Why phonometry

phonometry is a standards-based acoustic measurement toolkit. Its differentiator is not the list of features but how they are built: every metric is implemented from the governing standard’s text, and the standard’s own reference values and acceptance limits are transcribed into the test suite and enforced in CI. This page explains that approach with a concrete case study — time weighting under IEC 61672-1:2013 — and summarizes what is conformance-tested today. The time-weighting analysis was originally published in issue #38.

Design philosophy: a case study in time weighting

Standard time weighting is defined as a continuous function of time via the differential equation

$$\tau \frac{dy(t)}{dt} + y(t) = x^2(t)$$

which corresponds to a stable first-order low-pass filter with a pole in the left half-plane ($s = -1/\tau$).

	phonometry	python-acoustics
Output	Continuous time-weighted envelope (one value per sample)	Stepped output (one value every τ seconds)
Input units	Raw sound pressure (Pa); squared internally	Energetic quantity (Pa²) expected as input
Filter	Stable exponential averaging (pole at $s=-1/\tau$)	Theoretically unstable design (pole in the right half-plane) stabilized by resetting the filter state every τ seconds
Behavior	True IEC time weighting	Closer to a block integrator ($L_{eq,\tau}$)

A pole on the negative real axis corresponds to a decaying exponential impulse response ($h(t) \propto e^{-t/\tau}$) — exactly what “exponential time weighting” means: past events are forgotten exponentially. A pole on the positive real axis grows without bound; block-resetting hides this but changes the measurement’s nature.

Verification against IEC 61672-1 (tone bursts)

The rigorous test for time weighting is the Tone Burst Response (IEC 61672-1, Table 4), using a 4 kHz sine burst referenced to the steady-state level.

phonometry results (FAST):

Burst duration	IEC target (dB)	phonometry (dB)	Error (dB)	Status
200 ms	−1.0	−0.98	+0.02	✅ PASS
50 ms	−4.8	−4.82	−0.02	✅ PASS
10 ms	−11.1	−11.14	−0.04	✅ PASS
1 ms	−20.9	−20.99	−0.09	✅ PASS

python-acoustics results (FAST, squared signal passed as required by that library):

Burst duration	IEC target (dB)	python-acoustics (dB)	Error (dB)	Status
200 ms	−1.0	−0.97	+0.03	✅ PASS
50 ms	−4.8	−3.93	+0.87	⚠️ FAIL
10 ms	−11.1	−10.90	+0.20	✅ PASS
1 ms	−20.9	−20.90	+0.00	✅ PASS

phonometry maintains high precision across all test cases. The block-based approach deviates significantly (> 0.8 dB) for the 50 ms burst because 125 ms blocks cannot resolve short transient events accurately — the result depends on how the burst aligns with the block boundaries.

Measured Fast envelopes (blue) matching the Table 4 reference values (dashed) within 0.1 dB for 200/50/10 ms bursts.

What this means in practice

• If you need standard-compliant Fast/Slow/Impulse envelopes (sound level meter behavior, one level per sample), use phonometry’s time_weighting.
• If you need block-averaged Leq per interval, that is a different, equally valid metric — you can compute it with leq over consecutive slices.
• Both approaches are useful; they just answer different questions. The discrepancy reported in issue #38 comes from comparing a continuous envelope against a block integrator, not from an implementation error.

Conformance testing across the library

The tone-burst case above is not an isolated check. For each standard the library implements, the reference values and acceptance limits are transcribed from the official text into the test suite, so any regression fails CI:

Standard	What is verified	Test file
IEC 61672-1:2013 Table 3	A/C/Z weighting at all 34 nominal frequencies, class 1 limits, at 48 and 96 kHz	tests/test_iec_weighting_table3.py
IEC 61672-1:2013 Table 4	F/S tone-burst responses (1 s to 1 ms) and the LAE column for sel()	tests/test_iec_compliance.py
IEC 61672-1:2013 Table 5	lc_peak() one-cycle/half-cycle peak responses, class 1 limits	tests/test_levels.py
IEC 61260-1:2014 Table 1	Filter-bank class 1/2 acceptance limits via verify_filter_class()	tests/test_compliance.py
ISO 7196:1995 Table 2	G weighting (infrasound) at every nominal response value, 0.25–315 Hz	tests/test_g_weighting.py
ISO 226:2023 Annex B	Equal-loudness contours, loudness levels and hearing threshold against the Annex B tables	tests/test_equal_loudness.py
ECMA-418-1:2024	TNR/PR tone prominence: critical bandwidths, proximity spacing and prominence criteria against the worked examples in clauses 10–12	tests/test_tonality.py
ISO 1996-1:2016	lden(), ldn() and composite_rating_level() against hand-computed formula values	tests/test_environmental.py
IEC 60942:2017 Table 2	Calibrator short-term stability limits (frequency-dependent, class 1) in calculate_sensitivity()	tests/test_calibration_validation.py

Beyond IEC 61252-style noise dose (sound_exposure(), lex_8h()), the same standards-first mindset shows up in the numerics: filter banks place their −3 dB points on the ANSI S1.11 / IEC 61260-1 band edges for every architecture (including Chebyshev II and Bessel, where scipy’s raw parametrization would not), and A/C weighting stays within class 1 tolerances up to 16 kHz at common audio rates via internal oversampling (see Frequency Weighting).

Where phonometry fits in the Python ecosystem

• python-acoustics was archived in February 2024 and is no longer maintained. Its comparison above reflects the last released code.
• acoustic-toolbox, the community successor to python-acoustics, depends on phonometry for its weighted level computations rather than reimplementing them.
• MoSQITo focuses on psychoacoustic sound-quality metrics (loudness, sharpness, roughness). It complements rather than overlaps phonometry: it does not cover sound level metrology (weighting filters, ballistics, Leq/SEL/Lden, calibration) and does not claim conformance testing against the standards’ tolerance tables.

If your work needs numbers you can defend against a standard’s tolerance table — measurement reports, environmental assessments, instrument cross-checks — that verification layer is what phonometry is for.