STI and STIPA: how speech intelligibility is measured for PA, voice alarm and PAGA systems

What is STI?

Speech is a carrier sound whose loudness rises and falls several times a second: syllables, words and pauses. A listener understands speech by following those fluctuations. Anything that fills in the quiet gaps makes the fluctuations shallower and speech harder to follow. In practice that is reverberation, which smears each syllable into the next, background noise, which fills the gaps from below, and long echoes.

STI measures exactly that. A test signal with known fluctuations, its modulation, is played at the talker's or loudspeaker's position, and at the listener's position the meter measures how much of the modulation is left. This is done in seven octave bands from 125 Hz to 8 kHz, which together cover the frequencies that carry speech. The results are weighted by how much each band contributes to intelligibility and combined into one number. An STI of 1 means the speech arrives unchanged; an STI near 0 means none of the fluctuations survive.

Because STI is computed from the physics of the transmission path, and not from listeners scoring words, two engineers measuring the same position with the same method should get the same result. That repeatability is why it is used for acceptance testing.

Full STI, STIPA and the indirect method

IEC 60268-16 allows the index to be obtained in more than one way. The difference between them is how many modulation frequencies are tested, and therefore how long a measurement takes.

STIPA is the method used for almost all public address and voice alarm testing. The test signal is played through the system itself, so the result includes everything between the microphone input and the listener's ear: the amplifiers, the loudspeakers, the room and the noise. The older RASTI method was withdrawn from the standard and should not appear in a new specification.

What the numbers mean: the qualification bands

The current edition of IEC 60268-16 groups STI values into qualification bands, so a specification can call for a band rather than a bare number. Each band is 0.04 wide.

The figure most voice alarm specifications ask for, a mean STI of 0.50, sits in band G. Rooms meant for speech, such as lecture theatres and meeting rooms, are designed for higher bands. A value below about 0.45 means that a listener who does not know the message in advance will miss a significant part of it, which is why specifications put the floor for individual positions there.

What lowers STI, and what raises it

• Reverberation: in an atrium, terminal or factory hall each syllable is still ringing when the next one starts. More loudspeakers closer to the listener, directional loudspeakers, or absorption raise STI
• Background noise: plant, traffic, crowds and ventilation fill the gaps in speech. A higher level helps only up to a point; local loudspeakers near the listener or visual alarms are the next step
• Long-delay echoes: outdoors, and in very large spaces, the same message arrives from two distant loudspeakers more than about 50 to 100 ms apart. Delay alignment or closer loudspeaker spacing removes them
• Too much level: above a certain level the loudspeakers distort, the room reverberates more, and the ear itself masks part of the speech. STI often falls when an installer turns the system up
• The message itself: a fast, poorly recorded or badly equalised message loses intelligibility that no loudspeaker layout can recover

How STIPA is measured on site

The listener positions are agreed on plan before the test, in a grid that covers each zone, at seated or standing ear height. The system plays the STIPA test signal at the level of the real emergency message. The meter at each position records the result after 15 to 25 seconds. Each position is measured at least twice, and a third time if the two readings differ by more than 0.03.

Where the building or plant cannot run at full noise during the test, the ambient noise is measured separately in octave bands during normal operation, and the STIPA result is corrected for it as the standard allows. Without this step a test made in a quiet, empty building can report an intelligibility the system will never deliver when the building is in use.

The report gives STI per position and the mean and minimum per zone, compared with the criterion in the specification, together with the signal level, the instruments and their calibration. Geonoise carries out this test as an independent commissioning witness; see PAGA and voice alarm STIPA testing. To check a design before installation, see speech intelligibility prediction.

Which standards apply

Voice alarm specifications built on these documents commonly require a mean STI of at least 0.50 in each zone, with no position below about 0.45. Oil and gas and marine PAGA specifications usually set their own figure for intelligibility and a margin above ambient noise. The figure that governs a project is the one its specification names.

Audibility is a separate test

A fire alarm sounder or alarm tone does not need to be intelligible, but it does need to be heard. Audibility is measured as the A-weighted level of the alarm at listener height and its margin above the ambient noise. Typical requirements are a fixed minimum, such as 65 dB(A) in general areas and 75 dB(A) at the bedhead in BS 5839-1, together with a margin above ambient of around 10 to 15 dB, depending on the standard. A voice alarm has to pass both tests: loud enough to be heard, and clear enough to be understood.

Frequently asked questions