Upward Spread of Masking

Human hearing exhibits upward spread of masking: energy in a given frequency band partially masks signals in higher-frequency bands. A strong 250 Hz noise component makes it harder to hear speech at 500 Hz and 1 kHz, even if the SNR in those bands appears adequate. IEC 60268-16 models this effect to produce more accurate STI predictions.

Masking Model (Clause 5.2)

The Ed.5 masking model calculates an effective noise level in each band that includes contributions from the band below:

Lnoise,eff(k) = 10·log₁₀ [ 10^(Lnoise(k)/10) + 10^((Lspeech(k-1) - ΔM)/10) ]

Where ΔM is the masking attenuation (approximately 26 dB for Ed.5, slightly higher than the 18 dB used in earlier editions). The masking contribution propagates upward from band k-1 to band k, but does not propagate further to k+1 from this indirect source.

Effect on STI

The masking correction reduces the effective SNR in mid and high-frequency bands when low-frequency noise or speech levels are high. This means:

Spaces with strong HVAC rumble (high energy at 125-250 Hz) score lower than the raw per-band SNR would suggest
High speech levels can partially mask themselves in upper bands — relevant for very loud PA systems
The correction is most significant for the 500 Hz and 1 kHz bands, which carry the heaviest STI weighting

Absolute Threshold

The standard also includes an absolute hearing threshold correction. If the received speech level in any band falls below the hearing threshold (approximately 46 dB SPL at 125 Hz, 0 dB at 1 kHz), the MTF for that band is reduced accordingly. This matters for very quiet PA systems or distant listener positions where the speech level approaches the threshold of hearing.

Practical Impact

In a well-designed PA system with broadband background noise at 45 dBA, the masking correction typically reduces STI by 0.01–0.03 compared to the non-masked calculation. In spaces with dominant low-frequency noise (e.g., near traffic, industrial HVAC), the correction can reduce STI by 0.05–0.08 — enough to shift a rating category.

Noise Correction in Practice

The STIPA measurement inherently includes the effect of masking because the received signal contains both the transmitted test signal and the ambient noise. The masking model is primarily needed when computing STI from impulse response data with a separately measured noise spectrum.

SonaVyx Implementation

The STI tool applies the Ed.5 masking model when computing STI from impulse response or from the RT60+SNR simplified method. For direct STIPA measurements, masking effects are captured naturally. The per-band calculation display shows both raw and masking-corrected SNR values for transparency.