Skip to content

psychoacoustics: fluctuation strength

Every name on this page is published by phonometry.psychoacoustics: from phonometry import psychoacoustics, then psychoacoustics.<name>.

Fluctuation strength after Fastl & Zwicker / Osses et al.

Fluctuation strength F (unit: vacil) rates the slow loudness fluctuations of a sound. It has a band-pass dependence on modulation frequency peaking near 4 Hz, and its fixed point is defined (Fastl & Zwicker, Psychoacoustics: Facts and Models, Chapter 10) as 1 vacil for a 1 kHz tone at 60 dB, 100 % amplitude-modulated at 4 Hz. There is no ISO standard for fluctuation strength.

Two routes are provided:

  • fluctuation_strength_am_noise: the closed form (Fastl & Zwicker Eq. 10.2) for a sinusoidally amplitude-modulated broadband noise, exact in the modulation factor, level and modulation frequency.
  • fluctuation_strength: the signal model of Osses, García & Kohlrausch (2016), which estimates F from an arbitrary calibrated pressure signal. It sums specific contributions over 47 auditory filters, (Osses 2016 Eq. 1), where m* is a generalised modulation depth, k a cross-covariance between neighbouring bands and g(z) a frequency weighting.

The closed form and the 1-vacil calibration are exact. The signal model has no normative oracle; it is implemented clean-room from the Osses 2016 paper and cross-checked against the literature fluctuation-strength values of Fastl & Zwicker (Osses 2016 Table 1) and the open SQAT reference implementation. Like the ECMA-418-2 roughness, the model reproduces the reference trends and the 1-vacil anchor within a documented tolerance rather than to the last digit.

Front-end deviations from Osses 2016 (besides the re-fitted H(fmod) corners documented at _HP_LO): the paper analyses 2 s frames with 90 % overlap and 50 ms raised-cosine gating and screens components against the absolute hearing threshold; this implementation uses 50 % overlap, a Hann analysis window and relative floors (component magnitude > max/10^4, excitation level > max - 60 dB), with at the filter-bank edges. Measured consequences: a steady 1 kHz tone returns ~0.09 vacil instead of ~0 (analysis-window envelope leakage through the 2 Hz high-pass) and steady broadband noise returns ~0.15-0.19 vacil (partly physical level fluctuation of the noise itself).

ECMA-418-2:2025 (4th ed.) introduced a normative hearing-model fluctuation strength (its Clause 9, block size 65536); this module implements the Osses/Fastl & Zwicker model, not that method.

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

fluctuation_strength(
x: Signal | NumericArrayLike,
fs: float | None = None,
) -> FluctuationStrengthResult

Fluctuation strength of a calibrated signal (Osses 2016 model).

Estimates F in vacil from an arbitrary calibrated sound-pressure signal by the model of Osses, García & Kohlrausch (2016): an outer/middle-ear transmission, a 47-band excitation-pattern filter bank, the generalised modulation depth m* from the band-pass-filtered envelope, the cross-covariance k between neighbouring bands and the weighted sum of Eq. (1). The signal is analysed in 2 s frames; the overall value is the median across frames. The Osses 2016 model is defined for the free field only (its a0 transmission is the free-field characteristic); there is no diffuse-field variant, so this function takes no field argument.

Parameters

Name Description
x Calibrated sound-pressure signal (1-D), in Pa. Accepts a phonometry.io.Signal, which is where that calibration comes from without arithmetic: the model reads excitation levels, so an uncalibrated record is taken as if one digital unit were one pascal and F comes out wrong by however far that is from true.
fs Sample rate, in Hz (resampled to the model rate if needed).

Returns: A FluctuationStrengthResult.

Raises

Exception When
ValueError If the signal is invalid or fs is not positive.
TypeError if fs is not a single number, or x holds a text or a value that is not a real number.
fluctuation_strength_am_noise(
level: float,
modulation_factor: float,
mod_frequency: float,
) -> float

Fluctuation strength of AM broadband noise (Fastl & Zwicker Eq. 10.2).

in vacil, the closed form for a sinusoidally amplitude-modulated broadband noise of level L, modulation factor m and modulation frequency fmod. Exact; the result is clamped at 0 (the formula goes negative below ~20 dB or where the sensation vanishes).

Parameters

Name Description
level Broadband-noise level L, in dB.
modulation_factor Modulation factor m (0..1).
mod_frequency Modulation frequency fmod, in Hz.

Returns: Fluctuation strength, in vacil (>= 0).

Raises

Exception When
ValueError If modulation_factor is outside [0, 1] or mod_frequency is not positive/finite, or the level is not finite.
TypeError if level, mod_frequency or modulation_factor is not a single number.
FluctuationStrengthResult(
fluctuation_strength: float,
specific: NDArray[np.float64],
bark_axis: NDArray[np.float64],
time_dependent: NDArray[np.float64],
)

Fluctuation strength of a signal (Osses 2016 model).

Attributes

Name Description
fluctuation_strength Overall fluctuation strength F, in vacil (the median over the analysis frames).
specific Specific fluctuation strength f(z) per auditory filter, in vacil/Bark (averaged over frames), shape (47,).
bark_axis Centre critical-band rate z_i of each filter, in Bark, shape (47,).
time_dependent Fluctuation strength per analysis frame, in vacil.
FluctuationStrengthResult.plot(
ax: Axes | None = None,
*,
language: Literal['en', 'es'] = 'en',
**kwargs: Any,
) -> Axes

Plot the specific fluctuation strength against critical-band rate.

Parameters

Name Description
language The language of the labels: "en" or "es".