.25, which is at the boundary between Normal and Marginal privacy. This assumes a best-case specification: NRC 0.90 ceiling at 75%+ coverage, 1.5–1.8 m partitions with NRC 0.65 surfaces, and the most favourable workstation geometry. Under normal design constraints, the treatment-only solution delivers Marginal privacy for the majority of workstation pairs.

Normal privacy — reliably, for all workstation positions including face-to-face and 90°-offset — requires raising the background noise floor to approximately 40–42 dBA. That means sound masking.

What Sound Masking Actually Does to the AI

Sound masking raises the background noise floor from its natural level (28–35 dBA in a modern office) to a target level (40–45 dBA) using loudspeakers distributed in the ceiling plenum. The spectrum is shaped to match HVAC ambient noise — heavy in the low-to-mid frequencies (250 Hz–2 kHz) where speech intelligibility is highest.

The AI calculation at different background noise levels demonstrates the effect directly:

Scenario: Talker at normal level (60 dBA at 1 m), treated open office (NRC 0.80 ceiling), listener at 4 m behind 1.2 m partition

Estimated speech level at listener: 38 dBA (after reverberant treatment and partial barrier).

The shift from 35 dBA background (best-case acoustic treatment) to 42 dBA (standard sound masking) changes the AI from 0.28 (Marginal) to 0.07 (Normal). That 7 dB increase in background is doing more work than all the acoustic panels combined.

The Actual Mechanism: SNR, Not Absolute Level

The insight that most office designers and clients miss is this: the problem with open office speech privacy is not that offices are too loud. Modern offices are often too quiet for privacy. A well-engineered HVAC system with properly attenuated supply diffusers and return paths creates an ambient noise level of 28–32 dBA — far below the threshold needed for speech to be masked.

The WELL v2 Feature 74 requirement for background noise levels sets a maximum of 45 dBA — but sets no minimum. An office that achieves RC 20 background noise (approximately 28 dBA equivalent) is performing excellently for hearing conservation and workstation noise compliance, while simultaneously creating the worst possible speech privacy conditions.

This is the central paradox of modern acoustic office design: the better you are at controlling HVAC and mechanical noise, the worse your speech privacy becomes — unless you add masking.

The solution is never to allow HVAC noise to remain high for privacy purposes. That trades one problem (distraction from mechanical noise) for another (poor speech privacy). The correct solution is: control HVAC noise AND add masking to achieve the target background level for speech privacy.

Sound Masking Specification Fundamentals

If you accept that masking is necessary, the specification must address four parameters:

1. Target masking level. Typically 42–45 dBA for Normal/Confidential privacy. ASTM E1374 specifies test methods for measuring masking effectiveness. The target should be defined as an equivalent continuous level (LAeq) measured at workstation height, not a ceiling-level measurement.

2. Masking spectrum. The ANSI/ASA S12.2 standard provides the target spectrum for masking noise — a bell-shaped curve peaking around 500–1000 Hz. Products that deviate significantly from this spectrum sound harsh (too much high-frequency energy) or muffled (too much low-frequency energy) and may be perceived as distracting despite being at the correct level.

3. Spatial uniformity. Masking level must be uniform across the floor plate — within ±2 dBA at workstation positions. Variance above this threshold creates "privacy holes" where the masking is below target and speech is intelligible, and "hot spots" where the masking is above target and becomes distracting.

4. Loudspeaker spacing and placement. Typical commercial masking systems use directional in-ceiling loudspeakers in the plenum, pointing upward to reflect off the structural deck, typically spaced at 1.5–2.0× ceiling height apart. The plenum depth and deck material affect the reflected spectrum and must be accounted for in the system design.

Combining Treatment and Masking: The Correct Approach

Acoustic treatment and sound masking are complementary, not alternatives. The optimum specification uses both:

Ceiling treatment (NRC 0.80–0.85, 60–70% coverage): Reduces reverberant speech level by 6–9 dB compared to a reflective ceiling. This reduces the speech signal the masking must cover, meaning the masking system can be set lower (42 dBA rather than 46 dBA), improving occupant acceptance.

Workstation partitions (1.2–1.5 m height, NRC 0.65 surface): Provide direct-path attenuation for workstation pairs in line with the partition. Reduce speech level at shielded positions by 8–12 dB. Without these, the masking system must be set higher to achieve the same SNR at shielded positions.

Sound masking (40–43 dBA, shaped spectrum): Raises the noise floor to mask residual speech. At this target, the combined system achieves AI < 0.10 at 4 m for most workstation orientations — Normal to Confidential privacy.

The three-element system (ceiling + partitions + masking) achieves Confidential privacy at 4 m and Normal privacy at 2 m — which is the correct target for an open-plan collaborative workspace. Trying to achieve the same result with ceiling + partitions alone fails at the 2 m position, which is where the majority of overheard conversations occur.

Use AcousPlan's speech privacy calculator to model the AI at your specific workstation geometry, ceiling treatment level, partition height, and background noise floor. The calculator implements the ASTM E1130 Articulation Index method and outputs the privacy class at 2 m, 4 m, and 6 m distances, clearly showing whether your proposed specification achieves Normal or Marginal privacy and whether masking is required.

Addressing the Client Objection to Sound Masking

The most common client objection to sound masking is: "Our staff will find it annoying." This concern is legitimate but misplaced. The occupant acceptance problem with masking systems is almost always a design and commissioning problem, not an inherent property of masking.

Masking systems that sound harsh are misspecified — the spectrum is too high-frequency-heavy. Masking systems that are consciously perceived by occupants as "white noise" are at the wrong level — either too high (above 46 dBA) or with the wrong spectrum. A correctly specified masking system at 42 dBA with a speech-shaped spectrum is perceived as HVAC ambient. Occupants do not notice it.

The proof is that every occupied office has some background noise. The masking system replaces the unreliable, variable, spectrally uncontrolled noise of HVAC and mechanical systems with a reliable, spatially uniform, acoustically optimised background. When masking is removed in a commissioned system, the overwhelming occupant response is that the office feels uncomfortably quiet and private conversations feel exposed — which is exactly correct.

Summary

Normal speech privacy in an open office requires an ambient noise floor of approximately 40–42 dBA. Acoustic treatment alone — regardless of how extensively it is specified — cannot raise the ambient noise floor. It can only reduce the speech signal. In typical open office conditions, the speech signal after best-case acoustic treatment still produces Marginal privacy at 2 m distances and Normal privacy at best at 4 m distances.

Sound masking is not an optional upgrade. It is the mechanism by which speech privacy moves from Marginal to Normal. The Articulation Index calculation makes this explicit: a 7 dB increase in background noise (from 35 dBA to 42 dBA) reduces the AI more than all the acoustic treatment combined.

The next time a client asks for speech privacy through treatment alone, show them the SNR calculation. Then specify masking.