Impulsive-sound prominence (NT ACOU 112)

Standards: NT ACOU 112 ISO 1996 ISO/PAS 1996

Noise with prominent impulses (hammering, riveting, pile driving) is more annoying than a steady sound of the same equivalent level. NT ACOU 112:2002 (a Nordtest method) quantifies how prominent an impulse is and turns it into a graduated adjustment $K_{I}$ added to the measured $L_{A e q}$ . The prominence is read from two properties of each impulse’s onset in the A-weighted, time-weighting-F level history: how fast it rises (the onset rate) and how far it rises (the level difference).

1. Predicted prominence (clause 7)

For each candidate impulse the predicted prominence $P$ combines the onset rate (in dB/s) and the level difference (in dB) on a logarithmic scale (Formula 1):

P = 3 lo g_{10} (onset rate) + 2 lo g_{10} (level difference) .

The coefficients were fitted to listening tests and $P$ is designed to peak around 15 for very sudden, loud impulses. The impulse with the highest $P$ over a 30-minute period governs. A level rise only qualifies as an impulse when its onset rate exceeds 10 dB/s (clause 4.5; clause 8 applies the adjustment “for sounds with onset rates larger than 10 dB/s” only): impulse_prominence marks non-qualifying events in its qualifies mask, warns about them, and never lets them set the governing prominence or a $K_{I}$ (the adjustment is 0 dB when no event qualifies).

Download the animation (WebM)

from phonometry import environment

# Three candidate impulses: (onset rate dB/s, level difference dB).
result = environment.impulse_prominence([1200.0, 300.0, 60.0], [32.0, 18.0, 11.0])
print(result.per_impulse.round(2))  # [12.25  9.94  7.42]
print(round(result.prominence, 2))  # 12.25  (the governing impulse)
print(round(result.adjustment, 2))  # 13.05  dB

result.plot()   # the KI(P) curve of section 2 with these impulses marked

A single impulse can be evaluated directly with predicted_prominence:

from phonometry import environment

# P = 3*lg(1000) + 2*lg(30) = 9 + 2.95 = 11.95.
print(round(environment.predicted_prominence(1000.0, 30.0), 4))  # 11.9542

2. Adjustment to LAeq (clause 8)

Below a prominence of 5 the impulse is not audible enough to matter; above it the adjustment grows linearly (Formula 2):

K_{I} = 1.8 (P - 5) dB (P > 5), K_{I} = 0 (P \leq 5) .

from phonometry import environment

print(float(environment.impulse_adjustment(10.0)))  # 9.0 dB
print(float(environment.impulse_adjustment(5.0)))   # 0.0 dB (at the threshold)

The adjustment is applied to $L_{A e q, 30 min}$ from the single event with the highest prominence. The rating level over a longer reference time combines the impulse-adjusted equivalent levels of the sub-intervals (clause 8, Note 1):

L_{A r, T} = 10 lo g_{10} (\frac{1}{T} N \sum Δ t_{N} 1 0^{(L_{A e q, N} + K_{I, N}) /10}) .

from phonometry import environment

# Two 30-min periods: one impulsive (KI = 7.6 dB), one quiet.
print(round(environment.rating_level([72.0, 66.0], [7.6, 0.0], [30.0, 30.0], 60.0), 2))
# 76.78 dB

The ImpulseProminenceResult carries the per_impulse prominences, the governing prominence and its adjustment, and its .plot() draws the $K_{I} (P)$ curve with the impulses marked. The method is a supplement to the environmental-noise measurement of ISO 1996-2.

Left: the predicted prominence P rising with onset rate on a log axis for level differences of 5, 15 and 30 dB, reaching about 15 for a sudden loud impulse. Right: the adjustment KI, flat at zero up to the threshold P = 5 then linear, with three impulses marked and the governing one at KI = 13 dB

Both inputs matter but not equally: at a fixed onset rate, going from a 5 dB to a 30 dB level difference moves the prominence far less than an order of magnitude of onset rate does. Below P = 5 the adjustment is identically zero, so an impulse that does not qualify costs nothing at all — which is why the qualification test is the part of the method to get right.

Show the code for this figure

import numpy as np
import matplotlib.pyplot as plt
from phonometry import environment

# One line for the adjustment curve with the impulses marked:
environment.impulse_prominence([1200.0, 300.0, 60.0], [32.0, 18.0, 11.0]).plot()
plt.show()

# By hand, the left panel — P vs onset rate for three level differences:
orate = np.logspace(1, 4, 200)
for ld in (5.0, 15.0, 30.0):
    plt.plot(orate, environment.predicted_prominence(orate, np.full_like(orate, ld)),
             label=f"LD = {ld:g} dB")
plt.xscale("log"); plt.legend(); plt.show()

3. Objective measurement from a signal (ISO/PAS 1996-3)

NT ACOU 112 takes the onset rate and level difference as inputs. ISO/PAS 1996-3:2022 keeps the same prominence and adjustment formulae but adds the objective measurement chain that reads those quantities straight from a calibrated recording. impulsive_sound_adjustment A-weights the signal, applies time weighting F ( $τ = 125 ms$ ), samples the level history $L_{p A F}$ every 10-25 ms, detects each onset (the contiguous stretch whose gradient exceeds 10 dB/s, merging events less than 50 ms apart), measures its level difference $L_{D} = L_{e} - L_{s}$ and its least-squares onset rate $OR$ , and returns the governing adjustment $K_{I}$ with the source category of clause 7: not impulsive ( $K_{I} = 0$ ), regular impulsive ( $0 < K_{I} \leq 5$ ) or highly impulsive ( $K_{I} > 5$ ).

The onset detection can be exercised on a level history directly with detect_onsets, which is convenient for meters that already log $L_{p A F}$ :

import numpy as np
from phonometry import environment

# An LpAF history sampled every 20 ms: quiet, a 30 dB rise over 0.30 s, steady.
dt = 0.02
levels = np.concatenate([
    np.full(10, 40.0),
    40.0 + 30.0 * np.arange(1, 16) / 15,   # a straight 100 dB/s onset to 70 dB
    np.full(15, 70.0),
])
onset = environment.detect_onsets(levels, dt)[0]
print(round(onset.onset_rate), round(onset.level_difference))  # 100 30
print(round(onset.prominence, 2))                              # 8.95

What the recording has to be. The chain reads a level history, so the conditions that make one valid are the conditions of the whole method: a class 1 measurement chain calibrated before and after the session, and a signal in pascals, because although the adjustment itself is calibration-insensitive (it is built from level differences) the reported and adjusted $L_{A e q}$ are not. The level history has to be sampled fast enough to resolve the rise — the method works on a 10 ms to 25 ms grid, which is the dt argument below — and the governing impulse is the one with the highest $P$ over a 30-minute assessment period. Three things invalidate a period rather than lowering its score: no onset qualifying at the 10 dB/s gradient at all (the adjustment is then 0 dB and the period says nothing about the source), events overlapping so that one impulse starts on the decay tail of the previous one, and a level history that saturated or clipped at either end of the range.

From a calibrated time signal (in pascal) the whole chain runs end to end:

result = environment.impulsive_sound_adjustment(signal, fs)
print(result.category)                  # e.g. 'highly impulsive'
print(round(result.adjustment, 1))      # KI in dB (0.0 to about 9 dB in typical cases)
print(round(result.adjusted_laeq, 1))   # LAeq + KI

result.plot()   # the LpAF history with the detected onsets (needs matplotlib)

Three keyword arguments of that call decide whether the number is defensible, and none of them is optional in practice:

Argument	Default	What it does
`dt`	`0.02` s	The logging interval of the level history. The method needs 10 ms to 25 ms; a coarser grid flattens the fitted onset rate and under-reports $P$ , which is the caveat two paragraphs down made settable
`calibration_offset`	`0.0` dB	Added to the level history to bridge a digital scale to a calibrated one: the difference between the calibrator’s known level and the level this same processing reports for the calibration recording. It moves $L_{A e q}$ and the adjusted $L_{A e q}$ and leaves $K_{I}$ alone
`laeq`	`None`	Supply the $L_{A e q}$ measured over the assessment period instead of letting it be computed from the excerpt, which is what a real report does
`reference_pressure`	`2e-05` Pa	Only change it for a non-air medium
`onset_rate_method`	`'least_squares'`	The fit over the detected rise

environment.sound_pressure_level_history(signal, fs, *, dt=..., calibration_offset=...) is the same Clause 4 step on its own — A-weighting, F time weighting and resampling onto the 10-25 ms grid — for a caller who wants the history without the adjustment. It warm-starts the F integrator at the first window’s mean square, so the beginning of a record does not register as an onset of the processing itself.

A-weighted Fast level history of three hammer strikes over a 55 dB(A) background across six seconds: each strike rises from about 52 dB to 89 dB, the detected onset start and end points are marked with the least-squares onset line, the governing level difference of 36.8 dB is annotated, and the title reports a prominence of 11.34 with an adjustment of 11.42 dB, category highly impulsive

The three strikes are deliberately well separated, 1.6 s apart against a 0.08 s decay, so each onset starts from the same background and each fit has the whole rise to itself. That is the easy case: the two field errors described below both show up as a shallower fitted line on a visibly steep edge.

Show the code for this figure

import numpy as np
import matplotlib.pyplot as plt
from phonometry import environment

# Three hammer strikes over a 55 dB(A) background, 6 s at 48 kHz.
fs = 48000
rng = np.random.default_rng(7)
t = np.arange(int(6.0 * fs)) / fs
background = rng.standard_normal(t.size)
background *= 2e-5 * 10 ** (55 / 20) / np.sqrt(np.mean(background ** 2))
signal = background.copy()
for onset_time in (1.0, 2.6, 4.2):
    decay = np.exp(-(t - onset_time) / 0.08) * (t >= onset_time)
    strike = decay * rng.standard_normal(t.size)
    window = (t >= onset_time) & (t < onset_time + 0.1)
    strike *= 2e-5 * 10 ** (95 / 20) / np.sqrt(np.mean(strike[window] ** 2))
    signal += strike

res = environment.impulsive_sound_adjustment(signal, fs)
print(res.category, round(res.prominence, 2), round(res.adjustment, 2))
# highly impulsive 11.34 11.42

# One line: the level history with the onsets, the fitted onset lines and
# the governing level difference.
res.plot()
plt.show()

Two things are worth reading off the plot. The onset lines are the least-squares fits over the detected rise, so a shallow fit on a visibly steep edge means the level history was logged too coarsely to resolve it. And the level difference is measured from the level just before the onset, not from the long-term background, which is why a second strike arriving on the decay tail of the first one scores a smaller $L_{D}$ than it appears to deserve.

Because the onset rate and level difference are level differences, the adjustment is insensitive to the absolute calibration of the meter (clause 8); only the reported $L_{A e q}$ and the adjusted $L_{A e q}$ depend on it. The scope of the document states that the adjustment typically falls between 0,0 dB and 9,0 dB; the formula itself is not capped, so a very sudden, loud impulse can exceed that range. ImpulsiveSoundResult.plot() draws the $L_{p A F}$ history with the detected onsets, the least-squares onset lines and the governing level difference marked.

4. Where the method sits among the standards

Rating standards traditionally handle impulsiveness by category, not by measurement. ISO 1996-1:2016 (Table A.1) adds a fixed 5 dB when the source is regular impulsive and 12 dB when it is highly impulsive, and leaves the choice of category to the assessor’s judgement. NT ACOU 112 replaces that judgement with a measurement: the onset rate and level difference of the actual impulses produce a graduated $K_{I}$ that runs from 0 dB for barely perceptible onsets to about 18 dB for the most sudden, loud ones, crossing the two conventional figures on the way. The approach proved durable: the same onset-based prominence was later carried into ISO/PAS 1996-3:2022, which uses it to decide objectively whether a source counts as regular or highly impulsive in the ISO 1996-1 rating.

Two practical caveats when applying it. The onset is read from the A-weighted, time-weighting-F level history, and the 125 ms time constant of weighting F itself limits how fast a recorded level can rise: very short impulses arrive already smoothed, so the level history must be logged at the meter’s full output rate (not in coarse intervals); otherwise the onset rate, and with it $P$ , is underestimated. And the method rates the prominence of the impulses, not their energy: $K_{I}$ rides on top of the measured $L_{A e q}$ in the rating level, it never replaces it.

5. Assessment report (`.report()`)

An assessment report has to carry more than the fiche computes. State the standard applied and the edition; the source and the situation; the measurement position and the assessment period; the governing impulse with its onset rate and level difference; the prominence $P$ and the adjustment $K_{I}$ ; the source category the result puts the source in; and the $L_{A e q}$ the adjustment rides on, with the instrumentation and calibration behind it. The fiche below fills the computed half automatically — the per-impulse table, the governing $P$ , $K_{I}$ and the category — and takes the rest from ReportMetadata, which is why the metadata block is not decoration: without it the sheet documents an arithmetic result and not a measurement.

ImpulseProminenceResult.report(path) renders a one-page PDF fiche laid out like an impulsive-sound assessment report of an environmental-noise laboratory, following NT ACOU 112:2002 (carried into ISO/PAS 1996-3:2022): a standard-basis line, an optional metadata header block (source/situation, client, measurement position, instrumentation and date, with the 30-minute assessment period always shown), a full-width per-impulse table (onset rate, level difference, predicted prominence $P$ and whether the onset qualifies as an impulse) above the adjustment-curve plot $K_{I} (P)$ with the candidate impulses marked, the boxed governing prominence $P$ together with the derived $L_{A e q}$ adjustment $K_{I}$ (Formula 2), an optional PASS/FAIL verdict row and a prominence-category note, and a footer with the fixed disclaimer.

It uses the same ReportMetadata container and rendering engine as the ISO 532-1 loudness fiche; a supplied requirement is read as the maximum acceptable governing prominence $P$ (a less prominent impulse passes). Rendering needs reportlab and, for the figure the fiche embeds, matplotlib (pip install "phonometry[report,plot]"); only engine="reportlab" is supported. The fiche renders in English by default; pass language="es" for a Spanish fiche (translated fixed strings and a comma decimal separator), e.g. res.report("impulse_fiche_es.pdf", language="es").

from phonometry import environment, ReportMetadata

# The three-impulse pile-driving set from §1 (onset rate dB/s, level diff. dB).
res = environment.impulse_prominence([1200.0, 300.0, 60.0], [32.0, 18.0, 11.0])
res.report(
    "impulse_fiche.pdf",
    metadata=ReportMetadata(
        specimen="Pile-driving site, intermittent hammering",
        measurement_standard="ISO 1996-2",
        laboratory="Phonometry Reference Laboratory",
        requirement=10.0,           # maximum acceptable governing prominence P
    ),
)                                   # governing P and the LAeq adjustment KI (dB)

The example fiche is regenerated with make reports and kept rendered in the repository; click the preview to open the PDF.

NT ACOU 112 impulsive-sound prominence example report (PDF) — Impulsive-sound prominence fiche (ImpulseProminenceResult.report), governing prominence P with the LAeq adjustment KI.

What this guide covers

Covered
NT ACOU 112:2002: the onset definition (clauses 4.5 to 4.7), the predicted prominence $P$ (clause 7, Formula 1), the graduated adjustment $K_{I}$ (clause 8, Formula 2) and the rating level over a reference time (clause 8, Note 1), through impulse_prominence, predicted_prominence, impulse_adjustment and rating_level. ISO/PAS 1996-3:2022 is covered for the objective measurement chain built on the same prominence formula: $L_{p A F}$ sampling, onset detection and merging, and the least-squares onset rate and level difference, through impulsive_sound_adjustment and detect_onsets.
Not covered
NT ACOU 112’s Note 4 (clause 7) records an older Nordic guideline: a flat 5 dB adjustment based on subjective judgement, recommended when $K_{I} > 3$ for noise characteristic of working operations, offered as a fallback where the graduated method cannot be applied. Only the graduated $K_{I}$ of Formula 2 is implemented; the flat 5 dB alternative is not. ISO 1996-1’s Table A.1 category adjustments (5 dB regular, 12 dB highly impulsive) are covered only as the assessor’s-judgement baseline that this measurement replaces, not as a function of their own.

References

International Organization for Standardization. (2016). Acoustics — Description, measurement and assessment of environmental noise — Part 1: Basic quantities and assessment procedures (ISO 1996-1:2016). The rating framework and the Table A.1 category adjustments (5 dB regular, 12 dB highly impulsive) that the measured prominence calibrates against.
International Organization for Standardization. (2022). Acoustics — Description, measurement and assessment of environmental noise — Part 3: Objective method for the measurement of prominence of impulsive sounds and for adjustment of LAeq (ISO/PAS 1996-3:2022). The ISO successor built on the NT ACOU 112 onset-rate prominence.
Nordtest. (2002). Acoustics: Prominence of impulsive sounds and for adjustment of LAeq (NT ACOU 112:2002). The implemented method, freely downloadable from the Nordtest archive: the predicted prominence (clause 7, Formula 1), the adjustment to LAeq (clause 8, Formula 2) and the rating level (clause 8, Note 1), with the onset defined in clauses 4.5-4.7.

Created and maintained by José Manuel Requena Plens ORCID All projects