psychoacoustics.loudness.moore_glasberg_time

Time-varying loudness per ISO 532-3:2023 (Moore-Glasberg-Schlittenlacher).

Clean-room implementation of the time-varying loudness model of ISO 532-3:2023, the extension of the stationary Moore-Glasberg method of ISO 532-2:2017 to sounds whose loudness changes over time. The procedure of Clause 7 turns a calibrated pressure waveform into two time histories:

• the short-term loudness S'(t) in sone (clause 7.8), the momentary loudness of a short segment such as a spoken word or a musical note; and
• the long-term loudness S''(t) in sone (clause 7.9), the loudness of a longer segment such as a whole sentence or musical phrase.

The signal chain is:

• a running short-term spectrum computed from six parallel Hann-windowed FFTs whose segment durations (2, 4, 8, 16, 32 and 64 ms) trade temporal against spectral resolution, each contributing only its own frequency range, updated every 1 ms (clause 7.3);
• the level-dependent rounded-exponential excitation pattern of ISO 532-2 sampled on the ERB-number grid from 1.75 Cam to 39 Cam in 0.25 Cam steps (clause 7.4, Formulae 1-6);
• the compressive transformation of excitation into instantaneous specific loudness N'(i) in sone/Cam (clause 7.5, Formulae 7-9, Tables 2-4, with $C=0.063$ sone/Cam);
• an attack/release temporal smoothing of the specific loudness at every centre frequency to the short-term specific loudness (clause 7.6, Formulae 10-13);
• the across-frequency smoothing and binaural inhibition of ISO 532-2 (clause 7.7, Formulae 14-17), and integration over the ERB-number scale to the short-term loudness of each ear and their binaural sum (clause 7.8);
• a slower attack/release temporal smoothing of the short-term loudness to the long-term loudness (clause 7.9, Formulae 18-21).

The loudness of a sound lasting up to about 5 s is well predicted by the maximum of the long-term loudness (clause 7.9); this n_max and the corresponding loudness level in phon (Table 5) are the headline results the standard asks to report (clause 9). A steady 1 kHz tone at 40 dB SPL presented binaurally in a free field yields a long-term loudness of 1.000 sone (40 phon) by definition of the sone; the additive spectral calibration of clause 7.3 (nominally +3.32 dB per component) is set so this anchor holds exactly.

Conformance mode for the sampling rate. Clause 5 prescribes converting the input to 32 kHz and clause 7.3 uses fixed 2048-point FFTs; this implementation deliberately processes at the native sampling rate and grows the FFT to max(2048, next_pow2(segment)) so the 64 ms Hann window is never truncated at 44.1/48 kHz (see test_low_freq_band_not_truncated_across_sample_rates). The spectral calibration _SPECTRAL_CAL_DB was fixed against the 32 kHz anchor; at other rates the Annex C.1 anchor reproduces with a 0.3 % cross-rate spread (a 0.5 s tone reads n_max = 0.9918 / 0.9926 / 0.9900 at 32 / 44.1 / 48 kHz; a settled 1.3 s tone 1.0000 / 1.0000 / 0.9982), far inside the standard’s 2.8 phon expanded uncertainty. These values are pinned by the test suite; resample to 32 kHz first if letter-of-standard processing is required.

Auto-generated from the source docstrings by scripts/generate_api_docs.py (make api-docs). Do not edit by hand.

loudness_moore_glasberg_time(
    signal: Signal | Sequence[float] | np.ndarray,
    fs: float | None = None,
    *,
    field: Literal['free', 'diffuse', 'eardrum'] = 'free',
    presentation: Literal['binaural', 'diotic', 'monaural'] = 'binaural',
    percentiles: Sequence[float] = (1.0, 5.0, 10.0, 50.0, 90.0, 95.0),
) -> MooreGlasbergTimeVaryingLoudness

Time-varying loudness of a calibrated pressure signal (ISO 532-3:2023).

Implements the Moore-Glasberg-Schlittenlacher method: a running short-term spectrum (six parallel FFTs, clause 7.3) feeds the ISO 532-2 excitation and specific-loudness model (clauses 7.4, 7.5), which is integrated over time to the short-term loudness S'(t) (clauses 7.6-7.8) and long-term loudness S''(t) (clause 7.9). The signal must be calibrated so its samples are the instantaneous sound pressure in pascals.

Parameters

Returns: A MooreGlasbergTimeVaryingLoudness.

A steady 1 kHz tone at 40 dB SPL, binaural, free field, yields a peak long-term loudness of 1.000 sone (40 phon) by definition of the sone.

MooreGlasbergTimeVaryingLoudness(
    time: np.ndarray,
    short_term_loudness: np.ndarray,
    long_term_loudness: np.ndarray,
    short_term_loudness_level: np.ndarray,
    long_term_loudness_level: np.ndarray,
    n_max: float,
    loudness_level_max: float,
    percentiles: dict[float, float],
    field: str,
    presentation: str,
)

Result of an ISO 532-3:2023 time-varying loudness calculation.

time is the frame time axis in seconds (1 ms spacing, clause 7.3). short_term_loudness and long_term_loudness are the binaural short-term S'(t) (clause 7.8) and long-term S''(t) (clause 7.9) loudness traces in sone; short_term_loudness_level and long_term_loudness_level are the corresponding loudness levels in phon (Table 5). n_max is the maximum of the long-term loudness (sone) - the predictor of the loudness of sounds up to about 5 s (clause 7.9) - and loudness_level_max the phon value it maps to. percentiles gives the long-term-loudness values (sone) exceeded for the stated fraction of the active trace (e.g. percentiles[5] is the level exceeded 5 % of the time); the standard itself reports only the peak long-term loudness (clause 9), the percentiles are provided as a convenience. field and presentation echo the listening conditions.

MooreGlasbergTimeVaryingLoudness.plot(
    ax: Axes | None = None,
    *,
    language: str = 'en',
    **kwargs: Any,
) -> Axes

Plot the short-term and long-term loudness against time.

Requires matplotlib (pip install phonometry[plot]); returns the Axes. See phonometry._plot.psychoacoustics.