# Detailed Per-Band Prediction (ISO 12354)

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

Canonical: https://docs.jmrp.io/phonometry/buildings/design/detailed-prediction/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/buildings/design/detailed-prediction/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: 2026-07-30
Updated: 2026-11-16T21:36:14Z
Standards: ISO 12354-1:2017; ISO 12354-2:2017
Verified: 6 conformance checks (ISO 12354-1:2017; ISO 12354-2:2017) https://docs.jmrp.io/phonometry/reference/conformance/#room-building-acoustics
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: ISO 12354-1/-2:2017 detailed model: in-situ element and junction data, flanking band by band, and the per-path contributions to R'w and L'n,w.

---

The simplified model of
[Predicting Sound Insulation (EN 12354)](https://docs.jmrp.io/phonometry/buildings/design/insulation-prediction/) returns one
number, $R'_\mathrm{w}$ or $L'_\mathrm{n,w}$, and hides everything that happens inside the
spectrum. The **detailed model** (ISO 12354-1:2017 Clause 4.2 airborne,
ISO 12354-2:2017 Clause 4.2 impact) carries every quantity through the
one-third-octave bands instead: it converts the laboratory element data into
their *in-situ* values, forms each transmission path band by band, sums them
energetically and only then rates the spectrum through ISO 717. That is what a
consultant runs when the element spectra are known and the question is not
"does it pass" but "which path do I have to fix, and in which bands".

This page walks the standard's own worked building end to end, and closes with
the lightweight (Type B) branch and the checks the standard offers on its own
numbers.

## The chain, band by band

Before any of it: the standard splits every element in two, and the whole chain
below is one of the two branches. A **Type A element** (clause 3.3.5) is one
whose structural reverberation time is primarily determined by the elements it is
connected to, up to at least the 1 kHz one-third-octave band, and across which
the vibration level falls by less than 6 dB in the direction perpendicular to the
junction line. Cast in-situ concrete, solid wood including cross-laminated
timber, glass, plastic, metal and blockwork with a finish that mechanically
connects it are the standard's own examples. A **Type B element** (3.3.6) is
anything else, typically plasterboard or timber cladding on a timber or metal
frame. The distinction is not academic: it is exactly why the Type A chain
assembles the loss factor from the junction perimeter, because for a Type A
element the junctions *are* the damping. For a Type B element the standard takes
$T_\mathrm{s,situ} = T_\mathrm{s,lab}$, so the in-situ correction is 0 dB and
$R_\mathrm{situ} = R$.

The Annex L building, and steps 1 to 5 below, are the Type A branch. Two
consequences are worth carrying. The classification is **per frequency band**.
Notes 2 to both definitions say a masonry wall can be Type A low and mid and
Type B at the top of the range, so the binary heading of the lightweight section
further down is a simplification of a boundary that actually moves with
frequency. And at a mixed junction (clause 4.2.2.3, "for example Type B wall on
Type A floor") Formula (10) may still be used as an approximation, with the Type
B element's equivalent absorption length taken as its area over the reference
length, $a_{i,\mathrm{situ}} = S_{i,\mathrm{situ}}/l_0$ (Formula 13). Running
`in_situ_element` on a framed partition instead assembles a total loss factor
from junctions that do not control its damping, and nothing will warn you.

Both parts share the same machinery, so a building is described once and the
airborne and impact chains read the same in-situ element data.

**1. Element performance.** For a homogeneous element the standard calculates
the sound reduction index from the material properties (Annex B):

$$
\tau = \left(\frac{2\rho_\mathrm{o} c_\mathrm{o}}{2\pi f m'}\right)^2 \cdot
\frac{\pi f_\mathrm{c} \sigma^2}{2 f \eta_\mathrm{tot}} \quad (f > f_\mathrm{c})
$$

with the radiation factor for free bending waves $\sigma$ (Formulae B.4-B.6),
the radiation factor for forced waves $\sigma_\mathrm{f}$ (Formula B.3) below $f_\mathrm{c}$,
and a third branch for the band that straddles $f_\mathrm{c}$. Above about 1 kHz a
thick element stops improving, and Formula (B.10) floors the transmission
factor at a plateau set by $\rho c_\mathrm{L}$.

**2. In-situ conversion.** The element radiates and is damped differently in
the building than in the test frame. The total loss factor in situ is
(Formula C.1)

$$
\eta_\mathrm{tot} = \eta_\mathrm{int}
+ \frac{2\rho_\mathrm{o} c_\mathrm{o} \sigma}{2\pi f m'}
+ \frac{c_\mathrm{o}}{\pi^2 S \sqrt{f f_\mathrm{c}}} \sum_k l_k \alpha_k
$$

the three terms being the internal losses of the material, the radiation into
the air, and the losses at the perimeter. The perimeter coefficients follow
from the junctions themselves, $\alpha_k = \sum_j \sqrt{f_{\mathrm{c},j}/f_\mathrm{ref}}
\, 10^{-K_{ij}/10}$ (Formula C.4). From $\eta_\mathrm{tot}$ come the structural
reverberation time $T_\mathrm{s} = 2.2/(f\,\eta_\mathrm{tot})$ and the equivalent
absorption length $a_\mathrm{situ} = 2.2\pi^2 S \sqrt{f_\mathrm{ref}/f} /
(c_\mathrm{o} T_\mathrm{s,situ})$ (Formula 11).

**3. Junctions.** The situation-invariant $K_{ij}$ becomes the level drop the
junction actually produces (Formula 10):

$$
\overline{D}_{\mathrm{v},ij,\mathrm{situ}} = K_{ij}
- 10\log_{10}\!\left(\frac{l_{ij}}{\sqrt{a_{i,\mathrm{situ}}\,a_{j,\mathrm{situ}}}}\right)
\ \ge 0\ \text{dB}
$$

**4. Paths.** The direct path is $R_\mathrm{Dd} = R_\mathrm{s,situ} +
\Delta R_\mathrm{D,situ} + \Delta R_\mathrm{d,situ}$ (Formula 14) and every
flanking path (Formula 15)

$$
R_{ij} = \frac{R_{i,\mathrm{situ}}}{2} + \Delta R_{i,\mathrm{situ}}
+ \frac{R_{j,\mathrm{situ}}}{2} + \Delta R_{j,\mathrm{situ}}
+ \overline{D}_{\mathrm{v},ij,\mathrm{situ}}
+ 10\log_{10}\!\left(\frac{S_\mathrm{s}}{\sqrt{S_i S_j}}\right)
$$

The impact side runs in parallel: the bare slab's per-band level is
$L_\mathrm{n} = 155 - 30\log_{10} m' + 10\log_{10} T_\mathrm{s} + 10\log_{10}\sigma + 10\log_{10}(f/f_\mathrm{ref})$
(Part 2 Formula B.2), the direct path subtracts the covering and any ceiling
(Formula 11), and each flanking path is Formula (12).

**5. Assembly.** $R' = -10\log_{10}\sum 10^{-R/10}$ over the thirteen paths of a
four-flanking-element room, and $L'_\mathrm{n} = 10\log_{10}\sum 10^{L_\mathrm{n}/10}$ over the five
impact paths, then ISO 717.

## Starting from measured element data

Clause 4.1 is explicit about where the numbers should come from: acoustic data on
the elements involved *should be taken primarily from standardized laboratory
measurements*, and the sources of data used shall be stated. The Annex B
calculation the page opens with is the fallback for when no measurement exists.
A real project usually starts from a catalogue sheet or an ISO 10140 test report,
and that route has its own three steps.

**The transfer to the building is a loss-factor transfer.** The laboratory
measured the element in a test frame whose damping is not the building's, so
Formula (9) rescales the index by the ratio of the two structural reverberation
times, $R_\mathrm{situ} = R - 10\lg(T_\mathrm{s,situ}/T_\mathrm{s,lab})$, and Part 2
Formula (5) does the same for the impact level **with the sign reversed**. Read
the sign: an element more heavily damped in the building than in the test frame
has a shorter $T_\mathrm{s}$ in situ, which *raises* its index and *lowers* its impact
level; an element more lightly damped loses index. The standard's own note adds
that $R_\mathrm{situ} = R$ is a usable first approximation, and for a Type B
element it is exact.

**$T_\mathrm{s,lab}$ comes from the test report, or from Annex C.** If the
laboratory printed it, use it. If not, Formula (C.3) estimates the laboratory
total loss factor as
$\eta_\mathrm{tot,lab} \approx \eta_\mathrm{int} + m'/(485\sqrt{f}) \approx 0.01 +
m'/(485\sqrt{f})$, valid for elements below $m'$ = 800 kg/m², with
$\eta_\mathrm{int}$ normally taken as 0.01 (Annex B Table B.1 lists it per
material). `laboratory_total_loss_factor` is that formula and
`structural_reverberation_time` turns it into $T_\mathrm{s} = 2.2/(f\eta_\mathrm{tot})$;
$T_\mathrm{s,situ}$ comes from the same Formula (C.1) chain the page already runs.

**Flanking paths need one more correction than the direct path.** Below its
critical frequency a laboratory $R$ contains forced transmission, which a
flanking path does not carry: only free bending waves cross a junction. Annex B.1
therefore corrects a measured $R$ to resonant transmission only before it enters
a flanking path (the 8 dB estimate for a single frame element without a cavity)
while the **direct** path keeps the forced transmission the laboratory measured.
That is `resonant_sound_reduction_index`, and `in_situ_element(...,
resonant_only=True)` does the same on a calculated element.

The four names below (`laboratory_total_loss_factor`,
`structural_reverberation_time`, `in_situ_reduction_index` and
`in_situ_impact_level`) live in the same `building` module as the rest of the
page, and the block starts from your own measured spectra.

```python
import numpy as np
from phonometry import building

bands = np.array([50, 63, 80, 100, 125, 160, 200, 250, 315, 400, 500, 630,
                  800, 1000, 1250, 1600, 2000, 2500, 3150], float)

# The 220 mm separating floor in situ, the element the next section assembles
# from its junctions: its perimeter sum is 2.659 m (Formula C.4) and its
# `reverberation_time` is Ts,situ.
el = building.in_situ_element(building.HomogeneousElement(
    label="floor", area=20.0, length1=5.0, length2=4.0, mass_per_area=484.0, critical_frequency=76.8, internal_loss_factor=0.005, perimeter_absorption=2.659, density=2200.0, longitudinal_velocity=3800.0), bands)

# A measured laboratory R and Ln for that same floor, transferred to the
# building.
r_measured = ...    # measured R per band, dB (ISO 10140-2 test report)
ln_measured = ...   # measured Ln per band, dB (ISO 10140-3 test report)

eta_lab = building.laboratory_total_loss_factor(
    bands, mass_per_area=484.0, internal_loss_factor=0.005)   # (C.3)
ts_lab = building.structural_reverberation_time(eta_lab, frequencies=bands)  # (C.1)
print(np.round(ts_lab[[0, 10]], 3))                    # [0.301 0.089] s
print(np.round(el.reverberation_time[[0, 10]], 3))     # [0.53  0.152] s, in situ

r_situ = building.in_situ_reduction_index(r_measured, el.reverberation_time,
                                          ts_lab)
ln_situ = building.in_situ_impact_level(ln_measured, el.reverberation_time,
                                        ts_lab)
# This floor is *less* damped in the building than in the test frame (Ts,situ is
# the longer of the two), so Formula (9) takes 2.0 to 2.5 dB off its index across
# the range and Part 2 Formula (5) adds the same to its impact level.

# A flanking path takes the resonant-only form of the same measured index;
# the direct path does not.
r_flanking = building.resonant_sound_reduction_index(r_situ, frequencies=bands,
                                                     critical_frequency=76.8)
```

### What you need before you start

| Quantity | Where it comes from |
| :--- | :--- |
| $R$, $L_\mathrm{n}$ per band | ISO 10140-2 / ISO 10140-3 test report (**with the laboratory structural reverberation time from the same report**) |
| $\Delta R$ of a lining | ISO 10140-1 Annex G, or ISO 12354-1 Annex D |
| $\Delta L$ of a covering | [ISO 16251-1](https://docs.jmrp.io/phonometry/buildings/design/impact-improvement/) or ISO 10140-3 |
| $K_{ij}$ per junction | ISO 10848-1/-4, or the [Annex E catalogue](https://docs.jmrp.io/phonometry/buildings/design/insulation-prediction/) |
| $m'$, $f_\mathrm{c}$, $\eta_\mathrm{int}$, $c_\mathrm{L}$ | the material data sheet, or Annex B Table B.1 |
| Areas and coupling lengths | the drawings, measured **surface to surface** |

That last row changes an answer: a coupling length measured surface to surface
rather than centre to centre differs by a leaf thickness on a lined or
double-leaf element, and $l_{ij}$ enters Formula (10) directly. Clause 4.1 also
requires the source of every input to be stated, which on a fiche is what the
`notes` field of `ReportMetadata` is for.

## The worked building of Annex L / Annex G

ISO 12354-1:2017 Annex L and ISO 12354-2:2017 Annex G describe the **same**
building: two dwellings one above the other, 55 m³ rooms, a 5.00 m × 4.00 m
separating floor of 220 mm concrete carrying a 35 mm floating screed on
mineral wool, two 365 mm autoclaved aerated concrete external walls and two
200 mm calcium-silicate internal walls, meeting at rigid T and cross
junctions. Together they print about twenty per-band tables, one per
intermediate quantity, which makes the example a complete oracle for the
model. Eight defects of those printed tables are recorded in
[Errata](https://docs.jmrp.io/phonometry/reference/errata/), and the fixture below takes the
corrected readings.

![The ISO 12354-1 Annex L worked building: a section through two stacked dwellings showing the 220 mm separating floor with its floating screed, the two external and two internal walls with their masses and critical frequencies and the Kij of each junction, and a plan of the separating floor with the four coupling lengths dimensioned and the thirteen transmission paths counted](https://docs.jmrp.io/phonometry/media/diagram_iso12354_annexl.svg)

The drawing is what ties the element list to the numbers the snippet multiplies.
The separating floor's perimeter sum is 9 m of external edge plus 9 m of internal
edge because the 5.00 m × 4.00 m floor meets an external wall along each 4.00 m
edge and an internal wall along each 5.00 m edge, and each of those junctions
carries the floor above and below it. An external wall's own perimeter takes
8.0 m of horizontal junction (its top and bottom edges, 4.00 m each) and 2.75 m
of vertical. The floor-to-external-wall junctions are rigid T, the
floor-to-internal-wall junctions rigid cross. And the thirteen paths are one
direct path plus three per flanking element, $1 + 4\times3$.

```python
import numpy as np
from phonometry import building

bands = np.array([50, 63, 80, 100, 125, 160, 200, 250, 315, 400, 500, 630,
                  800, 1000, 1250, 1600, 2000, 2500, 3150], float)

# Annex E junctions (unrounded): floor-to-external-wall rigid T, external wall
# in-line across it, floor-to-internal-wall rigid cross, internal wall in-line.
k_floor_ext = building.junction_vibration_reduction("rigid_t", "corner", 484.0 / 219.0)
k_ext_ext = building.junction_vibration_reduction("rigid_t", "through", 484.0 / 219.0)
k_floor_int = building.junction_vibration_reduction("rigid_cross", "corner", 360.0 / 484.0)
k_int_int = building.junction_vibration_reduction("rigid_cross", "through", 484.0 / 360.0)
print(round(k_floor_ext, 1), round(k_ext_ext, 1))     # 6.4 11.2  (Table L.5)
print(round(k_floor_int, 1), round(k_int_int, 1))     # 8.8 11.0  (Table L.6)

# The floor's perimeter: it butts into an external wall above and below at each
# of its two external edges, and crosses the internal walls at the other two.
a_at_ext = building.perimeter_absorption_coefficient(
    [k_floor_ext] * 2, critical_frequencies=[92.6, 92.6])
a_at_int = building.perimeter_absorption_coefficient(
    [building.junction_vibration_reduction("rigid_cross", "through", 360.0 / 484.0),
     k_floor_int, k_floor_int],
    critical_frequencies=[76.8, 128.4, 128.4])
floor_perimeter = 9.0 * (a_at_ext + a_at_int)         # 2.659 m (Formula C.4)

floor = building.HomogeneousElement(label="floor", area=20.0, length1=5.0, length2=4.0, mass_per_area=484.0, critical_frequency=76.8, internal_loss_factor=0.005,
                                    perimeter_absorption=floor_perimeter, density=2200.0, longitudinal_velocity=3800.0)
el = building.in_situ_element(floor, bands)
print(np.round(el.total_loss_factor[[0, 10]], 4))     # [0.0831 0.029 ]  Table L.3
print(np.round(el.sound_reduction_index[[0, 10]], 1)) # [31.8 54.9]      Table L.3
print(np.round(el.impact_level[[0, 10]], 1))          # [57.3 63.6]      Table G.3
print(np.round(el.absorption_length[[0, 10]], 1))     # [10.8 11.9]      Table L.4
```

Every printed column of Tables L.2, L.3, L.4, G.3 and G.4 comes back within
0.06 dB, and the same fixture drives both totals.

The element's $m'$ and $f_\mathrm{c}$ follow from its material and its
thickness, $m' = \rho t$ and $f_\mathrm{c} = c_\mathrm{o}^2 / (1.8\, c_\mathrm{L} t)$
with the density and the quasi-longitudinal phase velocity of Table B.3, and
`HomogeneousElement.from_solid` works both out from a solid's catalogue row,
a published one or one of [your own catalogues](https://docs.jmrp.io/phonometry/io/material-catalogues/). The internal loss factor
is never read from the row, which may hold four, so it is given here as the
element specification gives it. The unrounded $f_\mathrm{c}$ is the one the
standard's footnote says its calculation used:

```python
from phonometry import solids

concrete = solids.SolidMaterial.from_printed(
    name="Concrete",
    source="ISO 12354-1:2017 Table B.3",
    density_kg_m3=2200.0,
    plate_longitudinal_speed_m_s=3800.0,
)
same_floor = building.HomogeneousElement.from_solid(
    concrete,
    thickness_m=0.22,
    internal_loss_factor=0.005,
    area=20.0,
    length1=5.0,
    length2=4.0,
    perimeter_absorption=floor_perimeter,
    label="floor",
)
print(round(same_floor.mass_per_area, 1), round(same_floor.critical_frequency, 2))   # 484.0 76.82
from_row = building.in_situ_element(same_floor, bands)
print(np.round(from_row.sound_reduction_index[[0, 10]], 1))   # [31.8 54.9]  Table L.3
```

```python
# The floating floor: 35 mm screed, m' = 73.5 kg/m2 on s' = 8 MN/m3.
f0 = 160.0 * np.sqrt(8.0 / 73.5)                      # 52.8 Hz (Formula C.2)
delta = building.floating_floor_improvement(bands, resonance_frequency=f0)  # 30 lg(f/f0)

wall = building.in_situ_element(building.HomogeneousElement(
    label="ext1", area=11.0, length1=4.0, length2=2.75, mass_per_area=219.0, critical_frequency=92.6, internal_loss_factor=0.0125, perimeter_absorption=2.375, density=600.0, longitudinal_velocity=1900.0), bands)

# Path D1 (Df: separating floor -> external wall 1), Table L.4.
d1 = building.airborne_flanking_path(
    label="D1", kind="Df", element_i=el, element_j=wall,
    vibration_reduction_index=k_floor_ext, coupling_length=4.0,
    separating_area=20.0, delta_r_i=delta)
print(np.round(d1.values[[0, 10]], 1))                # [ 41.2  93.6]  Table L.1
```

Assembling the direct path and all twelve flanking paths gives the apparent
index per band, its energy split and the ISO 717-1 rating in one call:

```python
# The other three flanking elements, built the way `wall` was above, each with
# its own area and perimeter sum (Formula C.4); `situ` then holds all five.
specs = (("ext2", 13.75, 5.0, 2.75, 219.0, 92.6, 0.0125, 2.548, 600.0, 1900.0),
         ("int1", 11.0, 4.0, 2.75, 360.0, 128.4, 0.01, 1.636, 1800.0, 2500.0),
         ("int2", 13.75, 5.0, 2.75, 360.0, 128.4, 0.01, 1.839, 1800.0, 2500.0))
situ = {"floor": el, "ext1": wall}
for spec in specs:
    situ[spec[0]] = building.in_situ_element(
        building.HomogeneousElement(*spec), bands)

# The twelve flanking paths, three per flanking element, each built the way
# `d1` was above.
paths = []
for tag, name, lij in (("1", "ext1", 4.0), ("2", "ext2", 5.0),
                       ("3", "int1", 4.0), ("4", "int2", 5.0)):
    flank = situ[name]
    cross = k_floor_ext if name.startswith("ext") else k_floor_int
    through = k_ext_ext if name.startswith("ext") else k_int_int
    paths += [
        building.airborne_flanking_path(label=f"D{tag}", kind="Df", element_i=situ["floor"],
                                        element_j=flank, vibration_reduction_index=cross,
                                        coupling_length=lij, separating_area=20.0,
                                        delta_r_i=delta),
        building.airborne_flanking_path(label=f"{tag}d", kind="Fd", element_i=flank,
                                        element_j=situ["floor"], vibration_reduction_index=cross,
                                        coupling_length=lij, separating_area=20.0),
        building.airborne_flanking_path(label=f"{tag}{tag}", kind="Ff", element_i=flank,
                                        element_j=flank, vibration_reduction_index=through,
                                        coupling_length=lij, separating_area=20.0),
    ]

res = building.detailed_airborne_prediction(
    bands, direct_index=building.direct_reduction_index(el.sound_reduction_index,
                                                        delta_r_source=delta),
    flanking_paths=paths)
print(np.round(res.r_prime[[0, 10]], 1))   # [28.8 55.9]   Table L.1 total
print(res.rating.rating, res.dominant[0])  # 57 'Dd'
res.plot()                                 # per-band path-contribution bars
```

![Stacked per-band shares of the transmitted energy for the thirteen transmission paths of the ISO 12354-1 Annex L building, with the apparent sound reduction index R' overlaid: the direct path Dd carries half the energy at 50 Hz, falls away through the 80 Hz to 160 Hz transition and is negligible from 200 Hz upwards, after which flanking paths across the external and internal walls take over](https://docs.jmrp.io/phonometry/media/detailed_prediction_paths.svg)

*This is the plot the whole detailed model exists for. Below 80 Hz the
separating floor itself is the problem, so a heavier slab or a better floating
floor would help; it fades through the 80 Hz to 160 Hz transition and from
200 Hz upwards it has left the budget entirely, the flanking paths across the
walls setting $R'$ on their own, so no amount of work on the floor would move
the result there. The single number $R'_\mathrm{w} = 57$ dB says none of that.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np
from phonometry import building

bands = np.array([50, 63, 80, 100, 125, 160, 200, 250, 315, 400, 500, 630,
                  800, 1000, 1250, 1600, 2000, 2500, 3150], float)
k = {
    "floor-ext": building.junction_vibration_reduction("rigid_t", "corner", 484.0 / 219.0),
    "ext-ext": building.junction_vibration_reduction("rigid_t", "through", 484.0 / 219.0),
    "floor-floor": building.junction_vibration_reduction("rigid_cross", "through", 360.0 / 484.0),
    "floor-int": building.junction_vibration_reduction("rigid_cross", "corner", 360.0 / 484.0),
    "int-int": building.junction_vibration_reduction("rigid_cross", "through", 484.0 / 360.0),
    "int-ext": building.junction_vibration_reduction("rigid_t", "corner", 360.0 / 219.0),
    "extT-ext": building.junction_vibration_reduction("rigid_t", "through", 360.0 / 219.0),
    "corner": building.junction_vibration_reduction("corner", "corner", 1.0),
    "int-int-x": building.junction_vibration_reduction("rigid_cross", "through", 1.0),
}
def a(critical_frequencies, indices):
    return building.perimeter_absorption_coefficient(
        indices, critical_frequencies=critical_frequencies)

floor_sum = 9.0 * (a([92.6, 92.6], [k["floor-ext"]] * 2)
                   + a([76.8, 128.4, 128.4],
                       [k["floor-floor"], k["floor-int"], k["floor-int"]]))
ext_top = a([76.8, 92.6], [k["floor-ext"], k["ext-ext"]])
ext_side = a([92.6], [k["corner"]]) + a([128.4, 92.6], [k["int-ext"], k["extT-ext"]])
int_top = a([76.8, 76.8, 128.4], [k["floor-int"], k["floor-int"], k["int-int"]])
int_side = a([92.6, 92.6], [k["int-ext"]] * 2) + a([128.4] * 3, [k["int-int-x"]] * 3)

specs = {
    "floor": (20.0, 5.0, 4.0, 484.0, 76.8, 0.005, floor_sum, 2200.0, 3800.0),
    "ext1": (11.0, 4.0, 2.75, 219.0, 92.6, 0.0125,
             8.0 * ext_top + 2.75 * ext_side, 600.0, 1900.0),
    "ext2": (13.75, 5.0, 2.75, 219.0, 92.6, 0.0125,
             10.0 * ext_top + 2.75 * ext_side, 600.0, 1900.0),
    "int1": (11.0, 4.0, 2.75, 360.0, 128.4, 0.01,
             8.0 * int_top + 2.75 * int_side, 1800.0, 2500.0),
    "int2": (13.75, 5.0, 2.75, 360.0, 128.4, 0.01,
             10.0 * int_top + 2.75 * int_side, 1800.0, 2500.0),
}
situ = {name: building.in_situ_element(building.HomogeneousElement(label=name, *spec), bands)
        for name, spec in specs.items()}
delta = building.floating_floor_improvement(bands,
                                            resonance_frequency=160.0 * np.sqrt(8.0 / 73.5))

paths = []
for tag, name, lij in (("1", "ext1", 4.0), ("2", "ext2", 5.0),
                       ("3", "int1", 4.0), ("4", "int2", 5.0)):
    wall = situ[name]
    cross = k["floor-ext"] if name.startswith("ext") else k["floor-int"]
    through = k["ext-ext"] if name.startswith("ext") else k["int-int"]
    paths += [
        building.airborne_flanking_path(label=f"D{tag}", kind="Df", element_i=situ["floor"],
                                        element_j=wall, vibration_reduction_index=cross,
                                        coupling_length=lij, separating_area=20.0,
                                        delta_r_i=delta),
        building.airborne_flanking_path(label=f"{tag}d", kind="Fd", element_i=wall,
                                        element_j=situ["floor"], vibration_reduction_index=cross,
                                        coupling_length=lij, separating_area=20.0),
        building.airborne_flanking_path(label=f"{tag}{tag}", kind="Ff", element_i=wall,
                                        element_j=wall, vibration_reduction_index=through,
                                        coupling_length=lij, separating_area=20.0),
    ]
res = building.detailed_airborne_prediction(
    bands,
    direct_index=building.direct_reduction_index(situ["floor"].sound_reduction_index,
                                                 delta_r_source=delta),
    flanking_paths=paths)

# One line — the per-band path contributions with R' overlaid:
res.plot()
plt.show()
```

The impact side of the same building runs on the same `situ` dictionary:

```python
from phonometry import building

k_floor_ext = building.junction_vibration_reduction("rigid_t", "corner", 484.0 / 219.0)
k_floor_int = building.junction_vibration_reduction("rigid_cross", "corner", 360.0 / 484.0)
direct = building.direct_impact_level(situ["floor"].impact_level, delta_l=delta)
flanking = [
    building.impact_flanking_path(label=f"Df{tag}", floor=situ["floor"], element_j=situ[name],
                                  vibration_reduction_index=(
                             k_floor_ext if name.startswith("ext") else k_floor_int),
                                  coupling_length=lij, delta_l=delta)
    for tag, name, lij in (("1", "ext1", 4.0), ("2", "ext2", 5.0),
                           ("3", "int1", 4.0), ("4", "int2", 5.0))
]
imp = building.detailed_impact_prediction(bands, direct_level=direct, flanking_paths=flanking)
print(np.round(imp.l_prime_n[[3, 10]], 1))     # [54.  35.9]   Table G.1 total
print(imp.rating.rating, imp.rating.ci)        # 41 2           printed 41.0 (2)
```

![Stacked per-band shares of the transmitted energy for the five impact paths of the ISO 12354-2 Annex G building with the apparent normalized impact sound pressure level overlaid, the direct path through the excited floor dominating in every band and the floating floor resonance marked on the frequency axis](https://docs.jmrp.io/phonometry/media/detailed_impact_paths.svg)

*The impact half of the same building, and it is a different picture from the
airborne one. There are five paths, not thirteen, because only the floor is
excited: the tapping machine stands on it, so there is one direct path and one
Df path into each of the four walls, and no Ff or Fd. The direct path through the
excited floor governs in every band. That is the practical lesson: an impact
problem is fixed at the source floor, not at the junctions.*

**Show the code for this figure**

```python
# One line — the per-band path contributions with L'n overlaid:
imp.plot()
plt.show()
```

Read the rating the same way. $L'_\mathrm{n,w} = 41$ dB with $C_\mathrm{I} = +2$ dB means the
unweighted, walking-relevant level sits above the weighted rating, which is the
signature of a floor whose improvement is concentrated high up, exactly what a
floating floor with $f_0 = 52.8$ Hz and a 30 lg law does. Three levers move it,
in the order the model responds to them: a *softer* resilient layer lowers $f_0$
and lifts the whole improvement curve; a *heavier* base slab lowers the bare
level directly through Part 2's Formula (B.2); and a better junction does almost
nothing until the direct path has been dealt with. Note too that the floating
floor's $\Delta L$ is subtracted from the direct path and from every flanking
path alike, because all five start at the same excited floor.

## Simplified against detailed

The standard applies its simplified model to the same building (Tables L.10 /
G.10): $R'_\mathrm{w} = 57.0$ dB and $L'_\mathrm{n,w} = 39.7$ dB against the detailed
model's $R'_\mathrm{w} = 57$ dB and $L'_\mathrm{n,w} = 41$ dB. The two agree well inside the
models' own stated spread, and the library's test suite pins that agreement.
The detailed model's advantage is not accuracy on the single number but the
spectrum behind it: the airborne prediction of the detailed model carries no
bias error and a standard deviation of 1.5 dB to 2.5 dB (Clause 5) against
about 2 dB for the simplified one.

## Lightweight constructions (Type B)

For elements whose structural reverberation time is *not* set by the connected
elements the standard takes $T_\mathrm{s,situ} = T_\mathrm{s,lab}$ and
describes the junction with the **normalized** velocity level difference
$\overline{D}_{\mathrm{v},ij,\mathrm{n}}$ instead of $K_{ij}$ (Formula 17), or with a laboratory
measurement of the flanking level difference $D_\mathrm{n,f}$ (Formula 16). Below
$f_\mathrm{c}$ the element indices must first be corrected to resonant transmission
only (Annex B.1/B.2, an 8 dB estimate for single frame elements without a
cavity). The impact side offers the same two routes: Part 2 Formula (14) from
$\overline{D}_{\mathrm{v},ij,\mathrm{n}}$ and Part 2 Formula (13) from a measured normalized
flanking impact level $L_\mathrm{n,f}$.

Unlike the Type A chain, this branch consumes **measured** junction data, so its
four inputs are per-band arrays over `bands` that come from somewhere specific.
`r_wall_leaf` is the leaf's laboratory sound reduction index per band
(ISO 10140-2), corrected below its critical frequency to resonant transmission
only because a flanking path carries free bending waves alone. `dv_n` is the
**normalized** direction-averaged velocity level difference of the junction in
dB, measured per
[ISO 10848-2](https://docs.jmrp.io/phonometry/buildings/insulation/flanking-lab/) and referred to a
reference length of 1 m. `dnf_13` is the normalized flanking level difference
$D_\mathrm{n,f}$ of a junction measured in a laboratory, in dB, which the call re-scales
by the ratio of the site and laboratory coupling lengths. And `lnf_13` is its
impact counterpart, the normalized flanking impact level $L_\mathrm{n,f}$ in dB, which
re-scales by both area and coupling length. The block below is an excerpt: it
starts from those four arrays.

```python
from phonometry import building

# ISO 12354-1 L.2.1: a wood frame building, floor 20 m2, junction 4 m.
r_wall_leaf = ...   # measured R of the inner leaf per band, dB (ISO 10140-2)
dv_n = ...          # normalized junction velocity level difference, dB (ISO 10848-2)
dnf_13 = ...        # laboratory normalized flanking level difference Dn,f, dB
lnf_13 = ...        # laboratory normalized flanking impact level Ln,f, dB

r_star = building.resonant_sound_reduction_index(r_wall_leaf, frequencies=bands,
                                                 critical_frequency=2200.0)   # +8 dB below fc
r_ff = building.flanking_reduction_index_from_normalized_difference(
    index_i=r_star, index_j=r_star, normalized_velocity_level_difference=dv_n,
    separating_area=20.0, coupling_length=4.0)                        # Table L.11

# ISO 12354-1 L.2.2: a measured junction between two timber frame walls.
r13 = building.flanking_reduction_index_from_flanking_level(
    dnf_13, separating_area=10.44, coupling_length=2.41,
    laboratory_coupling_length=2.5)                                   # Table L.15

# ISO 12354-2 Formula (13): the impact twin, from a measured Ln,f.
ln_13 = building.flanking_impact_level_from_flanking_level(
    lnf_13, area=20.0, laboratory_area=10.0,
    coupling_length=4.0, laboratory_coupling_length=4.5)
```

## Checks the standard gives you

Two identities let a spectrum check itself. For a homogeneous floor the
airborne index and the impact level add up to a function of frequency alone
(Part 2 Formulae B.3/B.4), $R + L_\mathrm{n} = 38 + 30\log_{10} f$ in one-third-octave bands
and $43 + 30\log_{10} f$ in octave bands, valid where forced transmission is
negligible; `reciprocity_impact_level` implements it. And Table B.2 of Part 2
tabulates the octave-band $L_\mathrm{n}$ of four monolithic floors calculated the way
this module does, a useful sanity target for a new floor build-up.

## `in_situ_element()` parameters

| Parameter | Type | Units | Range / default | Notes |
| :--- | :--- | :--- | :--- | :--- |
| `element` | `HomogeneousElement` | — | — | Area, side lengths, $m'$, $f_\mathrm{c}$, $\eta_\mathrm{int}$, $\sum l_k \alpha_k$, and optionally $\rho$/$c_\mathrm{L}$ |
| `frequencies` | array | Hz | > 0 | Band centres |
| `fraction` | int | — | `3` (one-third octaves, default) / `1` (octaves) | Sets the band that carries the $f \approx f_\mathrm{c}$ branch |
| `resonant_only` | bool | — | default `False` | Drops the forced-transmission term below $f_\mathrm{c}$ (Annex B.1, flanking paths) |
| `fluid` | `Fluid` | — | default `EN12354_AIR` | The 340 m/s of Annex A with the 1.29 kg/m³ of Annex B; pass a computed `fluids.air()` to predict in the air of the building |

## `airborne_flanking_path()` / `impact_flanking_path()` parameters

| Parameter | Type | Units | Range / default | Notes |
| :--- | :--- | :--- | :--- | :--- |
| `label` | str | — | — | Display name of the path |
| `kind` | str | — | `'Ff'` / `'Df'` / `'Fd'` | Airborne only; the impact builder is always `Df` |
| `element_i` / `element_j` | `InSituElementResult` | — | — | Source-room and receiving-room elements (`floor` / `element_j` for impact) |
| `vibration_reduction_index` | float or array | dB | — | $K_{ij}$ of this path |
| `coupling_length` | float | m | > 0 | Junction coupling length $l_{ij}$ |
| `separating_area` | float | m² | > 0 | $S_\mathrm{s}$ (airborne only) |
| `delta_r_i` / `delta_r_j` | float or array | dB | default `0` | Lining improvements |
| `delta_l` | float or array | dB | default `0` | Floor-covering improvement (impact only) |

## Detailed prediction report (`.report()`)

`DetailedAirborneResult.report()` and `DetailedImpactResult.report()` write the
per-band counterpart of the simplified prediction fiches: the same one-page
layout, a basis line naming ISO 12354-1/-2:2017 Clause 4.2, the per-path
share-of-energy table beside the per-band path-contribution figure, the boxed
$R'_\mathrm{w}$ / $L'_\mathrm{n,w}$, the detailed model's 1.5 dB to 2.5 dB standard deviation and a
PASS/FAIL verdict against a `requirement`. `verbose=True` annexes the band in
which each path peaks. Both need the ISO 717 rating, so the spectrum must cover
100 Hz to 3150 Hz (or 125 Hz to 2000 Hz in octaves).

```python
res.report("airborne-prediction.pdf")   # R'w, thirteen paths
imp.report("impact-prediction.pdf")     # L'n,w, five paths
```

The example fiches are regenerated with `make reports` and kept rendered in the
repository; click a preview to open the PDF. Both show the Annex L building of
this page: the same elements, junctions and floating floor the tables above are
built from.

**Report:** [ISO 12354-1 detailed airborne prediction example report (PDF)](https://docs.jmrp.io/phonometry/media/iso12354_detailed_airborne_example.pdf): One-page detailed airborne prediction fiche for the ISO 12354-1:2017 Annex L building: an identification header with the 20 m2 separating floor and the two 50 m3 dwellings, the share-of-energy table of all thirteen paths (the 2d flanking path largest at 18.0 %, the direct Dd path 6.3 %), the per-band figure stacking each path's contribution under the apparent sound reduction index curve from 50 Hz to 5 kHz, the boxed predicted R'w = 57 dB, the detailed model's 1.5 dB to 2.5 dB standard deviation, and a PASS against a requirement of 52 dB.

Detailed airborne prediction fiche (DetailedAirborneResult.report), the thirteen paths of the Annex L building and their energy shares.

**Report:** [ISO 12354-2 detailed impact prediction example report (PDF)](https://docs.jmrp.io/phonometry/media/iso12354_detailed_impact_example.pdf): One-page detailed impact prediction fiche for the same building in the impact direction: the identification header, the share-of-energy table of the direct path and the four flanking paths, the per-band figure of each path's contribution under the apparent normalized impact sound pressure level curve, and the boxed predicted L'n,w = 41 dB with the detailed model's standard deviation and the verdict against the requirement.

Detailed impact prediction fiche (DetailedImpactResult.report), the same building excited by the tapping machine.

## What this guide covers

- **Covered.** The Clause 4.2 per-band chain for **Type A** elements: the Annex B
  homogeneous-element sound reduction index and impact level and the Annex C
  in-situ conversion through `HomogeneousElement`, `in_situ_element` and
  `perimeter_absorption_coefficient`; the transfer of *measured* laboratory data
  through `laboratory_total_loss_factor`, `structural_reverberation_time`,
  `in_situ_reduction_index` and `in_situ_impact_level` (Formula 9 and Part 2
  Formula 5); the Formula (10) in-situ junction drop; the Formula (14)/(15)
  direct and flanking airborne paths through `direct_reduction_index` and
  `airborne_flanking_path` and their impact twins `direct_impact_level` and
  `impact_flanking_path`; the Formula (16)/(17) **Type B** routes and the Annex
  B.1/B.2 resonant correction through `resonant_sound_reduction_index`; the
  assembly and ISO 717 rating through `detailed_airborne_prediction` and
  `detailed_impact_prediction`; the Part 2 B.3/B.4 reciprocity identity through
  `reciprocity_impact_level`; and the two fiches through `.report()`. The
  airborne model is stated by Clause 5 to carry no bias error and a standard
  deviation of 1.5 dB to 2.5 dB for buildings of homogeneous elements, assuming
  good workmanship: the lower figure when every aspect is taken into account,
  the higher for complex situations or when the structural reverberation time is
  neglected.

- **Not covered.** Only homogeneous Annex B elements are calculated here; the spectrum of a
  lightweight, double or composite element is an input, from [Predicting Panel
  Sound Insulation](https://docs.jmrp.io/phonometry/buildings/design/panel-sound-insulation/) or
  from a test report. The Annex E junction catalogue itself lives on [Predicting
  Sound Insulation](https://docs.jmrp.io/phonometry/buildings/design/insulation-prediction/), and the
  Part 1 Annex D lining improvement on [Predicting Resilient-Layer
  Performance](https://docs.jmrp.io/phonometry/buildings/design/resilient-layers/); the Annex D and F
  default tables of EN 12354-5 are implemented nowhere. The Type B branch is
  shown from its standard inputs rather than worked end to end. Airborne
  transmission through cavities and suspended ceilings is not carried at all,
  and clause 3.3.4 Note 1 warns it "can contribute to or even dominate" there.
  And the ISO 717 rating needs the whole spectrum: 100 Hz to 3150 Hz in
  one-third octaves, or 125 Hz to 2000 Hz in octaves.

## See also

- [Predicting Sound Insulation (EN 12354)](https://docs.jmrp.io/phonometry/buildings/design/insulation-prediction/): the
  simplified single-number model and the Annex E junction catalogue this page
  builds on.
- [Laboratory Flanking Transmission (ISO 10848)](https://docs.jmrp.io/phonometry/buildings/insulation/flanking-lab/): where the
  measured $K_{ij}$ and the equivalent absorption length come from.
- [Insulation Ratings (ISO 717)](https://docs.jmrp.io/phonometry/buildings/insulation/insulation-ratings/): the reference-curve
  engine that turns the predicted spectrum into $R'_\mathrm{w}$ / $L'_\mathrm{n,w}$.
- [Field Insulation Measurement (ISO 16283)](https://docs.jmrp.io/phonometry/buildings/insulation/insulation-field/): the built
  result the prediction is checked against.
- [Dynamic stiffness of resilient materials (EN 29052-1)](https://docs.jmrp.io/phonometry/materials/resilient/dynamic-stiffness/):
  the $s'$ behind the floating floor's resonance frequency.
- API reference: [`building.prediction.detailed_model`](https://docs.jmrp.io/phonometry/reference/api/building/detailed-model/).
- Theory: [Sound insulation and absorption, predicted](https://docs.jmrp.io/phonometry/reference/theory/rooms-buildings/#sound-insulation-and-absorption-predicted-en-12354-1-2-6-bies-cremer-hopkins): the direct and flanking path algebra the detailed model itemises.

## References

- **Hopkins, C. (2007).** *Sound insulation*. Butterworth-Heinemann. https://doi.org/10.4324/9780080550473
  The physical background of every term of the detailed model: radiation efficiency, structural reverberation time, the equivalent absorption length and the statistical-energy-analysis footing of the path summation. ISBN 978-0-7506-6526-1.
- **International Organization for Standardization. (2017).** [*Building acoustics — Estimation of acoustic performance of buildings from the performance of elements — Part 1: Airborne sound insulation between rooms*](https://www.iso.org/standard/70242.html) (ISO 12354-1:2017).
  Clause 4.2 defines the detailed per-band airborne model; Annex B the calculated element performance and radiation factors, Annex C the structural reverberation time, Annex E the junction indices and Annex L the worked examples.
- **International Organization for Standardization. (2017).** [*Building acoustics — Estimation of acoustic performance of buildings from the performance of elements — Part 2: Impact sound insulation between rooms*](https://www.iso.org/standard/70243.html) (ISO 12354-2:2017).
  Clause 4.2 defines the detailed per-band impact model; Annex B the calculated impact level of a bare floor, Annex C the floating-floor improvement and Annex G the worked examples sharing ISO 12354-1 Annex L's building.
- **Vigran, T. E. (2008).** *Building acoustics*. CRC Press. https://doi.org/10.1201/9781482266016
  The homogeneous-element transmission model and the impact-level closed form behind Annex B of both parts. ISBN 978-0-415-42853-8.
