Calculate ANSI/ASA S3.5-1997 (R2024) Speech Intelligibility Index

Calculates the American National Standards Institute (ANSI) S3.5-1997 Speech Intelligibility Index (SII) (ANSI, 1997), a standard method for computing the intelligibility of speech from acoustical measurements of speech, noise, and hearing thresholds. This package includes data frames corresponding to Tables 1 - 4 in the ANSI standard as well as functions utilizing these tables and user-provided hearing threshold and noise level measurements to compute the SII score. The methods implemented here extend the standard computations to allow calculation of SII when the measured frequencies do not match those required by the standard by applying interpolation. The package also includes a native C++ implementation of the 'AUDMOD' specific loudness model for impaired hearing (Bramslow, 2004), ported from the 'bramslow2004' implementation in the Auditory Modeling Toolbox (AMT), to calculate loudness in sones. It provides methods to predict aided SII under hearing aid prescriptive rationales, including Open-NL (Open Non-Linear), a theoretical optimization framework that uses a Nelder-Mead simplex search to maximize the SII (by default with the Johnson and Dillon (2011) desensitization correction), subject to loudness penalties and simulated Maximum Power Output (MPO) constraints. Further functions calculate real-ear insertion gains, estimate maximum power output (SSPL90), and prescribe dynamic range compression. Open-NL and the prescription functions are research tools for theoretical modeling and simulation; they have not been clinically validated and are not intended for fitting hearing aids to patients. Development of this package was originally funded by the Center for Bioscience Education and Technology (CBET) of the Rochester Institute of Technology (RIT).

SII: Speech Intelligibility Index, Loudness, and the Open-NL Prescription Testbed

The SII package for R provides:

• The ANSI S3.5-1997 (R2024) Speech Intelligibility Index, with the standard's band-importance tables, extended by interpolation to audiograms measured at non-standard frequencies, and an optional Johnson & Dillon (2011) hearing-loss desensitization correction.
• A loudness model for impaired hearing: a C++ port of AUDMOD (Bramsløw, 2004), verified stage by stage against the bramslow2004 implementation in the Auditory Modeling Toolbox (AMT 1.6.0).
• Open-NL, an open, inspectable testbed for hearing aid prescription research. Open-NL searches for insertion gains that maximize the SII subject to a loudness limit and a simulated maximum power output, so that each assumption in a prescription can be changed or removed and its effect measured.

Open-NL and the prescription functions are research tools for modeling and simulation. They have not been clinically validated and are not intended for fitting hearing aids to patients.

Installation

From CRAN:

install.packages("SII")

Development version from GitHub (needs a C++ compiler: Rtools on Windows, Xcode Command Line Tools on macOS):

# install.packages("remotes")
remotes::install_github("r-gregmisc/SII")

Examples

SII for a speech and noise spectrum (ANSI S3.5 Annex C, Example C.1)

library(SII)

res <- sii(
  speech    = c(50, 40, 40, 30, 20, 0),
  noise     = c(70, 65, 45, 25, 1, -15),
  threshold = c(0, 0, 0, 0, 0, 0),
  method    = "octave"
)
res$sii          # 0.504, the value given in the standard
summary(res)     # band-by-band detail
plot(res)

SII with hearing loss desensitization

desensitization is either "none" (the ANSI SII, the default) or "johnson2011_desensitized" (the Johnson & Dillon, 2011, correction for reduced benefit from audibility at greater degrees of sensorineural loss).

sii(
  speech    = "raised",
  threshold = c(25, 25, 30, 35, 45, 45, 55, 60),
  freq      = c(250, 500, 1000, 2000, 3000, 4000, 6000, 8000),
  method    = "critical",
  interpolate = TRUE,
  desensitization = "johnson2011_desensitized"
)

Open-NL gain targets

target <- open_nl(
  speech    = 65,                               # input level, dB SPL
  threshold = c(0, 0, 10, 40, 70, 80),          # dB HL
  freq      = c(250, 500, 1000, 2000, 4000, 8000)
)
target          # insertion gain and maximum power output by frequency
target$gain

The optimization runs several Nelder-Mead searches and can take up to a minute per call. Useful arguments include objective_sii (which SII to maximize), cap_rule ("normal", the normal-hearing loudness of unaided speech, or the earlier "legacy" rule), cap_override, vent_floor, and switches for individual prescription rules (enable_severe_booster, disable_sdlfp, abg_fraction). See ?open_nl.

Aided SII and loudness for an Open-NL target

Pass the target to sii() through prescription, with the speech spectrum, thresholds and frequencies stored in the target:

aided <- sii(
  speech       = target$speech,
  threshold    = target$threshold,
  freq         = target$freq,
  prescription = target,
  interpolate  = TRUE,
  desensitization = "johnson2011_desensitized"
)
aided$sii          # aided SII
aided$unaided_sii  # unaided SII, same settings

calculate_loudness(target)$total  # loudness of the aided speech, sones

calculate_loudness_audmod() gives direct access to the AUDMOD model; see ?calculate_loudness_audmod.

Testing a prescription rule

Each rule in Open-NL can be switched on or off to measure its effect, for example:

target_booster <- open_nl(
  speech    = 65,
  threshold = c(0, 0, 10, 40, 70, 80),
  freq      = c(250, 500, 1000, 2000, 4000, 8000),
  enable_severe_booster = TRUE,  # extra gain for severe losses
  booster_onset = 60,            # dB HL at which the booster starts
  disable_sdlfp = TRUE           # turn off the slope-dependent low-frequency penalty
)
target_booster$gain - target$gain

Reproducing published results

Scripts that reproduce the results of studies using this package, and the checks of the loudness model against AMT, are in reproducibility_scripts/. Its README maps each table and figure to the script that produces it and says which package version each script runs on. These scripts are kept in the GitHub repository only; they are not part of the CRAN package.

What's new

See NEWS.md. Version 1.3.0 replaced the loudness model with the verified AUDMOD port, made the normal-hearing loudness cap the Open-NL default, and simplified desensitization to two options.

Authors

Gregory R. Warnes (original author of the ANSI S3.5 implementation) and Mark Shaver (maintainer; loudness model and Open-NL). Development of the original package was funded by the Center for Bioscience Education and Technology (CBET) of the Rochester Institute of Technology.

License

GPL-3.

References

• ANSI/ASA S3.5-1997 (R2024). Methods for Calculation of the Speech Intelligibility Index. Acoustical Society of America.
• Bramsløw, L. (2004). An objective estimate of the perceived quality of reproduced sound in normal and impaired hearing. Acta Acustica united with Acustica, 90(6), 1007–1018.
• Johnson, E. E., & Dillon, H. (2011). A comparison of gain for adults from generic hearing aid prescriptive methods: Impacts on predicted loudness, frequency bandwidth, and speech intelligibility. Journal of the American Academy of Audiology, 22(7), 441–459.