DOI: 10.61782/fa.2023.

0585

MODELING THE ISO 226:2023 EQUAL-LOUDNESS-LEVEL CONTOURS

BY STANDARDIZED LOUDNESS METHODS

Roland Sottek1∗ Thiago Lobato1 Moritz Bender2 Julian Becker1

1
HEAD acoustics GmbH, 52134 Herzogenrath, Germany
2
RWTH Aachen University, 52062 Aachen, Germany

ABSTRACT coustic analyses such as roughness [4] and sharpness [5].

Consequently, it is essential that modern loudness meth-
The standard ISO 226 [1] specifies combinations of sound ods can adapt to improvements in equal loudness con-
pressure levels and frequencies of pure continuous tones tours, as described in ISO 226. In this brief paper, we ex-
that are perceived as equally loud by human listeners. amine the performance of modern standardized loudness
These equal-loudness-level contours are part of the foun- methods – ISO 532-1, ISO 532-3 and ECMA 418-2 – with
dation of the psychoacoustic research and are constantly respect to the latest version of the ISO 226 standard.
evolving. The recent third version of ISO 226:2023 con-
tains some minor corrections to the second version of
ISO 226:2003. In this paper, we describe these changes 2. CHANGES IN ISO 226:2023
and evaluate how well current standardized algorithms The updates in ISO 226:2023 include clarification of the
perform in generating the new ISO 226:2023 curves. scope of the standard, an updated bibliography, an align-
To this end, we compare the loudness methods pub- ment with the 2019 revised edition of ISO 389-7 [6] re-
lished in ISO 532-1:2017 (Zwicker method) [2], ISO 532- garding the 0-phon data, and corrections of systematic er-
3:2023 (Moore, Glasberg and Schlittenlacher method) [3], rors in the equal-loudness-level contours. The updated in-
and the Sottek Hearing Model Loudness method pub- troduction emphasizes that the equal-loudness-level con-
lished in ECMA 418-2 2nd edition (2022) [4]. It should tours are only applicable to pure tones, since there are in-
be noted that the Zwicker method aims to match the sufficient data to estimate their validity for other sound
equal-loudness-level contours of the first edition of ISO types such as broadband noise and noise with promi-
226:1987, whereas the other methods aim to match the nent tones. Thus, while ISO 226 serves as a validation
data of the second version of ISO 226:2003. and requirement for loudness methods, appropriate and
validated loudness models should be employed for other
Keywords: Equal-Loudness-Level Contours, Sottek
sound types. The systematic errors previously present in
Hearing Model, ISO 226:2023, Loudness, ECMA 418-2.
the standard resulted in a maximum change of less than
0.6 dB, which is relatively small. These corrections re-
1. INTRODUCTION sulted in an updated formula for determining the sound
Equal-loudness-level contours provide the foundation for pressure level from the loudness level and an updated ref-
many loudness methods, which underpin various psychoa- erence table with new parameters. The new formula that
relates sound pressure level Lf in dB and loudness level
*Corresponding author: roland.sottek@head- LN in phon is as follows:
acoustics.com.
Copyright: ©2023 Sottek et al. This is an open-access
10 LN
article distributed under the terms of the Creative Commons Lf = · lg{(4 · 10−10 )(0.3−αf ) · [100.03 phon − 100.072 ]
Attribution 3.0 Unported License, which permits unrestricted αf
use, distribution, and reproduction in any medium, provided the Tf +LU

original author and source are credited. + 10αf 10 dB } dB − LU (1)

10th Convention of the European Acoustics Association

Turin, Italy • 11th – 15th September 2023 • Politecnico di Torino
6315
 ISO 532-1 Equal-Loudness-Level Contours
100
in which Tf is the threshold of hearing, αf is the ex-
ponent for loudness perception, LU is the magnitude of a 90
80 phon

linear transfer function normalized at 1 kHz in dB. The 80

new values of αf , LU and Tf are available in tabular for- 70

60 phon

mat in ISO 226:2023 [1]. 60

SPL/dB
50
40 phon

3. SOTTEK HEARING MODEL 40

30
20 phon
The ECMA 418-2 standard is based on the Sottek Hear-
20
ing Model [7], which accounts for numerous aspects of
10
auditory perception, including outer/middle ear filtering, ISO 532-1
ISO 226:2023

auditory filter bank, and the compressive non-linearity of 0

100 200 500 1k 2k 5k 10k
f/Hz
the human auditory system. The Sottek Hearing Model,
which serves as the foundation for the ECMA 418-2 stan- Figure 1: Equal-loudness-level contours of the ISO
dard, was recently employed in the development of a new
532-1 standard compared to ISO 226:2023.
standardized loudness metric that considers both the tonal
and noise loudness of signals. This new loudness stan- ISO 532-1 Equal-Loudness-Level Contours (1987)
100
dard effectively handles the loudness of noise signals with
90
sub-critical bandwidth [8] and harmonic components [9]. 80 phon

80

70
60 phon
4. METHODOLOGY 60
SPL/dB

To evaluate the performance of the algorithms, equal- 50

40 phon

loudness-level contours of 20 phon, 40 phon, 60 phon and 40

80 phon were calculated, for the frequency range from 30

20 phon
100 Hz to 10 kHz. All values for this selection are de- 20
fined as valid in the ISO 226:2023 standard. For each of
10
the three methods ISO 532-3, ISO 532-1, and ECMA 418- ISO 532-1
ISO 226:1987
0
2 we calculated the corresponding sone value at 1 kHz for 100 200 500 1k 2k 5k 10k
f/Hz
a given phon level and then determined the necessary dB
level for each frequency so that the algorithm yielded the Figure 2: Equal-loudness-level contours of the ISO
same sone value. This approach enabled us to generate the 532-1 standard compared to ISO 226:1987. Here the
equal-loudness-level contours of the algorithms. For all
RMS difference is 2.84 dB.
methods, we calculated the Root-Mean-Squared (RMS)
difference between their results and the reference values
of the ISO 226:2023 equal-loudness-level contours. Table 1: Difference of each method with respect to
the reference equal-loudness-level contours in ISO
226:2023.
5. RESULTS
The results of each method can be seen in Fig. (1) to Method RMS difference [dB]
(4), while the RMS difference for each result is shown in
ISO 532-1 5.55
Table 1.
Our comparison revealed that the Zwicker method, ISO 532-3 3.04
originally designed to match the contours of ISO ECMA 418-2 1.57
226:1987 (as can be clearly seen in Fig. 2), exhibits a
greater deviation from the new ISO 226:2023 standard
compared to the other two methods. We also verify a fluc- in Annex D of the DIN 45631/A1 standard [10] (the ba-
tuation in the results. This is a known effect described sis for ISO 532-1 [2]), from which we take the verbatim

10th Convention of the European Acoustics Association

Turin, Italy • 11th – 15th September 2023 • Politecnico di Torino
6316
 ISO 532-3 Equal-Loudness-Level Contours
100
a closer agreement with the updated contours. For middle
90
80 phon frequencies between 1 kHz and 5 kHz, the ECMA 418-2
80 standard has a considerably better match than all other ap-
70
60 phon
proaches, which overestimate the loudness of tonal com-
60
ponents in this range.
It is important to note that these results pertain to
SPL/dB

50
40 phon
raw sine waves under free-field conditions. For noise sig-
40
nals, particularly those with a sub-critical bandwidth, we
30
20 phon
recommend using the ECMA 418-2 standard, as it bet-
20
ter models this type of data, as demonstrated in [8]. The
10
ISO 532-3
Sottek loudness also has a slightly better accuracy for
ISO 226:2023
0 multi-tone components [9], and since it is additionally the
100 200 500 1k 2k 5k 10k
f/Hz approach with the lowest error here, we see it as a domi-
nant approach with respect to the results, being better or
Figure 3: Equal-loudness-level contours of the ISO equal in all possible type of signals we investigated so far.
532-3 standard compared to ISO 226:2023. Another interesting point of comparison between the
methods is their calulation time. We used the software
ECMA 418-2 Equal-Loudness-Level Contours
100
ArtemiS SUITE from HEAD acoustics to benchmark all
90
80 phon methods and to show their normalized calculation time re-
80 sult in Table 2. There we see that the ISO 532-1 is the
70
fastest method, followed by ECMA 418-2 and finally ISO
60 phon

60
532-3, which is one order of magnitude slower than the
other methods. For the ECMA 418-2 standard, most of
SPL/dB

50
40 phon
the calculation time is due to the separation of tonal and
40
noise components to calculate the corresponding partial
30
20 phon tonal and noise loudness values. The algorithm of ISO
20 532-3, on the other hand, is very time consuming, and
10
ECMA 418-2
even though the code in ArtemiS SUITE is highly opti-
0
ISO 226:2023
mized, the calculation still takes a long time.
100 200 500 1k 2k 5k 10k
f/Hz

Table 2: Normalized calculation time for each

Figure 4: Equal-loudness-level contours of the method as calculated with the software ArtemiS
ECMA 418-2 standard compared to ISO 226:2023. SUITE from HEAD acoustics.

text: ”For a sliding sine without superimposed noise, one Method Normalized calculation time
would initially expect a ”smooth” progression of loud- ISO 532-1 1
ness according to hearing. However, the actual calculated ISO 532-3 22
course is characterized by certain fluctuations. The rea-
ECMA 418-2 6
son for this is to be found in the signal processing of the
third-octave filter bank.” Depending on the characteristics
of the implemented filters, deviations of varying magni-
tude from the ”ideal value” can occur in the transition fre-
6. CONCLUSION
quency ranges in the addition of the partial loudnesses.
The method thus must be used with care, especially for In this paper, we discussed the changes made in ISO
sounds at the edges of the third-octave filters. 226:2023 compared to its 2003 predecessor and explained
In contrast, the Moore, Glasberg, and Schlittenlacher the main reasons for these changes. Using the new equal-
method (ISO 532-3:2023) and the Sottek Hearing Model loudness-level contours, we evaluated the efficiency of
Loudness method (ECMA 418-2 2nd edition) demonstrate various current loudness standards in replicating these

10th Convention of the European Acoustics Association

Turin, Italy • 11th – 15th September 2023 • Politecnico di Torino
6317
curves. [7] R. Sottek, Modelle zur Signalverarbeitung im men-
Our results show that ECMA 418-2 2nd Edition out- schlichen Gehör. PhD thesis, University of Aachen,
performes the others with an RMS error of 1.57 dB, fol- 1993.
lowed by ISO 532-3 with an error of 3.04 dB and ISO
[8] R. Sottek, T. Lobato, and J. Becker, “Loudness of
532-1 with an error of 5.55 dB. The first two methods re-
sounds with a subcritical bandwidth: improved pre-
produce the curves with reasonable accuracy, while ISO
diction with the concept of tonal loudness,” in DAGA,
532-1 shows larger deviations.
(Stuttgart), 2022.
The poorer performance of ISO 532-1 was to be ex-
pected since it was designed for matching ISO 226:1987, [9] T. Lobato and R. Sottek, “Modeling the perceived
which incorporated substantially different curves. More- tonal loudness of multiple tonal components,” in
over, ISO 532-1 exhibits fluctuations in its results due to DAGA, (Hamburg), 2023.
the steepness of the third-octave filters used, which can be
[10] “DIN 45631/A1: Calculation of loudness level and
particularly challenging for sounds located at the edges of
loudness from the sound spectrum – Zwicker method
the filters or those with broad frequency modulations.
– Amendment 1: Calculation of the loudness of time-
In the mid-frequency range of 1 kHz to 5 kHz, the
variant sound,” 2010.
Sottek loudness displays significantly superior accuracy
than other methods, which tend to overestimate the loud-
ness of signals at these frequencies. In terms of com-
putational efficiency, ISO 532-1 proves to be the fastest
method, followed by ECMA 418-2, which is six times
slower, and ISO 532-3, which is 22 times slower even
with a highly optimized code, making its application to
extended signals cumbersome.
Currently, the Sottek loudness defined in ECMA 418-
2 has shown equivalent or superior performance in all ex-
periments we have conducted [4, 8, 9] and has established
itself as the dominant variant in terms of result quality.

7. REFERENCES
[1] “ISO 226: Acoustics – Normal equal-loudness-level
contours,” 2023.
[2] “ISO 532-1: Methods for calculating loudness, Part 1:
Zwicker method,” 2017.
[3] “ISO 532-3: Methods for calculating loudness, Part 3:
Moore-Glasberg-Schlittenlacher method.” in prepara-
tion.
[4] “ECMA 418-2 2nd Edition: Psychoacoustic metrics
for ITT equipment, Part 2: Models based on human
perception,” 2022.
[5] “DIN 45692: Measurement technique for the simula-
tion of the auditory sensation of sharpness,” 2009.
[6] “ISO 389: Acoustics — Reference zero for the cal-
ibration of audiometric equipment — Part 7: Refer-
ence threshold of hearing under free-field and diffuse-
field listening conditions,” 2019.

10th Convention of the European Acoustics Association

Turin, Italy • 11th – 15th September 2023 • Politecnico di Torino
6318