Repository navigation
Conversation
…hness Three metrics of a one-channel SignalObj in Pa, each returning a dict of floats and NumPy arrays: - loudness_zwicker: ISO 532-1:2017, stationary and time-varying methods, free or diffuse field. N (sone), LN (phon), the specific loudness over 240 steps of 0.1 Bark and, time-varying, N(t), LN(t) and N'(t) every 2 ms with Nmax and N5. Port of SQAT's Loudness_ISO532_1 (BSD-3, Ella Manor's AARAE code after the C code of the standard). Signals not at 48 kHz are resampled as MATLAB's resample does (matches it to 2e-15). - sharpness_din45692: DIN 45692:2009, Formula (1) on that specific loudness, k = 0.11; S, or S(t) and S50 for the time-varying method. Written from the standard. - roughness_daniel1997: Daniel and Weber (1997), R(t), R'(t) over 47 half-Bark channels and Rmean (asper). Port of SQAT's Roughness_Daniel1997 (MIT, Dik Hermes et al.). The BSD-3 and MIT notices of the ported files are in the module header. Validation. Against SQAT on golden data from tests/golden/make_loudness_sqat.m and make_roughness_sqat.m (MATLAB R2026b, same float32 samples): stationary loudness within 3e-13, time- varying within 3.1e-8 sone (the nonlinear decay amplifies 3e-14 of filter rounding on the decay of a burst), roughness within 2.8e-5 asper (FFT rounding, 32 kHz input). ISO 532-1 Annex B: signal 1 (band levels) 83.2957 sone against 83.296; the standard's files of signals 2 to 5 within 0.034 sone (0.26 %); signals 6 to 25 keep N(t) within 5 % or 0.1 sone at every frame, max deviation 0.068 sone. DIN 45692: the 41 test signals of Tables A.2 and A.3 within 0.45 of the tolerance (worst 10.5 kHz band, -0.19 acum), reference noise 1.02 acum. 1 kHz at 40 dB gives 0.999976 sone; the reference AM tone 1.000001 asper. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
tonality_aures1985(signal, field='free', time_skip=0.0) gives the tonality K(t) and its mean in tonality units: the tonal components of Terhardt, Stoll and Seewann (1982) in Hann frames of 160 ms, weighted by bandwidth, frequency and level excess, and the loudness ratio WGr of the signal without and with them (Aures 1985). It is written from the two papers. SQAT's Tonality_Aures1985 is GPL-3.0 and served only as the test oracle. The private helper of the Terhardt excitation is shared with the roughness port, whose parity with SQAT is unchanged. Tests check the unit (a 1 kHz tone at 60 dB is 1 t.u.), the level excess weighting against its formula, parity with SQAT on golden data from tests/golden/make_tonality_sqat.m (tones within 3.4e-5 t.u., a tone in noise within 0.012 t.u.) and the rejection of invalid input. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…oacoustics fluctuation_strength_osses2016(signal, time_skip=0.0) gives FS(t) every 0.2 s over frames of 2 s, the specific fluctuation strength over 47 half-Bark channels and the mean FSmean, in vacil, after Osses, Garcia and Kohlrausch (2016): the roughness model of Daniel and Weber with frames of 2 s, a modulation weighting H(fmod), the generalised modulation depth compressed above 0.7 and FS = C_FS sum (m*)^1.7 |k k|^1.7 g(z). It is written from the paper. SQAT's FluctuationStrength_Osses2016 and its H(fmod) data are GPL-3.0 and served only as the test oracle. The paper does not give the filter type of H(fmod), so its cut-offs (1.75 and 13.5 Hz) were fitted to the AM tones of the paper's Table 1. The result differs from SQAT, as the docstring states: within 1 % for AM tones at 4 Hz, 4 % at 16 Hz, 14 % above for an FM tone, and about half of SQAT's value for AM at 1 Hz. Tests check the unit (1 vacil for the reference AM tone), the AM tones of Table 1 (rms deviation 0.09 vacil), the comparison with SQAT on golden data from tests/golden/make_fluctuation_sqat.m with the tolerances above, and the rejection of invalid input. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This was referenced Oct 31, 2026
Open
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Builds on #91 (and so on #79), so until those are merged this pull request also shows their commits.
Adds
fluctuation_strength_osses2016(signal, time_skip=0.0)topytta.psychoacoustics, the fluctuation strength model of Osses, Garcia and Kohlrausch (2016). It returns FS(t) every 0.2 s over frames of 2 s, the specific fluctuation strength over 47 half-Bark channels and the mean FSmean, in vacil.The model is the roughness model of Daniel and Weber, already in
roughness_daniel1997, with the changes of the paper:The function is written from the paper, because SQAT's
FluctuationStrength_Osses2016, with its H(fmod) data, is GPL-3.0. SQAT served only as the test oracle.The result differs from SQAT, and the docstring says so. The paper gives the passband of H(fmod) (3.1 Hz to 12 Hz) but not the filter type. Read as Butterworth cut-offs, those values miss the AM tones of the paper's own Table 1 by up to 0.5 vacil. The cut-offs used here, 1.75 Hz and 13.5 Hz, were fitted to that table, as the paper fits its filter, which gives C_FS = 0.2480 against the 0.2490 of the paper. SQAT takes H(fmod) from data files that the paper does not describe. Against SQAT:
The tests in
tests/test_psychoacoustics.pyprint the measured differences:tests/golden/make_fluctuation_sqat.m(MATLAB R2026b, same float32 samples), with tolerances of 1 %, 5 % and 15 %. The tone at 1 Hz is left out of this test and covered by test 2.Checked with Python 3.12, numpy 1.26 and scipy 1.17. The 7 failures of
development(3 intest_h5io.py, 4 intest_pyttasave.py) are the same with and without this change.Co-Authored-By: Claude Fable 5.1 noreply@anthropic.com