Speech Intelligibility: the 97% and 90% Requirements and How to Assess Them at the Design Stage

On 1 March 2026, Russia tightened its speech-intelligibility requirements for mass-notification (public address and voice alarm) systems. Here's what actually changed, why those percentages can't simply be "plugged into a formula," and how to assess intelligibility while the project is still on the drawing board — with references to the primary sources.

Can speech intelligibility be calculated with a formula?

Not the regulated percentage figures. Intelligibility expressed as a percentage is verified by a subjective articulation test — per GOST 16600-72, a Russian standard — run on the finished audio chain with a panel of trained talkers and listeners. At the design stage you can only predict it, via the Speech Transmission Index (STI) described in GOST R ISO 9921, the Russian adoption of the international ISO 9921. That is an estimate, not a measurement.

What intelligibility is required as of 1 March 2026?

Under GOST R 42.3.01-2021 as amended (Amendment No. 1, clause 5.5.3): word intelligibility must be at least 97%, and sound (syllable) intelligibility at least 90%. Before 1 March 2026 the thresholds were 95% and 86%.

Why does the test method still cite 95% and 86% while the requirement says 97% and 90%?

The requirement in GOST R 42.3.01 was raised by Amendment No. 1, but the companion test procedure — GOST R 42.3.03-2024, clause 5.7.2 — still keeps the old 95%/86% thresholds. The two documents aren't yet harmonised. The safe engineering choice is to write the stricter 97%/90% figures into the design brief.

What STI value counts as good enough?

On the GOST R ISO 9921 scale (Table F.1, the same rating bands used internationally): above 0.75 is "excellent," 0.60–0.75 "good," and 0.45–0.60 "fair." There is no direct conversion between STI and an intelligibility percentage — they are two different rating systems — so STI is used as a design target, not as a stand-in for the acceptance test.

How do you engineer intelligibility through equipment choice?

Three levers do the work: an even sound field (many quieter loudspeakers spaced close together rather than a few loud ones), adequate bandwidth in the speech chain, and a signal-to-noise margin of at least 10 dB over the ambient noise. You can predict coverage and the number of loudspeakers point by point across a room in our calculator.