ProAudium · Speech Transmission Index
See what the room does to speech
The Speech Transmission Index (STI) rates how much of the rise and fall of speech survives the journey from the announcer to the listener. Move the sliders to see what reverberation, noise and volume do to it.
The explorer
Reverberation is like snow
Reverberation keeps each sound hanging in the air, so it fills the dips before the next sound arrives. The fast rises and falls go first; with more reverberation, the syllables run together.
Noise is like a lake
Noise fills the dips up to the waterline. The peaks still show, but the gap between peak and dip shrinks. STI reads that lost depth, not the average level.
Inside the STI number
Set the room's reverberation and noise, then try turning the volume up. The picture shows what happens to the speech; the steps below show how the same two factors become the number.
How much of each rise and fall survives
STI tests 14 modulation rates, from 0.63 to 12.5 times a second. For each one it works out the share of the original depth that survives, called the modulation transfer, m. Reverberation and noise each take a share, and the two multiply.
m = m(reverberation) × m(noise) m(reverberation) = 1 ÷ √(1 + (2π × rate × T ÷ 13.8)²) m(noise) = 1 ÷ (1 + 10^(−S/N ÷ 10))
Turn each m into an apparent signal-to-noise ratio
Each surviving share is expressed as if it were a signal-to-noise ratio, then kept within ±15 dB: beyond +15 dB speech is treated as perfect, below −15 dB as lost.
apparent S/N = 10 × log₁₀(m ÷ (1 − m)), limited to −15 … +15 dB
Scale to a 0 to 1 index and average
transmission index = (apparent S/N + 15) ÷ 30 result = average of the 14 transmission indices
All 14 rates
What the full method adds
A real STI repeats this in seven octave bands from 125 Hz to 8 kHz, each with its own reverberation and noise, then combines the bands with weights that reflect how much each contributes to speech. It also corrects for masking between neighbouring bands and for the threshold of hearing, which is why turning a real system up keeps paying off less. This page does one band and leaves those corrections out, so treat its number as a guide to direction, not a figure for any real system.
A loud low sound hides quieter high sounds
The “z” in gazelle is a low buzz and a high hiss at the same moment. Turn the buzz up and its masking spreads further up the frequency range until the hiss disappears.
Illustrative masking curve; real curves depend on the sounds and the listener.
Estimate it for a real space
To put a range on the STI for a real room, from its size and finishes or its reverberation time and noise, use the estimator.
Written by Nigel Williams, ProAudium. The reverberation and noise figures use the modulation transfer function from Houtgast and Steeneken; speech shapes are drawn, not recorded.