Content uploaded by Anna Rieger
Author content
All content in this area was uploaded by Anna Rieger on Aug 28, 2023
Content may be subject to copyright.
The contribution of selected auditory sensations to the prediction of
preference judgements for consonant and dissonant sounds
Anna Rieger1
Laboratory for Development Applications, OTH Regensburg
Seybothstraße 2, 93053 Regensburg, Germany.
Acoustics Group, Cluster of Excellence "Hearing4all", Carl-von-Ossietzky University Oldenburg
Ammerländer Heerstraße 114-118, 26129 Oldenburg, Germany.
Steven van de Par2
Acoustics Group, Cluster of Excellence "Hearing4all", Carl-von-Ossietzky University Oldenburg
Ammerländer Heerstraße 114-118, 26129 Oldenburg, Germany.
Hans-Peter Rabl
Laboratory for Development Applications, OTH Regensburg
Seybothstraße 2, 93053 Regensburg, Germany.
Arne Oetjen3
ABSTRACT
Product sounds with clearly audible tonal components are often perceived as unpleasant or annoying.
If different simultaneously operating aggregates are present in a system, for example vehicle engines
and gearboxes, the interaction of tonal components, similar to music, can evoke additional sensations
in human auditory perception. Supplementary to a pronounced tonality, such sounds can also yield
distinct degrees of consonance or dissonance between tones. Previous studies showed that the
perceived dissonance had a high impact on preference judgements for sounds with similar tonality.
In experiments of the present study, sounds that differed in tonality were rated with respect to the
auditory sensations sharpness, tonality and dissonance by one group of participants while another
group only carried out a preference task. Thereout a model for predicting perceived preference is
derived from the subjective judgements of auditory sensations. The performance of the preference
predictions based on subjective judgements will be compared against purely model-based predictions
using different algorithms for acoustic attributes.
1. INTRODUCTION
The determination and optimization of sound quality is a mature aspect in developing technical
products. In the view of electric vehicles, new NVH (Noise Vibration Harshness) issues arise for
the developers, mostly due to the absence of common combustion engines. Powertrain-related
components like electric motors, inverters and gears, operating noise of smaller electric aggregates
1anna.rieger@st.othr.de, anna.rieger@uol.de
2steven.van.de.par@uol.de
3arne.oetjen@gmx.de
and also artificial sounds like the legally required AVAS (Acoustic Vehicle Alerting System) give the
vehicles their specific and complex tonal signature, in- and exterior [1,2]. It seems obvious that with
this combination of tonal orders also frequency intervals between adjacent tonal components can
occur, for example intervals between the orders of an electric motor and gearbox due to the gear ratio.
A concept for the interaction between multiple tonal components is known from music: consonance
and dissonance. This concept could also be applied as a sound quality measure for environmental
sounds with multiple tonal components. In a previous study [3], the authors examined the dissonance
perception for multi-tone sounds and therein have developed a psychoacoustic concept to predict the
perceived dissonance of sounds. It has also been questioned if a distinct perception of dissonance
influences the general pleasantness of multi-tone sounds. Zwicker and Fastl [4] state that the
annoyance of sounds is a measure that depends on their loudness, timbre and temporal structure. In
several studies, e.g. [5–7], it was found that the perceived annoyance increases with the amount of
tonal content. In previous research of the authors [3, 8] also the significant influence of dissonance,
among the psychoacoustic sensations loudness and sharpness, on preference judgements could be
determined for sounds with similar tonality. Due to the fact that sounds can only be consonant
or dissonant if there are audible tones and therefore a certain degree of tonality, the interaction of
tonality and dissonance is of special nature. Therefore, the influence of tonality for consonant and
dissonant sounds on preference judgements was analyzed more detailed. This contribution gives an
insight in parts of the results of two different experiments about the influence of the psychoacoustic
sensations sharpness, tonality and dissonance on preference judgements for multi-tone sounds. A
method to predict preference judgements based on subjective ratings for those selected sensations
along with the corresponding correlations of related psychoacoustic prediction models will be shown.
2. METHODOLOGY AND EXPERIMENTAL RESULTS
Two different experiments were carried out to analyze the influence of selected auditory sensations
on the general sound quality of multi-tone sounds. One group of participants had the task to rate
the sharpness, tonality and dissonance of sounds (group single attributes), while the other group
of participants rated those sounds in an adaptive alternative forced choice procedure regarding
their preference (group preference). Each group was not informed about the other’s task. By that
experimental design it became possible to analyze whether the preference of sounds can be modelled
by appropriate, selected auditory sensations using a relatively simple approach.
2.1. Stimuli
The complex sound character of an electric vehicle was simulated by synthetically generated sounds in
a somewhat simplified way. A distinction was made between test and reference sounds that all shared
f0=1046.5 Hz. This frequency range was chosen to roughly match the spectra of electric vehicles [1].
Each test sound contained two harmonic tone complexes (f0,2·f0,3·f0), with a 1/fovertone decay,
that were spaced by a certain frequency interval of the equal-tempered tuning system x=2n
/12 and had
a bandpass-filtered pink background noise. This scenario could be seen as an abstract version of two
tonal sounding machines operating simultaneously. In addition, there were four reference sounds,
all of them with the same background noise as described. Reference one and two were sinusoidal
conditions with f0and different SNRs (ref. 1 low , ref. 2 high SNR). Reference three was a harmonic
tone complex with f0,2·f0,3·f0and reference four a consonant dyad (constructed like the test
sounds). The SNR specified for the sounds is the ratio of the tone information to the background
noise and was approximately 16 dB for all sounds with intervals, resulting in a pronounced tonality.
Reference sounds two and three also had a pronounced tonality. As frequency intervals x, the intervals
or semitones n=±10,±6,±5,±1 known in music and n=+4 were applied in the sounds. The different
conditions were labeled correspondingly as ±S eventh(d),±T ritone(d),±F ourth(c),±S econd(d) for
test and +T hird(c) for reference sound four, with the indication whether the interval is basically
known to be consonant (c) or dissonant (d). All stimuli had a duration of 2 s and got played back
in a sound proof cabin by headphones of the type Sennheiser HD 650. The sounds were iteratively
brought to equal loudness according to DIN45631/A1 [9] to reference stimulus four (65 dB SPL)
at the beginning and during the experiments when changes of signal properties occurred, compare
sections 2.2. and 2.3. for the exact experimental designs. The influence of loudness can therefore
be neglected in these investigations what makes it thus possible to derive a detailed analysis of the
influence of other psychoacoustic sensations in sound quality perception.
By an analysis of the sounds concerning their relevant psychoacoustic features due to their
spectral properties and the fact that loudness is equal for every stimulus, possible remaining
influencing factors are determined by sharpness, tonality and dissonance. As an example: ±S eventh
shows that the interval tones were shifted to frequencies <f0and once >f0which produced an
additional effect of sharpness in the sounds. By having a different number of tones and different
SNRs, the tonality is assumed to play a role in the assessment of those sounds. With additional
frequency intervals between the tones, that are at least known to play a role in the perception of
music, also dissonance is suggested as an influencing factor when it comes to the assessment of the
preference or pleasantness of sounds.
2.2. Preference Judgements
In an adaptive, 2-interval Alternative Forced Choice (AFC) procedure (utilised toolbox by [10]), a
first group of N=26 subjects was asked about their preference between pairs of the test and the
reference sounds. The test sounds were presented in comparison to the reference sounds based on the
task Which interval would you prefer as a vehicle sound?. In every comparison, the participants had
to decide which one of the sounds they would prefer, what led to a stepwise reduction of the SNR in
the less preferred sound in the program background. Thus, with each decision, the tone level excesses
of all tonal components were reduced relative to the background noise. Thereby it is known that
the level excess of tonal components is a significant feature when rating the perceived tonality [5].
With every gradual reduction of the SNR, the loudness was again iteratively brought to the reference
loudness (cf. subsection 2.1.) defined in the beginning. Rating a pair of a test and a reference sound
consequently led to finding the Point of Subjective Equality (PSE). As a result, necessary reductions
in SNR, or in other words, the necessary reductions of tone levels to come to that point where a test
and a reference sound are equally preferred, are derived.
2.3. Dimensional Ratings
A second group consisting of N=24 participants was asked to rate all experimental stimuli (test
and references) with different discrete settings of SNR reductions (starting from 0 dB SNR reduction
in steps of 5 dB) on partially labeled categorical scales regarding their perceived sharpness, tonality,
and dissonance. Based on the SNR reductions group preference rated in the PSE experiment, the
range of the discrete SNR settings for the sounds in the categorical rating experiment was adapted,
where the largest reductions in SNR to be rated are 30 dB for the stimuli with +Tritone,+Seventh
intervals. Each scale ranged from 1 not sharp, tonal, dissonant to 9 very sharp, tonal, dissonant. As
the first group of participants adjusted the SNRs to find pairs of equal preference between test and
reference sounds in their test procedure, this second group rated different discrete settings of SNRs in
all test and reference sounds due to selected psychoacoustic sensations. In that way also the course
of different psychoacoustic sensations for sounds that get gradually changed in their tonal signature
could be analyzed.
Figure 1 shows the mean categorical ratings with standard errors of participant group single
attributes (N=24) from left to right: Sharpness, tonality and dissonance for all test sounds. A
reduced set of the data is shown here. The maximum standard errors are in the magnitude of
one category. By the fact that all test and reference sounds were evaluated with different discrete
reductions in SNR (from 0 up to 30 dB), a kind of map emerges for each test sound and each
psychoacoustic dimension. Out of this data it can be observed how the psychoacoustic dimensions
sharpness, tonality and dissonance are assessed for discrete reductions in tone level excesses. Note
that for all reductions in SNR shown in the experimental data, the tonal components were well above
their detection threshold. It can be observed that the sharpness of the sounds in general decreases
with decreasing SNR (follow the red to light blue line). The characteristic increase of sharpness
however arises from the difference of spectral components in the sounds. Here, the stimulus with
-Seventh as interval has the lowest spectral components overall, whereas the +Seventh has the highest
overall, compare the intervals described in subsection 2.1.. As expected, the tonality of all sounds
decreases with a decrease in SNR, follow again the red to light blue line. In principle, sounds with
the same SNR setting are perceived quite similar in tonality, follow every colored line for all test
sound intervals. It is visible that dissonance ratings significantly depend on the frequency interval in
the sounds. Just as in other studies, the interval theory generally known from music can be observed.
-Seventh
-Tritone
-Fourth
-Second
+Second
+Fourth
+Tritone
+Seventh
2
4
6
8
Categorical Ratings
Sharpness
0 dB
5
10
15
20
25
30
-Seventh
-Tritone
-Fourth
-Second
+Second
+Fourth
+Tritone
+Seventh
Testsound
Tonality
-Seventh
-Tritone
-Fourth
-Second
+Second
+Fourth
+Tritone
+Seventh
Dissonance
Figure 1: Categorical ratings (1-9) as mean and standard errors for N=24 participants. From left
to right, the dimensions sharpness, tonality and dissonance are plotted over the respective test sound
interval. Only the data of the test sounds is shown here.
Since the loudness of the stimuli was balanced and according to the previous analysis,
the main psychoacoustic factors influencing the sounds were suggested to be sharpness, tonality
and dissonance, a preference indicator was formed according to the following, relatively simple,
assumption:
P−Indicator ∼Sa+b·Tc+d·De(1)
For the application in sound quality assessment, the terms S,T, and Dshould be and are normally
substituted by computational algorithms. For the investigations shown here, however, the subjective
data on sharpness, tonality and dissonance, derived by the categorical rating experiment, is used. It is
analyzed how well the preference indicator is able to predict the preference judgements of participant
group preference on a subjective basis. As the performance of this predictor depends on the suitability
of the three chosen auditory sensations, an insufficient selection of input variables would result in a
poor prediction quality.
2.4. Preference Predictions "in-the-loop"
Using the categorical data from sharpness, tonality and dissonance (group single attributes), it
was attempted to predict the preference judgements (group preference) by applying the rule from
equation 1. To test the model performance, the preference indicator is embedded in-the-loop in the
interface of the PSE experiment. Instead of a participant evaluating the pairs of sounds in the program
interface by its selections, the numerical output of the preference indicator is used to evaluate the
preference differences of the sounds. As for the experimental run with the participants of group
preference, after each evaluation of a pair of sounds by the preference indicator, the SNR of the
less preferred sound is reduced in the program background in order to reach the point of subjective
equality in preference. The preference indicator would rate a pair of sounds until the adaptive
procedure converges to the PSE. This will be the case if the numerical outputs for both sounds
are approximately equal for a certain reduction in SNR in one of them. In that way, the necessary
reductions in tone level excesses or reductions in SNR are derived by the algorithm and can be
compared to the mean ratings of participant group preference. All values for sharpness, tonality and
dissonance can be either taken directly or get interpolated from the data shown in figure 1 for each
setting of SNR and each psychoacoustic dimension. The smallest reduction in SNR to be evaluated
by the predictor was 0.125 dB, what makes the prediction very accurate compared to a real human
subject. A higher numerical output of the indicator would indicate a lower preference and vice versa.
Figure 2 shows the mean results and standard errors of the PSE experiment conducted by
participant group preference (solid lines) and the corresponding results for the experiment when the
model-in-the-loop would have rated the pairs of sounds and thereby adjusted the SNRs (circles in alike
colors). A reduced set of data is shown here. Every data point depicts the state when a pair of sounds
(a test and a reference sound) are equally preferred. Thus, the SNR reductions are plotted over the
intervals in the test sounds. As an example: The SNR in the sound with the -Seventh as interval had
to be reduced by more than 15 dB to be equally preferred to reference stimulus one. Consequently,
positive values indicate a reduction of the SNR in the test sounds, while negative values indicate the
reduction of the SNR in a reference sound. 0 dB would indicate that the SNR in none of the sounds
had to be reduced, as they were found to be equally preferable initially. That was roughly the case for
-Seventh in comparison to reference three.
-Seventh
-Tritone
-Fourth
-Second
+Second
+Fourth
+Tritone
+Seventh
Testsound
-10
0
10
20
30
SNR reduction in dB
Ref. 1
Ref. 2
Ref. 3
Ref. 4
Figure 2: Solid lines: Results of the PSE experiment as mean and standard errors for participant
group preference (N=26), plotted as necessary reductions in SNR to build pairs of equal preference
in dB over the respective test sound interval. Circles in alike colors: Prediction of preference
judgements by the preference indicator (equation 1) relying on subjective data of participant group
single attributes (N=24, figure 1).
With respect to reference one, the SNRs in all test sounds had to be reduced to achieve the
points of equal preference. For reference two, the equivalent sinusoidal condition as reference one
only with higher SNR, it is observable that the test sounds needed less reduction in SNR for being
equally preferred. As the tone level excess of the sinusoidal component in reference one was much
smaller than that of reference two, the influence of tonality is assumed to be dominating here. For
the harmonic tone complex (reference three) no remarkable difference is observable compared to
reference two, except for the -Seventh,-Tritone conditions. These differences could be explained by
the influence of harmonics in the harmonic tone complex (ref. 3) compared to the sinusoidal condition
(ref. 2). The results for reference four, the consonant dyad, show the influence of additional frequency
intervals between tones. There were some test sounds (-Seventh,-Tritone,-Fourth) for which the
reference four had to be reduced in SNR to reach the points of equal preference. As also previous
results have shown [3, 8], sounds with the same frequency intervals and therefore theoretically
similar degrees of consonance or dissonance, e.g. ±Seventh,±Tritone,±Fourth respectively, are rated
very differently with respect to preference. Note that for e.g. ±Seventh (2−10
/12)−1=210
/12 holds.
This is mainly due to the different placement of the spectral components, i.e. the differently rated
sharpness of the sounds. Likewise, the influence of dissonance on preference judgements can again
be observed in the data, for example directly by the abrupt increase in the curves from -Fourth to
-Second, a commonly known consonant and dissonant interval. It can be observed that almost all
necessary reductions in SNR to build pairs of equally preferred sounds (results of participant group
preference) can be predicted within the standard error by the preference indicator. It should be
emphasized that the indicator was formed by the the categorical ratings of group single attributes,
where sharpness, tonality and dissonance have been measured independently from the preference
judgements of group preference. Those results suggest, as well as Fastl and Zwicker proposed, that
the preference or general pleasantness of sounds can in principle be modeled by an interaction of
relevant psychoacoustic dimensions.
3. PSYCHOACOUSTIC PREDICTION MODELS
The prediction of preference judgements by using the preference predictor as described by equation 1
turned out to function very well on a subjective basis. That would confirm the preliminary assumption
that the sounds are mainly judged due to their differences in sharpness, tonality and dissonance.
In the following, correlations between existing psychoacoustic predictions models, that are either
standardized or in the development and validation stage, and the categorical data for sharpness,
tonality and dissonance are shown in order to foresee the quality of purely model-based predictions,
as it is usually done in sound design.
In the German DIN45692 [11], a computational model for sharpness is standardized that
relies on the calculation of psychoacoustic loudness that again is standardized in DIN45631/A1 and
ISO532-1 [9, 12]. Three different weighting curves for the prediction of sharpness are presented that
consequently lead to (slightly) different sharpness predictions and therefore will get compared for
their performance. Subsequently, the weighting method according to DIN in the main part of the
standard will be specified by DIN/DIN. The method by Aures and v. Bismarck in the annex of the
standard will be labeled by DIN/Aur and DIN/Bis. Standarized versions for tonality calculations can
be found in the German DIN45681 and a model for IT equipment in ECMA-418-2 [13] annex B.
In the literature, drawbacks are described for both models that either concern a limited temporal
resolution for highly transient sounds or the fact that in complex tonal structures with multiple
tones just the most prominent tonal component is evaluated [14, 15]. Therefore, two additional
model attempts for tonality that have been presented in [16, 17] (later labeled as Oetjen et al.) and
in [15, 18, 19] (later labeled as Volk et al.) that were co-developed by the authors2,3are also tested
with regard to their coefficients of determination. For the dissonance phenomenon different theories
about the underlying perception mechanisms are described in literature, a broad overview is given
in [20]. Besides calculation procedures that give a measure for dissonance for known frequency
intervals between two tones, the authors focused on the development of a psychoacoustic concept
for dissonance perception that can be applied on arbitrary audio signals. The basic model consists
of an auditory front-end and a following evaluation of features of beating products in the envelope
domain. Basic model parameters have been set, while dependencies like level scaling are free
parameters to be determined in further validation processes. Therefore, the current basic concept is
also checked for consistency with this new categorical data. In [3, 8] excerpts about the basic model
concept and studies on the influence of dissonance in sound quality measures got described. For a
better visualization and comparison, the model calculations mod got normalized by the range of the
respective categorical data cat by equation 2.
modrescaled =mincat +mod −minmod
maxmod −minmod
·(maxcat −mincat ) (2)
Figure 3 shows the correlation of three selected models from left to right for: Sharpness (DIN45692,
weighting and method according to Aures [11]), Tonality (DIN45681 [5]) and Dissonance in the
current state based on Rieger et al. [3, 8]. The scatter plots show data for the test sounds only and for
discrete settings of SNR reductions from 0 to 20 dB. That range was chosen for a first comparison
as those settings in SNR were rated consistently for all different frequency intervals in the sounds,
compare figure 1.
2468
2
4
6
8
Rescaled Model Predictions
Sharpness DIN/Aur
2468
Categorical Ratings
Tonality DIN
2468
Dissonance Rieger et al.
Figure 3: Correlation of psychoacoustic prediction models and categorical data for sharpness, tonality
and dissonance with SNR reductions from 0 to 20 dB.
The coefficients of determination R2for the model predictions in figure 3 and for other tested
models describing the same sensations are drawn in table 1.
Table 1: Coefficients of determination of psychoacoustic prediction models for sharpness and tonality.
Sharpness Tonality
Model R2Model R2
DIN/DIN 0.03 DIN 0.9
DIN/Aur 0.12 ECMA-418-2 0.79
DIN/Bis 0.09 Oetjen et al. 0.9
Volk et al. 0.93
It can be noted that tonality models achieve the best concordance with the corresponding
experimental data. Nevertheless, the largest difference in explained variance is at about 14% for
the different tested models. All test sounds were stationary sounds with 2 s length with a maximum
of six tonal components. The altered parameter was the tone level excess of all tonal components
in discrete steps, directly influencing the perception of tonality. Therefore, it had also been assumed
that tonality models would yield a quite good accordance to the data shown. A benchmark of tonality
models for more complex sounds concerning their transient behaviour and the suitability to predict
the tonality of more complex tonal structures would give a broader insight in the differences of the
model performances. In the current status, the dissonance model approach already yields a correlation
of 0.7 with the experimental data, although some effects like the dependence on the overall level have
not yet been parameterized. The best-performing model for sharpness predictions shows a coefficient
of determination for this reduced set of data of 0.12. Further analyses are suggested to clarify the
poor correlation between models and data, as the models have reportedly proven to work for a wide
range of sounds, e.g. [4, 11]. In a first analysis it was found that participants more likely rated the
tonal sharpness in the sounds rather than an overall sharpness that can also arise from a broadband
noise. The bandpass-filtered pink background noise was the same for all stimuli at any time. If tone
levels got reduced, the loudness of the sounds was balanced iteratively, what resulted in a stepwise
amplification of the background noise which in turn influenced the calculated sharpness values in that
way. The subjective data suggests that sounds with increasing frequency intervals and therefore higher
spectral tone components get higher sharpness ratings, this can be observed in the general increase
for sharpness from -Seventh to +Seventh in figure 1. The background noise did not seem to play a role
in the assessment of the sharpness for the depicted experiments and those exact sounds. By carrying
out purely model-based predictions of the preference judgements according to equation 1 consisting
of calculated values for sharpness, tonality and dissonance, the sum of squared errors between model-
based predictions and subjective preference data (solid lines, figure 2) at least doubles in comparison
to the predictions by subjective judgements shown in figure 2 in circles. The deterioration of the
model-based predictions in comparison to the subjectively obtained results is mainly due to the major
deviations between calculated sharpness to subjectively rated sharpness.
4. SUMMARY AND OUTLOOK
The contribution of selected auditory sensations on preference judgements for sounds with equal
loudness, multiple tonal components and additional frequency intervals that made them either
consonant or dissonant got analyzed. By a preceding analysis of the stimuli, it was suggested that
sharpness, tonality and dissonance are most probably the relevant psychoacoustic sensations for
forming a judgement for their preference. In addition, the authors assumed that the preference or the
general pleasantness of sounds can be modelled by an interaction of several, relevant psychoacoustic
dimensions. A relatively simple approach to build a preference indicator (eq. 1) was introduced.
In order to derive subjective data, the sounds were rated due to their preference (experiment one:
2-interval Alternative Forced Choice) and perceived sharpness, tonality and dissonance (experiment
two: categorical ratings) by two separate groups of participants that did not know about the task
of the other group. The performance of the preference predictor was then tested using subjective
data for sharpness, tonality and dissonance derived by group single attributes and comparing the
preference predictions of the predictor to those that had been derived from experiment one by
participant group preference. For this purpose, a model-in-the-loop paradigm using categorical
ratings in a predictor algorithm simulating an AFC-experiment was developed. It has been found that
the preference predictions by the model-in-the-loop for the sound set (on a subjective basis) showed
a high similarity with the preference ratings of experiment one and participant group preference. A
comparison between preference predictions based on subjective ratings for different psychoacoustic
sensations and their modeled pendants showed that the approach shows a much higher performance
using purely subjective data. In future work, detailed analyses aiming towards improvements in
the prediction algorithms for sharpness, tonality and dissonance will be necessary for a reliable set
of description variables. For the application in sound design, a reliable prediction of sound quality
is desirable that provides similarly good predictions as the preference predictor has shown on a
subjective basis. Based on the good agreement of the predictions derived by the basic model concept
for dissonance, a validation of further free parameters is aspired in order to exploit the full potential
of the analysis.
REFERENCES
1. H. Tschöke, P. Gutzmer and T. Pfund. Elektrifizierung des Antriebsstrangs: Grundlagen - vom
Mikro-Hyrid zum vollelektrischen Antrieb. Springer, Berlin-Heidelberg, Germany, 2019.
2. D. Lennström, T. Lindbom and A. Nykänen. Prominence of tones in electric vehicle interior
noise. In Proceedings of INTER-NOISE 2013, pages 508-515, Innsbruck, Austria, September
2013.
3. A. Rieger, A. Oetjen, S. van de Par and H.-P. Rabl. Gehörbezogene Modellierung der Dissonanz.
In Proceedings DAGA 2021, pages 1034-1037, Wien, Austria, August 2021.
4. H. Fastl and E. Zwicker. Psychoacoustics: Facts and Models. Springer, Berlin, Germany, 2007.
5. DIN e.V. Besimmung der Tonhaltigkeit von Geräuschen und Ermittlung eines Tonzuschlages für
die Beurteilung von Geräuschimmissionen, DIN45681:2006-08.
6. E. Parizet, A. Bolmont and S. Fingerhuth. Subjective evaluation of tonalness and relation between
tonalness and unpleasantness. In Proceedings of INTER-NOISE 2009, pages 1500-1507, Ottawa,
Canada, August 2009.
7. S. Fingerhuth. Tonalness and Consonance of Technical Sounds. Dissertation, Aachen, Germany,
2009.
8. A. Rieger, A. Oetjen, S. van de Par and H.-P. Rabl. Einfluss der Dissonanz auf die Wahrnehmung
der Geräuschqualität. In Proceedings DAGA 2022, pages 963-966, Stuttgart, Germany, March
2022.
9. DIN e.V. Berechnung des Lautstärkepegels und der Lautheit aus dem Geräuschspektrum,
DIN45631/A1:2010-03.
10. S. Ewert. AFC - A modular framework for running psychoacoustic experiments and
computational perception models. In Proceedings AIA - DAGA 2013, pages 1326-1329, Merano,
Italy, March 2013.
11. DIN e.V. Messtechnische Simulation der Hörempfindung Schärfe, DIN45692:2009-08.
12. ISO 532-1:2017. Acoustics — Methods for calculating loudness — Part 1: Zwicker method.
13. ECMA-418. Psychoacoustic metrics for ITT equipment - Part 2: Models based on human
perception, 2nd Edition, December 2022.
14. T. Lobato and R. Sottek. Modeling the perceived tonal loudness of multiple tonal components. In
Proceedings DAGA 2023, pages 1186-1188, Hamburg, Germany, March 2023.
15. A. Oetjen, P. Volk, S. van de Par. Tonality perception of stationary and transient signals. In
Proceedings of EuroNoise 2015, pages 1913-1918, Maastricht, Netherlands, May-June 2015.
16. A. Oetjen, U. Letens and S. van de Par. Modellierung der Wahrnehmung von nichtstationären
tonalen Komponenten. In Proceedings Motor- und Aggregate-Akustik: 10. Magdeburger
Symposium, pages 62-74, Magdeburg, Germany, July 2018.
17. U. Letens, A. Oetjen, D. Goecke and D. Maiberger. Practical experience with psychoacoustics in
automotive engineering. In Proceedings of the 23rd International Congress on Acoustics, pages
7703-7710, Aachen, Germany, September 2019.
18. P. Volk, A. Oetjen and S. van de Par. Hörmodell zur Detektion von Tonhaltigkeit in transienten
Signalen. In Proceedings DAGA 2015, pages 902-905, Nuremberg, Germany, March 2015.
19. E. Claaßen, A. Oetjen, S. van de Par. Modell zur Detektion und Quantifizierung von zeitlich
variablen tonalen Komponenten. In Proceedings DAGA 2020, pages 792-795, Hannover,
Germany, March 2020.
20. Peter M. C. Harrison and Marcus T. Pearce. Simultaneous Consonance in Music Perception and
Composition. Psychological review, 127(2):216-244, 2020.
