Testing and fault-finding · Series, part 2
One STI reading is not a pass mark
Why the figures on a commissioning report are really ranges, and the one question to ask at your next service visit.
Reviewed on
In 1676 the astronomer Ole Rømer noticed that the eclipses of Jupiter's moon Io ran late when the Earth was further from Jupiter, and early when it was nearer. Light was taking time to cross the gap. The speed worked out from his timings came out about 27 % too slow. His method was sound. The orbit sizes he had to plug in were wrong, and with today's figures the same method lands almost exactly on the modern value.
A figure is only as good as what goes into it. That's worth remembering the next time you look down the STI column of a commissioning report.
A figure on a report looks exact
Most commissioning reports list one Speech Transmission Index (STI) figure for each position. 0.52 here, 0.48 there. It looks like a measurement to two decimal places, and it reads like a pass or a fail.
It's really the middle of a range. There are three reasons, and none of them is anyone's fault.
The meter wobbles from one reading to the next
Step on the bathroom scales twice and you'll get two weights. Nobody thinks they gained half a kilo between steps.
STIPA, the quick version of STI used on site, works the same way. Its test signal is pseudo-random, so repeat readings at the same spot with nothing changed vary by about 0.02 to 0.03 (Mapp, 2006). Different instruments disagree by more. In one comparison, six measuring platforms at a single position spread across 0.15 (Mapp, c. 2002).
The code of practice already allows for this. BS 5839-8:2023 asks for at least three STI or STIPA readings at each assessment point, not one (Clause 20.2.2 d).
The building on the day isn't the building in a fire
Readings are usually taken out of hours, in an empty building. The code itself notes that this can change the acoustics, and asks for any difference from normal conditions to be recorded (Clause 20.2.2 b). It also asks for the noise expected in an emergency, such as smoke-extract and stair-pressurisation fans, to be taken into account (Clause 20.2.2 g).
Every one of those is an input, like Rømer's orbits. A room that's a little more reverberant than assumed, or a fan that's a little louder, moves the true figure. Our STI estimator shows this as a shaded band around each input.
So the honest figure is a range
Put the meter's wobble and the day's conditions together and each figure on the sheet is a range. Two positions reading 0.51 and 0.49 may well be the same.
The code doesn't judge single positions anyway. For an area, it looks at the average of all the assessment points, which should normally be 0.50 or more, the level at which a voice alarm will be reasonably intelligible. It also asks that at least nine in ten points reach 0.45 (Clause 20.3.2). We've set out those figures in full here. One position at 0.48 isn't a fail on its own. An average of 0.51, built from single readings, is fragile.
The useful question isn't "did it pass?" It's "how likely is this area to read 0.50 or more next time?" The estimator answers that as a percentage.
"If the report says pass, why should I care?"
Because handover isn't the last time anyone measures. A pass that sits right on the line can read as a fail at the next check with nothing changed, and then you're paying to investigate a fault that isn't there. Knowing which areas are close lets you watch those, and leave the rest alone.
Won't asking start an argument with your contractor? It shouldn't. It's a question about method, and the code backs it.
The figure can be low too
The range runs both ways. On a new Sydney underground platform, measured STIPA averaged 0.46, about 0.1 below what word tests with real listeners showed at every position (Fong, Hanrahan-Tan and Leembruggen, 2024). The platform passed on the word tests.
What to take away
- A single reading wobbles by a few hundredths, and meters can differ by more.
- The conditions on the day are inputs, and they move the figure too.
- So each figure is a range, and the code judges the area, not one spot.
Rømer's method was right and his inputs weren't. A single STI reading is the same: the meter does its job, and the figure is still only as exact as everything that went into it.
Next step. At your next service visit, ask one question: "How many readings did you take at each assessment point, and how far apart were they?" A good answer is at least three, averaged, with the noise conditions and the loudspeakers that were on written down alongside (Clause 20.2.2 a and j). Then read the next article: What STI will we get? Ask the room first.
Questions
How much do repeat STIPA readings vary?
STIPA's test signal is pseudo-random, so repeat readings at the same spot with nothing changed vary by about 0.02 to 0.03 (Mapp, 2006). Different instruments disagree by more. In one comparison, six measuring platforms at a single position spread across 0.15 (Mapp, c. 2002).
Is one position below 0.50 a fail?
One position at 0.48 isn't a fail on its own. For an area, BS 5839-8:2023 looks at the average of all the assessment points, which should normally be 0.50 or more, and asks that at least nine in ten points reach 0.45 (Clause 20.3.2).
What should I ask at my next service visit about STI readings?
Ask how many readings were taken at each assessment point, and how far apart they were. A good answer is at least three, averaged (Clause 20.2.2 d), with the noise conditions and the loudspeakers that were on written down alongside (Clause 20.2.2 a and j).
Where this comes from
Sources: BS 5839-8:2023, Clauses 20.2.2 and 20.3.2 (paraphrased); Mapp (2006), Proc. IOA 28(6); Mapp (c. 2002), "Some further thoughts on STI"; Fong, Hanrahan-Tan and Leembruggen (2024), Proc. IOA 46(3); Today I Found Out, "How the speed of light was first measured".