Testing and fault-finding · Reference

Six assessment points, no misses allowed: the voice alarm intelligibility figures, correctly stated

What British Standard BS 5839-8:2023 recommends for speech intelligibility, how it is measured, and when a listening test is enough.

By Nigel Williams · Published 16 October 2026

Everyone has stood on a station platform and heard an announcement without catching a word of it. The voice was loud enough. You could tell it was English. You still had to ask the person next to you which platform.

Loud is not the same as understood. For voice alarm systems, the code of practice puts numbers on the difference.

Two ways to judge whether a message is understood

BS 5839-8:2023 gives two routes to assessing intelligibility. One is objective, measured with an instrument (Clause 20.3). The other is subjective, a structured listening test (Clause 20.4). The 2023 foreword records that the intelligibility clause was substantially rewritten to cover both.

Which route applies is not left to the day of commissioning. The method should be agreed by the interested parties and written into the system specification (Clause 20.2.1). In a simple acoustic space the assessment is usually subjective, unless the specification says otherwise. In a dispute, the code points to an objective test (Clause 20.2.1).

A "simple acoustic space" has a definition. Among other conditions, its average reverberation time across the 500 Hz and 1 kHz octave bands is not greater than 1.3 s (Clause 3.25). A hard-surfaced atrium or concourse may well not meet it.

The objective figures

The objective route uses the Speech Transmission Index (STI), measured to BS EN IEC 60268-16:2020. STI runs from 0 to 1. Voice alarm commissioning normally uses the Speech Transmission Index for Public Address systems (STIPA), a condensed version of the method (BS EN IEC 60268-16:2020, Annex B).

BS 5839-8:2023 recommends two things across an area (Clause 20.3.2):

• the average STI across the measurement positions should be at least 0.50
• no more than 10 % of positions should fall below 0.45

The average and the 10 % rule are new in the 2023 edition. The 2013 edition set 0.5 as the minimum value, and 0.45 appeared only in a note, as a lower figure the interested parties might agree for very reverberant or noisy spaces, or where the listeners know the messages from regular tests (BS 5839-8:2013, Clause 22.3.3 and Note 1). Pages that still quote a single figure are quoting half of it.

The change runs in both directions. In 2013, unless a lower figure had been agreed, the code recommended that every position reach 0.5. Now a position between 0.45 and 0.50 is acceptable as long as the average still reaches 0.50, and in a larger area a few positions may fall below 0.45. For a small area measured at the minimum number of points, the 0.45 floor applies to every position.

One more point that surprises people. The measurement standard does not set the pass mark. BS EN IEC 60268-16 says in its scope that it does not provide criteria for certifying a voice alarm system. Its Annex G calls 0.50 the "target value for VA systems", meaning voice alarm systems, and both its band annexes are informative. The pass mark comes from BS 5839-8.

How many positions, and how many may miss

The number of assessment points depends on the size of the area (Clause 20.2.2 c), Table 1). Each point gets at least three STI or STIPA measurements (Clause 20.2.2 d)). We ran the 10 % rule against the table:

• up to 10 m²: one point, none may fall below 0.45
• 10 m² to under 100 m²: three points, none
• 100 m² to under 500 m²: six points, none
• 500 m² to under 1 500 m²: ten points, one
• 1 500 m² to under 2 500 m²: 15 points, one, because 10 % of 15 is 1.5 and a position either passes or does not

So for any area under 500 m² assessed at the Table 1 minimum, the 10 % allowance is worth nothing. Every position should reach 0.45, and the average should reach 0.50.

At the other end, one low position still costs the rest. With 15 points and one at 0.44, the other 14 need to average about 0.504 to bring the area to 0.50.

Where the points go matters as much as how many. They should be spread evenly, with no preference for good or bad spots, and no more than a third on a loudspeaker's axis (Clause 20.2.2 e)). Listening height is 1 m to 1.2 m seated, or 1.5 m to 1.7 m standing (Clause 20.2.2 f)).

When a listening test is enough

The subjective route is a simple listening test. The result is agreed as "intelligible" or "not intelligible", not given a score (Clause 20.4, commentary). The code recommends reserving it for people who know the messages and the evacuation strategy, or where other measures are in place, such as non-public areas or areas where the public are always escorted by staff (Clause 20.4.1).

It still has recommendations. The listeners should not have a known hearing impairment, the messages should be varied so they do not become familiar, every pre-recorded emergency message is assessed, and the test is run under real or simulated emergency noise (Clauses 20.4.2 to 20.4.4).

From 2023 there is also an annual listening test in every loudspeaker zone, as part of servicing (Clause 40.2 j)). That checks the system still does what it did at handover. It is not a fresh choice of route.

"Can't we just add loudspeakers until it passes?"

Sometimes. But more loudspeakers won't fix the room.

STI falls with reverberation and with background noise. Both belong to the space. Adding loudspeakers can raise the level and shorten the distance to the listener, and in a hard, echoing hall it can also add more sound for the room to smear. Where the room is the limit, the honest answer may be acoustic treatment, or an agreed acceptable level documented under Clause 20.3.3, rather than more hardware. Outside a simple acoustic space, the code recommends that the objective assessment is done by a competent acoustician (Clause 20.3.4).

Two examples

The small office. A 400 m² open-plan office gets six assessment points. Five read 0.55 and one reads 0.44. The average is about 0.53, comfortably above 0.50. The area still does not meet Clause 20.3, because six points allow no position below 0.45.

The concourse. A 2 000 m² concourse gets 15 points. One position near a glazed wall reads 0.43. That is the single position the 10 % allowance covers, provided the average across all 15 is at least 0.50. A second low position would not be.

In short

• Two routes: objective STI (20.3) or a subjective listening test (20.4), agreed in the specification beforehand (20.2.1).
• Objective: average at least 0.50, and no more than 10 % of positions below 0.45 (20.3).
• Under 500 m², at the minimum number of points, that 10 % allows no misses at all (Table 1 and the arithmetic above).
• The pass mark comes from BS 5839-8, not from the measurement standard.

The platform announcement was loud enough. The code asks a different question: could you tell which platform?

Next step. Find the system specification and check whether it states which intelligibility route was agreed (Clause 20.2.1). If it is silent, that is worth settling in writing before the next commissioning or dispute, not after. Questions on a figure someone has quoted to you: contactus@proaudium.com or 01273 034630.