Voice Evacuation Intelligibility: NFPA 72 Annex D Explained

The requirement: intelligible, not just audible

Audibility and intelligibility are different tests. Audibility is a sound level: NFPA 72-2022 §18.4.4.1 requires the signal to be at least 15 dBA above the average ambient (or 5 dBA above a sustained maximum), and the 110 dBA cap of §18.4 still applies. Intelligibility is whether the words can be understood: §18.4.11 requires voice communications systems to reproduce prerecorded, synthesized, or live messages with voice intelligibility, and Chapter 24 (emergency communications systems) applies that requirement to each acoustically distinguishable space where intelligibility is required. The code states the requirement; Annex D explains how to measure it and what passes.

What STI and CIS measure

The Speech Transmission Index (STI) is a 0-to-1 score of how well a transmission path preserves the modulations of speech — the rises and falls in loudness that carry syllables. Reverberation smears those modulations in time; background noise buries them. A path that preserves them perfectly scores 1.0; one that destroys them scores near 0. STIPA is the quick, speech-like test signal that modern meters use to measure STI in the field. The Common Intelligibility Scale (CIS) is a logarithmic remapping, CIS = 1 + log10(STI), that spreads out the useful range: STI 0.45 is CIS 0.65, STI 0.50 is CIS 0.70. The IEC 60268-16 qualification bands referenced by Annex D are:

The Annex D pass criteria

Annex D.2.4.1 sets two conditions that must both be met within each acoustically distinguishable space: at least 90 percent of the measurement locations have a measured STI of at least 0.45 (CIS 0.65), and the average of all measurement locations in the space is at least 0.50 STI (0.70 CIS). The first rule catches dead spots — a corner behind a column, a spot between two speakers. The second keeps a space from passing on a handful of great readings under the speakers. Both are per space; you cannot average a bad corridor into a good lobby.

Acoustically distinguishable spaces

An acoustically distinguishable space (ADS) is any emergency communications system notification zone, or any subdivision of one, that is distinguished from other spaces by its acoustic, environmental, or use characteristics — size, finishes, ambient noise, ceiling height, or function. Walls are not required: an open office, the corridor that runs past it, and the stair at the end are three spaces. Designating the ADSs is the designer's first job on a voice system, and the designer also records which of them require intelligibility. Mechanical rooms, private restrooms, and small closets are commonly designated audible-only; assembly spaces, corridors, lobbies, and open offices are not.

Why loud rooms fail: reverberation and noise

Two things destroy intelligibility, and turning up the amplifier fixes only one of them.

Noise. If the speech at the listener is only a few dB above the ambient, the quiet consonants that distinguish words disappear. Audibility needs +15 dBA; good intelligibility generally wants the speech well above that. Shutting down HVAC on alarm, where the design allows, is often the cheapest intelligibility improvement available.

Reverberation. In a hard, tall room, each syllable keeps bouncing for a second or two and arrives on top of the next one. Louder speakers make louder echoes; the ratio of direct to reverberant sound does not change. The Sabine equation estimates reverberation time from volume and absorption — RT60 ≈ 0.049 × V ÷ (α × S) in feet — and rooms above about 1.2 seconds get difficult, above 2 seconds very difficult. The fixes are acoustic: add absorption, use directional speakers aimed at listeners so the direct sound dominates, or distribute many speakers at low taps so every listener is close to one. The helper's Panel 2 uses a simplified speech transmission model (Houtgast and Steeneken's modulation transfer function) to estimate STI from these two inputs. It is an estimate for planning, not a code value and not a prediction a modeling program would sign.

Laying out speakers that pass

NFPA 72 does not give a speaker spacing. The industry's working rules come from geometry. A ceiling speaker with a 90° dispersion angle mounted 10 ft above the floor covers a circle at the 5 ft ear plane with a diameter of 2 × (10 − 5) × tan(45°) = 10 ft to its −6 dB points. Layouts are described by how those circles overlap: edge-to-edge (centers one diameter apart, cheap but with holes on the diagonals), minimum overlap (centers at D ÷ √2, the usual voice-evacuation choice), and center-to-center (centers at D ÷ 2, the most uniform). The sound level under a speaker follows the sensitivity equation: SPL = sensitivity (1 W at 1 m) + 10·log10(tap watts) − 20·log10(distance in meters). An 88 dB speaker on a 1 W tap, 5 ft (1.52 m) above the ear plane, delivers about 84 dB directly below and 78 dB at the coverage edge. Doubling the tap adds 3 dB; halving the height above the ear plane adds 6.

How the acceptance test is run

Annex D describes measurement with a STIPA-capable meter playing the system's own test signal through the actual speakers, at representative listener locations throughout each ADS at ear height, with the building in its normal operating condition (HVAC running unless it shuts down on alarm). Several readings are averaged at each location, each location is compared with 0.45, and the location averages are compared with 0.50. The report lists the ADSs, the locations, the meter, the signal, the ambient conditions, and the results. Zion runs these tests as part of commissioning; the helper's Panel 1 is the arithmetic the report has to show.

How the helper applies the rules

Panel 1 parses the readings you paste (STI or CIS), converts if needed, and applies the two Annex D.2.4.1 tests with a pass-fail result and the IEC qualification band of the average. Panel 2 computes volume and surface area, estimates RT60 by Sabine from the absorption preset you choose, takes the signal-to-noise ratio from your speech and ambient levels, checks the §18.4.4.1 audibility rule, and estimates STI from reverberation and noise together, naming the limiting factor. Panel 3 computes the coverage diameter from the dispersion angle and height above the 5 ft ear plane, the grid for the overlap pattern you choose, the sound level under a speaker and at the coverage edge from sensitivity and tap, the audibility margin at the edge, and the total amplifier load.

Worked example

A school commons 90 × 60 ft with a 24 ft deck, exposed structure, acoustic tile on part of the ceiling (α ≈ 0.15), ambient 55 dBA. Ceiling speakers at 24 ft are not practical, so pendant speakers are hung at 14 ft. 88 dB sensitivity, 90° dispersion.

Common errors

• Designing for audibility only. A system that meets +15 dBA everywhere can still fail STI in every hard room.
• Averaging across spaces. Annex D criteria apply within each ADS.
• Turning up a reverberant room. The echo rises with the signal; STI barely moves.
• Testing with HVAC off when it runs during an alarm. Measure the normal condition.
• Edge-to-edge layouts for voice. The diagonal holes become the 10 percent of locations that fail.
• Forgetting the design margin. Aim for an estimated 0.55 to pass a 0.50 average in the field.

Frequently asked

Is intelligibility required everywhere a voice system plays?

No. It is required in each acoustically distinguishable space the designer designates as requiring it, in consultation with the AHJ. Spaces designated audible-only still must meet the §18.4.4.1 sound level.

What meter do I need?

A STIPA-capable sound analyzer and the matching STIPA test signal played through the system. Annex D describes the method; the meter's manual describes the averaging.

Can I calculate STI instead of measuring it?

For design, yes — modeling software predicts it, and the helper's estimate flags trouble early. For acceptance, Annex D is built around measurement, and most AHJs expect measured results.

What does a 520 Hz tone have to do with voice systems?

§18.4.6.3 requires the low-frequency signal where audible appliances are intended to wake sleeping occupants. Voice systems serving sleeping areas typically precede the message with a 520 Hz square-wave tone to meet it.

Which code edition applies in Texas?

This guide uses NFPA 72-2022. Texas jurisdictions adopt different editions and local amendments govern; some adopt stricter STI thresholds. Confirm with your AHJ.

Related