Metrics › Roughness

Roughness from the Sottek Hearing Model (ECMA-418-2)

Unit: asper_HMS · Also known as ECMA-418-2 roughness, asper_HMS, Sottek roughness, SHM roughness

Roughness from the Sottek Hearing Model, Clause 7 of ECMA-418-2. The 2nd edition (2022) added an optional entropy weighting for cases where rotational speed is measured; the 3rd edition (2024) adapted zero-padding and segmentation, and its refinements were mainly to roughness.

Editions and who implements them

Project map

The project map, with only the lines for Sottek Hearing Model roughness between the projects listed for it. Agreement between a project and the code it was ported or adapted from only checks the port.

• ported or adapted from another project
• the author's own code
• used at run time
• results checked against
• same contributor
• SQAT SQAT, MoSQITo, and sottek-hearing-model

Scroll the picture sideways.

Implementations

Projects in bold and other established projects come first, then newly released, developing and legacy ones. Validation is what each project says about its own testing. The details are below, and the kinds of evidence are explained on the About page. The project map above shows which project was ported or adapted from which.

Also available through tools that call one of these implementations: rota-order (via sottek-hearing-model, ECMA-418-2:2025 (4th ed.)), pySQAT (via sottek-hearing-model, ECMA-418-2:2025 (4th ed.)), and PsychoBox (via MoSQITo, ECMA-418-2:2022 (2nd ed.)).

See also: Roughness, Daniel & Weber model, Loudness from the Sottek Hearing Model (ECMA-418-2).

How they were validated

As reported by each project; LISQM has not run the code. Two implementations that agree can still share the same error. Kinds of evidence.

SQAT · ECMA-418-2:2025 (4th ed.)

compared with ArtemiS SUITE (HEAD acoustics)

Ported or adapted from refmap-psychoacoustics (the author's own code).

• Checked for the 4th edition for the dependence of the roughness of AM tones on modulation frequency, and against reference results from HEAD acoustics ArtemiS SUITE 17.1 (named in the source of SQAT's validation page) for a binaural recording of a train station.

SQAT · ECMA-418-2:2024 (3rd ed.)

compared with ArtemiS SUITE (HEAD acoustics)

Ported or adapted from refmap-psychoacoustics (the author's own code).

• Checked for the dependence of the roughness of AM tones on modulation frequency, and against reference results from HEAD acoustics ArtemiS SUITE 15.7 (named in the source of SQAT's validation page) for a binaural recording of a train station.

MoSQITo · ECMA-418-2:2022 (2nd ed.)

standard or paper data also compared with ArtemiS SUITE (HEAD acoustics)

• A unit test checks the standard's reference signal (1 kHz tone, 100 % amplitude-modulated at 70 Hz, 60 dB), which should give 1 asper, within ±0.1 asper.
• A validation script compares the specific roughness of amplitude-modulated tones (carriers from 125 Hz to 4 kHz, 65 dB) with values computed in HEAD acoustics ArtemiS SUITE, within ±10 %.

sottek-hearing-model · ECMA-418-2:2025 (4th ed.)

standard or paper data (reference signal and one recording) also compared with ArtemiS SUITE (HEAD acoustics)

Ported or adapted from refmap-psychoacoustics (the author's own code).

• Unit tests check the standard's reference signal, which should give 1 asper_HMS, at 48 and 44.1 kHz.
• RefMap's verification against ArtemiS v17.1, as for loudness.

Checked by the maintainer of LISQM: In personal use, Zhenxian LI found its results in very good agreement with those of HEAD acoustics ArtemiS SUITE. This is a personal check, not a published comparison.

refmap-psychoacoustics · ECMA-418-2:2025 (4th ed.)

compared with ArtemiS SUITE (HEAD acoustics)

• Verified against HEAD acoustics ArtemiS v17.1 with the same three signals as loudness.

MetaSona new · ECMA-418-2:2025 (4th ed.)

compared with SQAT (its source)

Ported or adapted from SQAT. It was compared with the code it was ported from, which checks the port only.

• The project says that detailed validation and reference comparisons are kept outside the repository; the results are not published.

ecma418-torch new · ECMA-418-2:2025 (4th ed.)

compared with sottek-hearing-model and ArtemiS SUITE (HEAD acoustics)

• The 1 kHz tone amplitude-modulated at 70 Hz (60 dB SPL) gives 0.9999987 asper for 10 s (the test allows 1e-4 around 1), and the average specific roughness in band 17 is 0.37348, against 0.374 from ArtemiS v17 as quoted in the tests of sottek-hearing-model.
• Against sottek-hearing-model run as an oracle on three signals (the calibration tone, noise and a mixture of modulated tones and noise), the time-dependent values differ by at most 2e-5 asper and the single values by at most 2e-5; the README traces the residual to the rounded filter coefficients.
• The tests also check that two identical channels give the mono result and that input at 44.1 kHz gives the 48 kHz result within 0.2 %.

phonometry new · ECMA-418-2:2025 (4th ed.)

standard or paper data (calibration signal only)

• Checked at the calibration signal only.

rota-order new via sottek-hearing-model · ECMA-418-2:2025 (4th ed.)

own tests only

• Tests, which run only where sottek-hearing-model is installed (the project's continuous integration does not install it), check that the wrapper returns what a direct call of the package gives.

Mosqito.NET new · ECMA-418-2:2022 (2nd ed.)

compared with MoSQITo (its source)

Ported or adapted from MoSQITo. It was compared with the code it was ported from, which checks the port only.

• Every public function is compared with the output of Python MoSQITo for seven WAV files (three of them ISO 532-1 test signals), within 1 % (0.5 dB for levels); MoSQITo's unit tests are ported too.

pySQAT developing via sottek-hearing-model · ECMA-418-2:2025 (4th ed.)

compared with SQAT (also ported from refmap-psychoacoustics)

Code ported from the same source is expected to agree, so that comparison is not an independent check.

• Compared by validation/FINAL_VAL.py with the bundled SQAT v1.3 MATLAB code, which follows ECMA-418-2:2024 (3rd edition), so the comparison crosses editions.

Not stated by PsychoBox.

Before you compare numbers

Implementations of the same metric can give different numbers without either being wrong. Check these choices first.

• Edition. Compare implementations of the same edition: MoSQITo, Mosqito.NET and tools that call MoSQITo follow the 2nd edition, the others the 3rd or 4th.
• Overall value. The single value is the 90th percentile of the time-dependent roughness after the first 320 ms. NumPy's percentile (sottek-hearing-model) and MATLAB's prctile (RefMap, SQAT) interpolate differently, so single values can differ slightly.
• Calibration and time grid. SQAT uses a slightly adjusted calibration factor and a time grid that ends at the signal duration; MetaSona uses the standard's nominal factor (0.0180685) and its 20 ms grid.
• Binaural. Two channels can be combined into one binaural value; per-channel and combined values differ.

What the projects state:

SQAT

• The single value is the 90th percentile from MATLAB's prctile, which interpolates; the other statistics take the nearest rank (get_exceeded_value). The first time_skip seconds are left out; the code enforces at least 320 ms (it cites ECMA-418-2:2025, Section 7.1.8).

sottek-hearing-model

• Input at other sample rates is resampled to 48 kHz.
• Overall roughness and fluctuation strength are 90th percentiles computed with NumPy (linear interpolation, Hyndman–Fan type 7). The MATLAB code it was translated from uses prctile, which interpolates differently, so single values can differ slightly from RefMap and SQAT.

MetaSona new

• Expects calibrated sound pressure in pascals.
• The C functions require 48 kHz input; the Python interface resamples other rates to 48 kHz (SciPy resample_poly).
• time_skip_s (default 0 s) leaves out the start of the signal from the reported frames; the computation still runs over the whole signal. Time-varying results include every complete frame, labelled at frame centres, which differs from SQAT's handling of the last frame and of time stamps.

ecma418-torch new

• Input at other sample rates is resampled to 48 kHz (band-limited, by the Fourier method).
• The single value is the 90th percentile from torch.quantile (linear interpolation) of the time-dependent values, after the first 16 (roughness) or 36 (fluctuation strength) values at 50 Hz are left out.
• The outer and middle ear filter uses the coefficients as printed in Table 1 of the standard, where sottek-hearing-model uses unrounded ones.

rota-order new

• For the ECMA-418-2 metrics, vs_rpm takes by default the mean of the loudness and the tonality in each speed bin, the tonality only above 0.02 tu_HMS, and the 90th percentile of the roughness, after about 0.3 s (0.6 s for SQAT's tonality).
• Input in any unit is turned into sound pressure in Pa by an explicit calibration factor and resampled to 48 kHz.

Mosqito.NET new

• Time-varying loudness and ECMA-418-2 roughness resample input at any other rate to 48 kHz, and stationary loudness resamples rates below 48 kHz; MoSQITo rejects rates above 48 kHz for time-varying loudness.

References

1. ECMA-418-2:2020 (1st ed.). Psychoacoustic metrics for ITT equipment — Part 2 (models based on human perception). https://ecma-international.org/wp-content/uploads/ECMA-418-2_1st_edition_december_2020.pdf
2. ECMA-418-2:2022 (2nd ed.). Psychoacoustic metrics for ITT equipment — Part 2 (models based on human perception). https://ecma-international.org/wp-content/uploads/ECMA-418-2_2nd_edition_december_2022.pdf
3. ECMA-418-2:2024 (3rd ed.). Psychoacoustic metrics for ITT equipment — Part 2 (methods for describing human perception based on the Sottek Hearing Model). https://ecma-international.org/wp-content/uploads/ECMA-418-2_3rd_edition_december_2024.pdf
4. ECMA-418-2:2025 (4th ed.). Psychoacoustic metrics for ITT equipment — Part 2 (methods for describing human perception based on the Sottek Hearing Model). https://ecma-international.org/wp-content/uploads/ECMA-418-2_4th_edition_june_2025.pdf