# Objective audibility of tones in noise (ISO/PAS 20065)

> One page of the phonometry documentation, as Markdown. Index: https://docs.jmrp.io/phonometry/llms.txt

Canonical: https://docs.jmrp.io/phonometry/perception/psychoacoustics/tone-audibility/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/perception/psychoacoustics/tone-audibility/index.md
Version: phonometry 4.0.0
Install: pip install "phonometry[full]"
Author: José Manuel Requena Plens (https://orcid.org/0000-0003-1250-6212)
Published: 2026-07-11
Updated: 2026-11-16T21:36:14Z
Standards: ISO/PAS 20065:2016; ISO 1996-2:2017; DIN 45681:2005-03
Verified: 12 conformance checks (ISO/PAS 20065:2016; ISO 1996-2:2017) https://docs.jmrp.io/phonometry/reference/conformance/
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: ISO/PAS 20065:2016 audibility ΔL of a tone above its masking threshold: critical band, masking, peak detection and the ISO 1996-2 adjustment.

---

A steady tone embedded in broadband noise stands out when it rises audibly above
the noise that would otherwise mask it, the objective precondition for the tonal
penalties applied in noise assessment. **ISO/PAS 20065:2016** is the *engineering
method* that quantifies this audibility: from a narrow-band FFT spectrum it
derives, for every prominent tone, the **audibility** $\Delta L$: how many
decibels the tone level exceeds the masking threshold of the surrounding
noise. (Whether a tone is *annoying* is a separate, downstream rating
judgement.) It is the detailed method that **ISO 1996-2:2017** defers to (the
simpler Annex C route lives in
[environmental measurement](https://docs.jmrp.io/phonometry/environment/assessment/environmental-levels/)); the
mean audibility $\Delta L$ it produces feeds the ISO 1996-2 tonal adjustment
$K_\mathrm{t}$.

![Per-tone audibility ΔL of the nine tones of the ISO/PAS 20065 Annex E combustion-engine spectrum, with the decisive tone at 137.3 Hz highlighted and the ΔL = 0 dB threshold marked](https://docs.jmrp.io/phonometry/media/tone_audibility.svg)

*The nine tones of the ISO/PAS 20065 Annex E combustion-engine spectrum, each
evaluated in its own critical band. A bar above the dashed $\Delta L = 0$ dB
line is audible over the masking noise around it, and the tallest of them
(the 137.3 Hz tone at 5.0 dB) is the largest *per-tone* audibility of this
spectrum. It is not yet the spectrum's decisive audibility: §3 shows why the
standard's own answer is 9.18 dB.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
from phonometry import psychoacoustics

# ISO/PAS 20065 Annex E combustion-engine spectrum 1: nine tones (fT, LT, LS)
# from a narrow-band spectrum with line spacing 2.7 Hz
fT = [118.4, 137.3, 158.8, 314.9, 433.4, 592.2, 629.8, 643.3, 1582.7]
LT = [64.56, 67.96, 68.63, 68.50, 73.17, 78.31, 75.00, 79.75, 71.07]
LS = [48.91, 49.22, 50.50, 52.85, 58.29, 59.53, 59.71, 61.98, 54.16]

res = psychoacoustics.assess_tones(LT, LS, 2.7, tone_frequencies=fT)
print(round(res.decisive_audibility, 2), res.decisive_frequency)  # 5.01 137.3
res.plot()   # per-tone audibility bars, decisive tone highlighted
plt.show()
```

## 0. The spectrum this method needs

Every later section consumes a narrow-band spectrum, and ISO/PAS 20065 is
specific about which one. A reader can follow this page end to end and produce
an audibility the standard would reject, because the analyser was set up
differently: the arithmetic gives no sign of it.

**Analyser settings (clause 4.2).** A class 1 chain per IEC 61672-1 with a
lower limiting frequency at or below 20 Hz, an anti-aliasing filter ahead of
the digitisation, and an amplitude resolution of at least 0.1 dB. The constant
line spacing $\Delta f$ shall lie between 1.9 Hz and 4.0 Hz inclusive: the
2.7 Hz of the Annex E example sits in the middle of that window, which is what
`assess_tones(LT, LS, 2.7, ...)` carries as its line spacing. The Hanning window is
*mandatory*, and it is where the −1.76 dB bandwidth correction of §2 comes
from: another window has another effective bandwidth, so that constant, and
with it every level in the chain, would be wrong.

**A-weighting, before anything else (clause 5.3.2).** The method is defined on
an A-weighted narrow-band spectrum; the library is weighting-agnostic and
applies nothing, so the caller weights the lines first. It matters because the
audibility is a difference of two levels taken over one critical band: a
weighting flat across the band would cancel out, but the A-curve is steepest
exactly where tonal complaints live. At the 137.3 Hz tone of the example it is
about −15 dB, and it climbs nearly 9 dB across that tone's own critical band
(−19.8 dB at 95.7 Hz to −11.0 dB at 197.0 Hz), so both $L_\mathrm{T}$ and $L_\mathrm{S}$ move,
and by different amounts. The Annex E numbers on this
page are already A-weighted, which is the tell-tale: anyone reproducing them
from an unweighted analyser will find low-frequency tones reading high.

**Averaging (clauses 4.3 and 5.1).** The analyser's own basic spectra are
shorter than a second at this line spacing, so they are merged line by line
(Formula 1) into spectra of approximately 3 s. That 3 s is a floor, not a
preference: clause 5.1 states that shorter averaging yields unjustified
audibilities, too high *and* too low. At least 12 time-staggered 3 s spectra
are then taken, because that is what generally keeps the extended uncertainty
inside ±1.5 dB, and every alternating operating state of the source must appear
among them.

![The two-level time structure of an ISO/PAS 20065 assessment: a 36 second record split into idle, full load and idle operating states; the analyser's sub-second basic spectra merged line by line into twelve merged spectra of 3 seconds each; each merged spectrum giving one decisive audibility, and the twelve energy-averaged by Formula 20 into the mean audibility that becomes the tonal adjustment Kt; a footer lists the class 1 chain, the 1.9 to 4.0 hertz line spacing, the mandatory Hanning window, the 0.1 decibel amplitude resolution, the A-weighting and the uncertainty rule](https://docs.jmrp.io/phonometry/media/diagram_tone_audibility_acquisition.svg)

*The time structure behind a defensible tonal penalty. The library starts at
the third band: it takes merged spectra and returns audibilities, so the first
two bands are the caller's responsibility.*

**When the method does not apply.** A source whose level or tone frequency
moves too fast to be represented by a 3 s average cannot be assessed this way
at all: start-up transients, strongly speed-varying machinery. Clause 5.1 puts
that exclusion in the standard rather than leaving it to judgement.

**What the report has to carry** (clause 7): date and place; a description of
the measurement environment with the source and measurement positions and a
sketch of the surroundings; air temperature, air pressure and relative
humidity; mean wind speed and direction; the make, model and serial number of
every instrument; the line spacing and the frequency range investigated; for
every spectrum with a decisive audibility above zero, the tone frequencies and
their audibilities; the mean audibility; the extended uncertainty whenever
fewer than 12 spectra were averaged; and a plot of the narrow-band levels of
the 3 s spectrum with the greatest $\Delta L$. `ToneAudibilityResult.report()`
prints the acoustic half of that list and embeds that plot; the environment and
instrument halves arrive through `ReportMetadata`.

## 1. The critical band about the tone

Each tone of frequency $f_\mathrm{T}$ is evaluated inside a critical band whose width
is (Formula 2)

$$
\Delta f_\mathrm{c} = 25.0 + 75.0\left(1.0 + 1.4\left(\tfrac{f_\mathrm{T}}{1000}\right)^{2}\right)^{0.69}\ \mathrm{Hz}.
$$

With a geometric placement of the corner frequencies about the tone
(Formulae 3–5), $\sqrt{f_1 f_2} = f_\mathrm{T}$ and $f_2 - f_1 = \Delta f_\mathrm{c}$, so
$f_1 = -\Delta f_\mathrm{c}/2 + \sqrt{\Delta f_\mathrm{c}^2 + 4 f_\mathrm{T}^2}/2$ and
$f_2 = f_1 + \Delta f_\mathrm{c}$.

```python
from phonometry import psychoacoustics

print(round(psychoacoustics.critical_bandwidth_engineering(137.3), 2))   # 101.36 Hz
f1, f2 = psychoacoustics.critical_band_corners(137.3)
print(round(f1, 2), round(f2, 2))                            # 95.67 197.04
```

## 2. Audibility of a tone

The mean narrow-band level $L_\mathrm{S}$ of the masking noise (Formula 6, an iterative
energy average of the lines in the critical band) and the tone level $L_\mathrm{T}$
(Formula 8, the energy sum of the tonal lines) are derived from the narrow-band
spectrum; `mean_narrowband_level` and `tone_level` do this directly (see §4).
The critical-band level of the masking noise spreads $L_\mathrm{S}$ over the critical
bandwidth (Formula 12), the masking index accounts for the ear (Formula 13) and
the audibility is their difference (Formula 14):

$$
L_\mathrm{G} = L_\mathrm{S} + 10\log_{10}\!\frac{\Delta f_\mathrm{c}}{\Delta f}, \qquad
a_\mathrm{v} = -2 - \log_{10}\!\Big[1 + \big(\tfrac{f}{502}\big)^{2.5}\Big], \qquad
\Delta L = L_\mathrm{T} - L_\mathrm{G} - a_\mathrm{v} .
$$

A supplied tone is *audible* when $\Delta L > 0$. $\Delta f$ is the line
spacing (frequency resolution); the energy sums over $K > 1$ lines carry a
window correction of $10\log_{10}(\Delta f/\Delta f_\mathrm{e})$ (−1.76 dB for the
recommended Hanning window, $\Delta f_\mathrm{e} = 1.5\,\Delta f$, Formula (8)), while
a single-line tone ($K = 1$) takes its level unchanged (Formula (7), no
bandwidth correction).

```python
from phonometry import psychoacoustics

# ISO/PAS 20065 Annex E, tone at 137.3 Hz (Δf = 2.7 Hz):
#   LS = 49.22 dB (Formula 6), LT = 67.96 dB (Formula 8).
print(round(psychoacoustics.tone_audibility(67.96, 49.22, 137.3, 2.7), 2))   # 5.01 dB
print(round(psychoacoustics.masking_index(137.3), 2))                        # -2.02 dB
```

The whole method is one chain from the spectrum to the penalty, and every
intermediate quantity above is a stop on it. The diagram walks the Annex E
tone through that chain, down to the $K_\mathrm{t}$ the mean audibility earns.

![Flow diagram of the ISO/PAS 20065 engineering method on the Annex E combustion-engine spectrum: a narrow-band FFT spectrum with 2.7 hertz line spacing yields a tone at 137.3 hertz, its critical band of 101.36 hertz spans 95.67 to 197.04 hertz, the masking noise level of 49.22 decibels and tone level of 67.96 decibels give a masking threshold of 64.97 decibels with a masking index of minus 2.02 decibels, and the audibility is 5.01 decibels, the decisive value of the spectrum; a closing note maps the 6.98 decibel energy-mean audibility of the five spectra to a tonal adjustment Kt of 4 decibels through ISO 1996-2 Table J.1](https://docs.jmrp.io/phonometry/media/diagram_tone_audibility.svg)

## 3. Decisive and mean audibility

The **decisive** audibility of one narrow-band spectrum is the largest tone
audibility in it (clause 5.3.8). Over $J$ staggered spectra the **mean
audibility** is their energy mean (Formula 20); a spectrum in which no tone is
found contributes $\Delta L_j = -10\ \text{dB}$ (Formula 21). `assess_tones` is
the *per-tone* entry point: it applies Formulae (2)-(5) and (12)-(14) to each
tone it is given and reports the largest, which is the decisive audibility only
when no two audible tones share a critical band.

**How the decisive band is selected.** The method does not scan a fixed set of
bands: each detected *tone* defines its own critical band (§1), the audibility
is evaluated tone by tone, and, after Step 3 has merged audible same-band
tones into `FG` groups rated at their most audible member (§5), the decisive
audibility is simply the largest $\Delta L$ left standing (Step 4). The "decisive
band" is therefore the critical band centred on whichever tone or group wins,
and it is free to move from spectrum to spectrum as the source runs through
its operating states; the energy mean of Formula 20 then lets the loudest
(most audible) spectra dominate the reported value, which is deliberate: a
tone that is clearly audible part of the time is not excused by intervals in
which it disappears.

```python
from phonometry import psychoacoustics

# Annex E combustion-engine spectrum 1: nine tones (fT, LT, LS), Δf = 2.7 Hz.
fT = [118.4, 137.3, 158.8, 314.9, 433.4, 592.2, 629.8, 643.3, 1582.7]
LT = [64.56, 67.96, 68.63, 68.50, 73.17, 78.31, 75.00, 79.75, 71.07]
LS = [48.91, 49.22, 50.50, 52.85, 58.29, 59.53, 59.71, 61.98, 54.16]
res = psychoacoustics.assess_tones(LT, LS, 2.7, tone_frequencies=fT)
print(round(res.decisive_audibility, 2), res.decisive_frequency)  # 5.01 137.3

# The five spectra of the Annex E campaign, rated with Step 3 applied
# (Table E.3 decisive values), energy-averaged by Formula 20:
print(round(psychoacoustics.mean_audibility([9.18, 6.04, 7.46, 2.67, 7.17]), 2))  # 6.98 dB

res.plot(view="levels")   # tone levels above their critical-band masking noise
```

The two numbers in that block are not the same spectrum rated twice by the same
rule. The tones at 118.4, 137.3 and 158.8 Hz all fall in one critical band, so
clause 5.3.8 Step 3 merges them into an `FG` group at their combined level
$L_\mathrm{T} = 72.15$ dB, and the standard's decisive value for spectrum 1 is the
9.18 dB that opens the `mean_audibility` list, not the 5.01 dB the per-tone
call prints, which is the largest of the nine individual audibilities
`[0.75, 0.89, 1.02, 1.79, 1.93, 3.47, 4.39, 4.57, 5.01]`. Those 4.2 dB survive
the averaging: the per-tone value in place of 9.18 pulls the campaign mean from
6.98 dB to 5.98 dB, which is $K_\mathrm{t} = 3$ dB instead of 4 dB in ISO 1996-2
Table J.1. So use `analyze_spectrum` (§5), which runs the grouping,
whenever a decisive audibility is going to be reported, and `assess_tones` only
when the grouping has already been resolved.

The same assessment reads two ways. The audibility bars at the top of this
page answer "how far above the masking threshold is each tone"; the levels
view answers "what did the analyser see", which is the view an assessment
report has to defend:

![Tone levels and critical-band masking noise of the ISO/PAS 20065 Annex E combustion-engine spectrum on a logarithmic frequency axis from about 96 Hz to 1.8 kHz: each tone is a stem from its critical-band masking level up to its tone level, and the decisive 137.3 Hz tone is highlighted with its critical band shaded](https://docs.jmrp.io/phonometry/media/tone_audibility_levels.svg)

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
from phonometry import psychoacoustics

# The Annex E combustion-engine spectrum of the snippet above.
fT = [118.4, 137.3, 158.8, 314.9, 433.4, 592.2, 629.8, 643.3, 1582.7]
LT = [64.56, 67.96, 68.63, 68.50, 73.17, 78.31, 75.00, 79.75, 71.07]
LS = [48.91, 49.22, 50.50, 52.85, 58.29, 59.53, 59.71, 61.98, 54.16]
res = psychoacoustics.assess_tones(LT, LS, 2.7, tone_frequencies=fT)

# One line: the levels view, the same one the .report() fiche embeds.
res.plot(view="levels")
plt.show()

# The default view is the per-tone audibility instead:
res.plot()      # or res.plot(view="audibility")
plt.show()
```

Reading it left to right: each horizontal segment is the critical-band
masking-noise level $L_\mathrm{G}$ drawn across the band it applies to, each marker is
the tone level $L_\mathrm{T}$, and the gap between them, less the masking index, is the
audibility. A tone whose marker sits *below* its segment is masked, and no
amount of level justifies a penalty for it.

### 3.1 Extended uncertainty of the audibility

Clause 5.4 attaches a 90 % bilateral extended uncertainty $U$ to every
audibility, and clause 6 makes it mandatory whenever fewer than 12 spectra
have been averaged. `analyze_spectrum` computes it per tone
(`res.extended_uncertainties`) from the spectrum lines it holds;
`assess_tones`, which receives levels and no lines, only carries the values
passed to it as `extended_uncertainties=`. `audibility_uncertainty` evaluates it
straight from the spectrum lines, and `mean_audibility_uncertainty` propagates
it to the energy-averaged audibility of a spectrum set (Annex E Table E.2:
$U = 2.80\ \text{dB}$ for the 137.3 Hz tone against the printed 2.79).

**How to read $U$.** A 90 % *bilateral* interval leaves 5 % in each tail, so
$\Delta L - U$ is a one-sided 95 % statement: when the whole interval
$\Delta L \pm U$ sits above 0 dB the tone is audible with at least 95 %
confidence, and when the interval straddles zero the verdict is not
statistically secured; the remedy is more spectra, since $U$ shrinks with the
number averaged (which is exactly why clause 6 makes reporting it mandatory
below 12 spectra). The same logic guards the downstream penalty:
**ISO 1996-2:2017 Annex J** converts the mean audibility into the tonal
adjustment $K_\mathrm{t}$ in 1 dB steps (Table J.1: $K_\mathrm{t} = 0$ for $\Delta L \le 0$, up
to $K_\mathrm{t} = 6\ \text{dB}$ for $\Delta L > 12\ \text{dB}$, or the coarser
0/3/6 dB ladder of its note), so an uncertainty that spans a table boundary
propagates straight into a 1–3 dB question mark on the rating level. Quoting
$\Delta L \pm U$ alongside $K_\mathrm{t}$ shows whether the adjustment is robust or
hinges on one borderline spectrum.

The Annex E campaign is exactly that case. Its five 3 s spectra (Table E.4)
carry decisive audibilities of 9.18, 6.04, 7.46, 2.67 and 7.17 dB with
individual uncertainties of 3.21, 2.95, 2.44, 2.52 and 2.14 dB; the energy mean
is 6.98 dB and `mean_audibility_uncertainty` propagates those into
$U = 1.38$ dB.

![The five ISO/PAS 20065 Annex E measured spectra plotted as decisive audibility against run index, each with its extended uncertainty as an error bar, over horizontal bands marking the ISO 1996-2 Table J.1 tonal adjustment steps; the energy-mean audibility of 6.98 decibels is drawn as a line with a shaded band of plus or minus 1.38 decibels that crosses the boundary between Kt equal to 3 and Kt equal to 4 decibels](https://docs.jmrp.io/phonometry/media/tone_audibility_uncertainty.svg)

*Five spectra, one penalty. The mean audibility earns $K_\mathrm{t} = 4$ dB, but its
uncertainty band reaches below the 6 dB boundary of ISO 1996-2 Table J.1, so
$K_\mathrm{t} = 3$ dB is not excluded, and the remedy is more spectra, not a rounder
number.*

**Show the code for this figure**

```python
# `psychoacoustics` is imported by the snippets above.
# ISO/PAS 20065 Annex E Table E.4: one decisive audibility and one extended
# uncertainty per measured 3 s spectrum.
decisive = [9.18, 6.04, 7.46, 2.67, 7.17]
uncertainty = [3.21, 2.95, 2.44, 2.52, 2.14]

mean = psychoacoustics.mean_audibility(decisive)
u = psychoacoustics.mean_audibility_uncertainty(decisive, uncertainty)
print(round(mean, 2), round(u, 2))     # 6.98 1.38 — Kt = 4 dB, Kt(ΔL − U) = 3 dB
```

## 4. From the narrow-band spectrum

Given the FFT lines of the critical band about a tone, `mean_narrowband_level`
runs the iterative Formula 6 procedure (energy average, dropping any line more
than 6 dB above the running $L_\mathrm{S}$, until stable within ±0.005 dB or fewer than
five lines remain each side, Annex D) and `tone_level` sums the tonal lines
contiguous with the peak (above both $L_\mathrm{S} + 6\ \text{dB}$ and $L_\text{peak} - 10\ \text{dB}$). The
mean always carries the −1.76 dB Hanning bandwidth correction; the tone level
carries it only when the run spans more than one line (Formulae (7)/(8)).

```python
from phonometry import psychoacoustics

# Annex E Table E.1: the 38 lines of the 137.3 Hz critical band (Δf = 2.7 Hz).
freqs = [96.9, 99.6, 102.3, 105.0, 107.7, 110.4, 113.0, 115.7, 118.4, 121.1,
         123.8, 126.5, 129.2, 131.9, 134.6, 137.3, 140.0, 142.7, 145.3, 148.0,
         150.7, 153.4, 156.1, 158.8, 161.5, 164.2, 166.9, 169.6, 172.3, 175.0,
         177.6, 180.3, 183.0, 185.7, 188.4, 191.1, 193.8, 196.5]
levels = [49.40, 50.68, 50.09, 53.37, 44.47, 50.91, 51.41, 59.40, 64.54, 57.57,
          51.02, 50.76, 59.93, 62.94, 58.49, 65.87, 62.66, 50.25, 51.32, 52.30,
          52.58, 53.15, 67.04, 67.27, 57.40, 57.17, 52.56, 51.39, 52.49, 47.68,
          51.26, 49.03, 61.42, 59.52, 48.43, 50.84, 48.20, 55.95]

ls = psychoacoustics.mean_narrowband_level(levels, 137.3, frequencies=freqs)
lt = psychoacoustics.tone_level(levels, 137.3, ls, frequencies=freqs)
print(round(ls, 2), round(lt, 2))                         # 49.22 67.96
print(round(psychoacoustics.tone_audibility(lt, ls, 137.3, 2.7), 2))   # 5.01 dB
```

## 5. Whole-spectrum detection

`analyze_spectrum` runs the full front-end over a spectrum (mean narrow-band
level per line, peak detection (Clause 5.3.8 Step 1, a tone cannot sit on a
slope), tone level, the distinctness test (Clause 5.3.4: bandwidth
$\le 26\,(1 + 0.001 f_\mathrm{T})$ Hz and edge steepness $\ge 24\ \text{dB}$), and audibility) and
returns the distinct, audible tones. It then applies **Step 3**: audible
tones sharing a critical band have their tone levels energy-summed
(Formula 17, shared lines counted once, via `combined_tone_level`) into a
combined "FG" entry rated at the most audible member, unless the
exactly-two-tones-below-1000-Hz exception of §5.1 keeps them separate. The
result's `group_sizes` tells individual tones (`1`) from FG entries ($N \ge 2$),
and the decisive audibility (Step 4) is the maximum over all entries.

```python
from phonometry import psychoacoustics

# Annex E Table E.1: the 38 lines of the 137.3 Hz critical band (Δf = 2.7 Hz).
freqs = [96.9, 99.6, 102.3, 105.0, 107.7, 110.4, 113.0, 115.7, 118.4, 121.1,
         123.8, 126.5, 129.2, 131.9, 134.6, 137.3, 140.0, 142.7, 145.3, 148.0,
         150.7, 153.4, 156.1, 158.8, 161.5, 164.2, 166.9, 169.6, 172.3, 175.0,
         177.6, 180.3, 183.0, 185.7, 188.4, 191.1, 193.8, 196.5]
levels = [49.40, 50.68, 50.09, 53.37, 44.47, 50.91, 51.41, 59.40, 64.54, 57.57,
          51.02, 50.76, 59.93, 62.94, 58.49, 65.87, 62.66, 50.25, 51.32, 52.30,
          52.58, 53.15, 67.04, 67.27, 57.40, 57.17, 52.56, 51.39, 52.49, 47.68,
          51.26, 49.03, 61.42, 59.52, 48.43, 50.84, 48.20, 55.95]

# Same Table E.1 spectrum as above.
res = psychoacoustics.analyze_spectrum(levels, 2.7, frequencies=freqs)
singles = res.group_sizes == 1
print([round(f, 1) for f in res.tone_frequencies[singles]])  # [118.4, 137.3, 158.8]

# Step 3 already combined the three same-band tones into an FG entry:
fg = res.group_sizes > 1
print(int(res.group_sizes[fg][0]), round(float(res.tone_levels[fg][0]), 2))  # 3 72.15

# The same Formula 17 combination, called directly (LS from Table E.2):
lt_fg = psychoacoustics.combined_tone_level(levels, [48.91, 49.22, 50.50], frequencies=freqs,
                                            tone_frequencies=[118.4, 137.3, 158.8])
print(round(lt_fg, 2))                                # 72.15

res.plot()   # the detected entries, FG groups included, as audibility bars
```

Reproducing a *decisive* audibility exactly needs the **complete** narrow-band
spectrum: Table E.1 is truncated to the 137.3 Hz critical band, so the 158.8 Hz
tone's mean narrow-band level is under-estimated from it (the algorithm itself
matches the parent standard DIN 45681:2005-03 reference program). The peak
detection and FG combination are verified against the Annex E worked example
(the three tone frequencies and $L_\mathrm{T} = 72.15\ \text{dB}$).

### 5.1 Two tones below 1000 Hz

When **exactly two** tones share a critical band and both lie below 1000 Hz, the
ear can still tell them apart (so they are rated *separately* rather than
FG-combined) if their frequency difference $|f_{\mathrm{T}1} - f_{\mathrm{T}2}|$ (Formula 18)
exceeds

$$
f_\mathrm{D} = 21 \cdot 10^{\,1.2\,\left|\log_{10}(f_\mathrm{T}/212)\right|^{1.8}}\ \text{Hz}
\qquad (\text{Formula 19},\ 88\ \text{Hz} < f_\mathrm{T} < 1000\ \text{Hz})
$$

evaluated at the more prominent tone $f_\mathrm{T}$ (the larger audibility $\Delta L$).
The threshold bottoms out at 21 Hz at $f_\mathrm{T} = 212\ \text{Hz}$ and grows on
either side. `two_tone_separation_frequency` gives $f_\mathrm{D}$;
`resolve_tones_separately` applies the decision.

![The ISO/PAS 20065 Formula 19 separation threshold fD plotted against the frequency of the more audible tone on a logarithmic axis from 88 hertz to 1 kilohertz: a U-shaped curve with its minimum of 21 hertz at 212 hertz, rising to 34 hertz at 88 hertz and 82 hertz at 1 kilohertz; the region above the curve is shaded and labelled rated separately, the region below labelled energy-summed into one FG entry, and the Annex E pair at 137.3 hertz with a separation of 18.9 hertz is marked below the curve](https://docs.jmrp.io/phonometry/media/two_tone_separation.svg)

*Where the ear stops resolving two tones in one critical band. The threshold is
narrowest around 212 Hz and widens steeply above 500 Hz: at 1 kHz two tones
must be 82 Hz apart to be rated separately, which is most of the critical band.
The Annex E pair sits just below the curve and is therefore combined.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np

# `psychoacoustics` is imported by the snippets above.
ft_grid = np.logspace(np.log10(88.0), np.log10(1000.0), 400)
fd = [psychoacoustics.two_tone_separation_frequency(f) for f in ft_grid]

plt.semilogx(ft_grid, fd, label="Threshold fD (Formula 19)")
plt.plot([212.0], [21.0], "o")          # the 21 Hz minimum
plt.plot([137.3], [18.9], "s")          # the Annex E pair, 118.4 and 137.3 Hz
plt.xlabel("Frequency of the more audible tone fT [Hz]")
plt.ylabel("Frequency separation |fT1 - fT2| [Hz]")
plt.legend()
plt.show()
```

```python
from phonometry import psychoacoustics

psychoacoustics.two_tone_separation_frequency(212.0)             # 21.0 Hz (minimum)
psychoacoustics.resolve_tones_separately(200.0, 260.0, 3.0, 2.0) # True  → rate separately
psychoacoustics.resolve_tones_separately(118.4, 137.3, 4.0, 5.0) # False → combine (Δf < fD)
```

**Note**

No ISO/PAS 20065 worked example exercises this branch; the Annex E band groups
*three* tones, so the "exactly two tones" rule never fires there. The formula and
decision are implemented clean-room from the text and verified against the
**DIN 45681:2005-03** Annex J reference program
(`fD = 21 * 10 ^ (1.2 * Abs(Log(fT / 212) / Log(10)) ^ 1.8)`). Reassuringly,
evaluated at the Annex E tones the threshold ($\approx 24\ \text{Hz}$ at 137.3 Hz) keeps them
combined, consistent with that example's FG grouping.

## 6. Tonal assessment report (`.report()`)

The fiche uses the ISO 1996-2 reporting symbols: the decisive audibility is
written $\Delta L_\mathrm{ta}$ and the tonal adjustment $K$, which are the
$\Delta L$ and $K_\mathrm{t}$ of the sections above.

`ToneAudibilityResult.report(path)` renders a one-page PDF fiche laid out like a
tonal-assessment report of an environmental-noise laboratory, following the
**ISO 1996-2:2017 Annex J** engineering method: a standard-basis line, an
optional metadata header block (source/situation, client, measurement position,
instrumentation and date, with the analysis line spacing $\Delta f$ read from the
result), a full-width table of the key quantities for every detected tone
(tone frequency $f_\mathrm{T}$, entry type, tone level $L_{p\mathrm{t}}$, critical-band
masking-noise level $L_{p\mathrm{n}}$, critical bandwidth $\Delta f_\mathrm{c}$ and the
audibility $\Delta L_\mathrm{ta}$) above the level-versus-frequency analysis plot with
the tones and their critical-band masking noise marked, the boxed decisive
audibility $\Delta L_\mathrm{ta}$ together with the derived tonal adjustment $K$
(Table J.1), an optional PASS/FAIL verdict row and a prominence note, and a
footer with the fixed disclaimer.

It uses the same `ReportMetadata` container and rendering engine as the
[ISO 532-1 loudness fiche](https://docs.jmrp.io/phonometry/perception/psychoacoustics/loudness/#iso-532-1-report-report);
a supplied `requirement` is read as the maximum acceptable decisive audibility
$\Delta L_\mathrm{ta}$ in dB (a quieter tone 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("tone_fiche_es.pdf", language="es")`.

```python
from phonometry import psychoacoustics, ReportMetadata

# The Annex E combustion-engine spectrum from §4/§5 (analyze_spectrum).
res = psychoacoustics.analyze_spectrum(levels, 2.7, frequencies=freqs)
res.report(
    "tone_fiche.pdf",
    metadata=ReportMetadata(
        specimen="Combustion engine, steady operation",
        measurement_standard="ISO 1996-2",
        laboratory="Phonometry Reference Laboratory",
        requirement=6.0,             # maximum acceptable ΔL_ta (dB)
    ),
)                                    # decisive ΔL_ta (dB) and K (dB, Table J.1)
```

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

**Report:** [ISO 1996-2 tonal audibility example report (PDF)](https://docs.jmrp.io/phonometry/media/iso1996_tone_audibility_example.pdf): One-page tonal-assessment fiche: a metadata header, a per-tone table of the tone level Lpt, the critical-band masking-noise level Lpn, the critical bandwidth and the audibility, the level-versus-frequency analysis plot with the tones and their masking noise marked, the boxed decisive ΔL_ta = 9.1 dB with the tonal adjustment K = 5 dB (ISO 1996-2:2017 Table J.1) and a FAIL verdict against a 6 dB audibility limit.

Tonal audibility fiche (ToneAudibilityResult.report), decisive ΔL_ta in dB with the tonal adjustment K.

## What this guide covers

- **Covered.** The ISO/PAS 20065:2016 engineering method in full: the clause 4.2 analyser
  requirements and the clause 4.3 merging of basic spectra that the input rests
  on, with the clause 5.1 acceptance conditions and the clause 7 reporting list;
  the critical bandwidth $\Delta f_\mathrm{c}$ and corner frequencies (Formulae 2 to 5),
  the mean narrow-band level $L_\mathrm{S}$ and tone level $L_\mathrm{T}$ (Formulae 6 and 8, Annex
  D), the critical-band masking level $L_\mathrm{G}$, the masking index $a_\mathrm{v}$ and the
  audibility $\Delta L$ (Formulae 12 to 14), peak detection and the distinctness
  test (clauses 5.3.8 and 5.3.4), the multi-tone FG combination (Formula 17) and
  the two-tones-below-1000-Hz exception (Formulae 18/19), the decisive and
  energy-mean audibility (Formula 20) and the clause 5.4/6 extended uncertainty
  $U$, all through `psychoacoustics.analyze_spectrum` and `assess_tones`. ISO
  1996-2:2017 Annex J is covered as far as it maps the mean audibility to the
  tonal adjustment $K_\mathrm{t}$ (Table J.1).

- **Not covered.** The distinctness edge-steepness test follows the DIN 45681:2005-03 reading
  (its executable reference program), not the asymmetric formulas printed in the
  ISO/PAS 20065 text, which contradict it, recorded in the [errata
  registry](https://docs.jmrp.io/phonometry/reference/errata/). Building the narrow-band FFT
  spectrum itself from a raw time-domain recording is not part of this module:
  every function here takes an already-computed spectrum (levels and
  frequencies), and §0 states what that spectrum has to be, including the
  clause 5.3.2 A-weighting, which the caller applies because the module is
  weighting agnostic.

## See also

- [Environmental noise measurement](https://docs.jmrp.io/phonometry/environment/assessment/environmental-levels/):
  the ISO 1996-1 rating level this audibility feeds, and the simpler ISO 1996-2
  Annex C route for a single spectrum.
- [Tone prominence: TNR and PR](https://docs.jmrp.io/phonometry/perception/psychoacoustics/tone-prominence/):
  the ECMA-418-1 prominence question asked of product noise, with the table
  that picks between the library's four tonality metrics.
- [Wind turbine noise](https://docs.jmrp.io/phonometry/environment/sources/wind-turbine-noise/):
  the IEC 61400-11 per-bin audibility, the same formula applied to a
  bin-by-bin wind-speed campaign.
- [Theory](https://docs.jmrp.io/phonometry/reference/theory/perception/#tone-prominence-tnr-and-pr-ecma-418-1):
  the neighbouring family of tone metrics, TNR and PR, derived from the same
  critical-band picture.
- API reference: [`psychoacoustics`: tone audibility](https://docs.jmrp.io/phonometry/reference/api/psychoacoustics/tone-audibility/).

## References

- **Deutsches Institut für Normung. (2005).** [*Akustik — Bestimmung der Tonhaltigkeit von Geräuschen und Ermittlung eines Tonzuschlages für die Beurteilung von Geräuschimmissionen*](https://www.dinmedia.de/en/standard/din-45681/76463232) (DIN 45681:2005-03).
  The parent standard the −1.76 dB Hanning bandwidth correction, the iterative masking-level procedure and the detection/combination logic are confirmed against (its Annex J reference program).
- **International Organization for Standardization. (2016).** [*Acoustics — Objective method for assessing the audibility of tones in noise — Engineering method*](https://www.iso.org/standard/66941.html) (ISO/PAS 20065:2016).
  The implemented engineering method: every formula on this page follows the 2016 PAS edition. The critical bandwidth Δfc (Formula 2) and its corner frequencies (Formulae 3–5), the critical-band level LG (Formula 12), the masking index av (Formula 13), the audibility ΔL = LT − LG − av (Formula 14) and the energy-mean mean audibility (Formula 20); the mean narrow-band level LS (Formula 6, iterative Annex D) and tone level LT (Formula 8) come from the critical-band spectrum, and analyze_spectrum adds peak detection (Clause 5.3.8) with the distinctness criteria (Clause 5.3.4), the multi-tone FG combination (Formula 17) and the separate evaluation of two tones below 1000 Hz (Formulae 18/19). Conformance is anchored on the Annex E combustion-engine worked example (Tables E.1/E.2/E.3). Withdrawn, superseded by ISO/TS 20065:2022 (https://www.iso.org/standard/81518.html).
- **International Organization for Standardization. (2017).** [*Acoustics — Description, measurement and assessment of environmental noise — Part 2: Determination of sound pressure levels*](https://www.iso.org/standard/59766.html) (ISO 1996-2:2017).
  The environmental-noise standard this method serves: its Annex J adopts the engineering method and maps the mean audibility to the tonal adjustment Kt (Table J.1) discussed in §3.1.
