Perceptual acoustic space of tyre noise Thibaut Marin-Cudraz 1 Laboratoire Vibrations Acoustique INSA-Lyon, 25 bis avenue Jean Capelle, 69621 Villeurbanne cedex, France Juan Jesus García 2 Applus+ IDIADA PO Box 20 Santa Oliva, 43710 Tarragone, Espagne Etienne Parizet 3 Laboratoire Vibrations Acoustique INSA-Lyon, 25 bis avenue Jean Capelle, 69621 Villeurbanne cedex, France

ABSTRACT

Road traffic noise accounts for the majority of perceived urban environmental noise and has important health consequences. The rolling noise of vehicle tyres is a major contributor to per- ceived road noise. The tread pattern of light vehicle tyres is already designed to minimize the amplitude of the noise emitted, but this is not the case for heavy vehicles. The European LEON- T project aims to minimize the nuisance of heavy vehicle tyres, especially noise. Prior to a study of the effects of tyre noise on sleep, an experiment was conducted to determine the timbre pa- rameters of such noise. The data set used was obtained by a series of recordings on a standard- ized track using tyres of various sizes. These stimuli were presented to headphones in a free sorting task.

The poster will show the results of this experiment, including the correlations between acoustic parameters and perceptual space structure determined from the groups formed by the participants. 1. INTRODUCTION

Road traffic noise has negative impacts on sleep and can generate physiological stress responses: (changes in cardiac function, blood pressure, viscosity, coagulation and lipid and carbohydrate con- tent of blood) [1,2]. In the long term, the risk of cardiovascular diseases such as heart attacks may increase [3].

1 thibaut.marin-cudraz@insa-lyon.fr

2 juanjesus.Garcia@idiada.com

3 etienne.parizet@insa-lyon.fr

worm 2022

The noise of combustion engines has been reduced in the last decades [4] and will even tend to disappear with the progressive electrification of the car fleet. It is therefore necessary to study the second source of noise from vehicles: the rolling noise of tyres.

One of the objectives of the European project LEON-T (Low particle Emissions and lOw Noise Tyres) is to understand and minimize the impact of tyre rolling noise on sleep. The interest will be focused on the effect of the different psychoacoustic parameters of these noises on cardiovascular health during sleep according to different road traffic scenarios.

The tyres available on the market have a great diversity of uses and tread patterns, influencing the generated rolling noise [5]. Therefore, it is necessary to study the acoustic diversity of these noises in order to create representative road traffic scenarios. This poster presents a preliminary free-sorting listening task experiment to identify the psychoacoustic parameters used to distinguish tyre noises. Since the scenario devised for the sleep experiments is indoor (in a bedroom), a comparison of two listening scenarios (exterior and interior) will be performed to study the influence of the scenario on the perception of rolling tyre noise.

2. STIMULI

2.1. Simulation of an exterior listening scenario

33 sound stimuli were taken from a serie of recordings made by Applus+ IDIADA for the LEON-T project on their ISO 10844:2014 compliant track. These rolling sounds were obtained for different types of vehicles (car, van, and heavy truck). For the heavy truck tyres, various uses (summer tyres, alpine winter tyres and Nordic winter tyres) and positions (steering or traction). Since the mi- crophone was located 7.5 meters from the recording track, the resulting sounds were ideal for simu- lating a listening scenario where the listener would be close to the road (level between 67.46 dB(A) and 79.93 dB(A)), i.e., the "exterior" condition.

2.2. Simulation of an exterior listening scenario

The 33 stimuli of the "interior" condition were produced to simulate a scenario in which the listener would be in a bedroom (noise level between 26.69 dB(A) and 38.94 dB(A)). Using special- ized building acoustics software (AcouBAT from Cypres), the sound insulation of a facade was sim- ulated to represent a wall with characteristics that can be found anywhere in Europe (10cm thick concrete wall and a double-glazing window equipped with an air intake and a rolling shutter box) while allowing an overall attenuation of 43 dB, which meets all current minimum regulations for exterior noise in Europe [6]. The frequency profile of the sound attenuation R was transformed into a frequency filter and applied to the 33 stimuli of the outdoor condition. 3. FREE SORTING TASK PROTOCOL

53 students from INSA-Lyon volunteered to participate in the experiment. The free sorting task was done with headphones (Sennheiser HD650) and used a graphical interface developed in- house (using Python [7] and tKinter [8]) where participants could freely move and listen to the tyre noises (converted to 16-bit WAV with a sampling rate of 44100 Hz), represented by numbers (Figure 1). The experimental device was previously calibrated with a binaural recording head associated with measurement microphones (Neutrik-Cortex Instruments 'MANIKIN MK1') connected to an OROS OR38 system, so that the levels heard by the participants were identical to reality.

worm 2022

Figure 1: A finished free sorting task. The different noises groups are identified according to the black rectangles. To validate the task, each sound had to be moved and listened to at least once.

worm 2022

For each listening scenario (outdoor and indoor), the participant had to sort the sounds ac- cording to their similarity: each group represented sounds that seemed similar, homogeneous accord- ing to his/her perception and sufficiently different from the rest (see Figure 1 for the state of the interface when the task is completed). When the first sorting task was validated, the participant had to repeat the task for the other listening scenario. The order of the two listening scenarios as well as the assignment of sounds to each number in each particular scenario was randomized. At the end of the experiment, the participant was briefly interviewed to explain in his/her own words what criteria he/she had used to group the sounds. 4. RESULTS

At the end of the experiments, individual accordance matrices were built from the participants’ results of the free sorting tasks for each condition. The matrices were calculated using the following rule: if the participant grouped two noises together, then the value in the matrix for this pair is 0, if the noises are in different groups, the value is 1. We then obtained 106 triangular matrices. For each condition, we averaged the corresponding 53 matrices, resulting in two mean accordance matrices, with values inside representing the number of times two sounds were grouped together: the closer to 0, the more often they were together.

The two mean accordance matrices were used as distance matrices to perform hierarchical agglomerative clustering [9], an algorithm that organizes data into groups (clusters) by merging them according to their similarity using the mean distance aggregation criterion (the mean distance of the different cluster is compared). This resulted into classification trees (Figure 2 and Figure 3). In our case, the more the noises were grouped together by the participants, the more likely they belong to the same cluster. The number of clusters, i.e. the place to where the trees could be cut, was found

Pour valider Fexpérience vous devez avoir écouté et bougé au moins une fois les sons. VALIDER ers v 1810 m 16 2.20 er)

using the maximum mean silhouette criterion [10] for each possible cut of the trees: it ranges from -1 (all clusters are poorly separated) to 1 (all clusters are well defined and homogenous). We found that the maximum silhouette criterion was obtained when cutting the tree in 5 clusters for the exterior condition (mean silhouette: 0.27, Figure 2) and 2 clusters for the interior condition (mean silhouette: 0.32, Figure 3).

worm 2022

Figure 2: Classification tree of the tyre noises for the exterior condition. The identification number of the noise as well as the engine status (Engine On or Off (Coastdown)) is indicated at the end of each branch (the leaves). The lower the height between two noises, the more they were grouped to- gether during the experiment. The dashed line shows the cut of the tree selected using the silhouette criterion, resulting in 5 clusters, whose names are indicated on the tree and are also designated by different colors.

Exterior condition

worm 2022

Figure 3: Classification tree of the tyre noises for the interior condition. The identification number of the noise as well as the engine status (Engine On or Off (Coastdown)) are indicated at the end of each branch (the leaves). The lower the height between two noises, the more they were grouped together during the experiment. The dashed line shows the cut of the tree selected using the silhouette criterion, resulting in 2 clusters, whose names are indicated on the tree and they also are designated by different colors. 5. PSYCHOACOUSTIC PROFILES OF THE TYRE NOISE GROUPS

Different psychoacoustic parameters were measured with Artemis 13.2 (Head acoustics) to study the acoustic space of tyre noises, i.e. the differences in each psychoacoustic parameter between the different clusters identified in the experiment. The roughness, sharpness, loudness, pitch, center frequency and lowest pitch were measured. These parameters were chosen in regards to the answers obtained in the post-experiment interview: the participants used criteria closely related or similar to those selected.

The spectral centroid frequency is the center of mass, i.e., the weighted average, of a frequency spectrum. The lowest tonal frequency (tonal frequency in Figure 4 and Figure 5) was defined as follow: if the noise is atonal (tonality < 0.1 tuHMS), it is set to 0 Hz, if the sound is tonal (tonality ≥ 0.1 tuHMS), then the specific tonality - the tonality as a function of frequency - is calculated and the tone with the lowest frequency is taken. This parameter allows us to distinguish between tonal and atonal noises, but also to identify an effect of the motor (usually present in the very low frequencies) on the perceived tone.

The different clusters identified for each scenario have different psychoacoustic profiles (Fig- ure 4 and Figure 5). It is interesting to note that cluster 4, the least tonal, contains all light vehicles, and is found in cluster 2 of the indoor condition.

Interior condition DQOONRBOKQOOQOGTNOONK OPN MK DK DDON DROW AANAANNAA NAA NAAINAAIANAINAAAAIAAAAINNO oon eecooenooeooeohenooonenneenon Engine On Coastdown & 0.0 0.2 0.4 0.6 0.8 Height

worm 2022

Figure 4: Repartition of the psychoacoustics parameters in each cluster for the exterior listening con- dition. The p-values on top of each boxplot are the result of a non-parametric ANOVA (Kruskal- Wallis test) to test the main effect of each parameter, i.e. it indicates that each parameter is affected by the cluster.

pm fe gt fe pels, eal, i 2 er ee ae ae ee i ~a} i a |- tO Ge Ge Gel oe Oe Oe (euog) sseupno7 (2H) Aouanbes, jeuo] ysemo7, tt: —7- | | a 1 a (wnoy) sseudieys (2H) Aouenbasy je9UeQ. , i = i = 020 si0 O10 S00 (uedsy) sseuyBnoy

Figure 5: Repartition of the psychoacoustics parameters in each cluster for the interior listening con- dition. The values on top of each boxplot are the result of a non-parametric comparison test (Wicoxon test).

oe oo } » [fs i ay ‘owes ssourrey (up fveroruelisanoy |e |, | 4 |- Se Bi 5 Em} | ee! pL S 5 Ell OF 60 80 008 004 009 00S ™ — (wnoy) sseudieys: (2H) Aouenbes4 enue a ae of. zo 800 +00 vo €0 zo 10 jedsy) ssouy6noy (SWHM) Ayjeuo, Cluster ‘Cluster

6. CONCLUSIONS

This study showed that tyre rolling noises are diverse and that this diversity is perceived by human listeners. The listeners used different psychoacoustic parameters to group the noises according to perceived similarity.

The listening scenario also influenced the perception of tyre noises since the number of iden- tified groups is different. However, the tonality and loudness played a major role in the definition of noise groups by participants in both scenarios. Thus, different combinations of these two parameters will be used to synthesis artificial stimuli to create realistic and diverse road traffic scenarios, simu- lating different vehicles passing by. The influence of these scenarios on sleep will be studied in the following sleep experiment of LEON-T project.

7. ACKNOWLEDGEMENTS

Thanks to all 53 participants in the experiment. This article is part of a research project that received funding from the European Union's Horizon 2020 research and innovation program under grant agreement No. 955387. 8. REFERENCES

1. T. Münzel, T. Gori, W. Babisch et M. Basner. Cardiovascular effects of environmental noise

exposure. European heart journal, 35.13 , 829-836 (2014). 2. W. Babisch. The noise/stress concept, risk assessment and research needs. Noise and health , 4.16 ,

1 (2002). 3. W. Babisch. Updated exposure-response relationship between road traffic noise and coronary

heart diseases: a meta-analysis. Noise and Health , 16.68 , 1 (2014). 4. B. Favre, & E. Parizet. L’Acoustique des Véhicules Routiers. Acoustique et Techniques , 40

(2005). 5. J. A. Ejsmont, U. Sandberg & S. Taryma. Influence of tread pattern on tire/road noise. SAE Trans-

actions , 632-640 (1984). 6. A. Alonso, R. Suárez, J. Patricio, R. Escandón & J. J. Sendra. Acoustic retrofit strategies of win-

dows in facades of residential buildings: Requirements and recommendations to reduce exposure to environmental noise. Journal of Building Engineering, 41 , 102773 (2021). 7. G. Van Rossum, & F. L. Drake Jr. Python reference manual, Centrum voor Wiskunde en Infor-

matica Amsterdam (1995). www.python.org 8. F. Lundh. An introduction to tkinter (1999). www.pythonware.com/library/tkinter/introduc-

tion/Index.htm 9. J.A. Hartigan. Clustering Algorithms, Wiley, New York (1975). 10. P.J. Rousseeuw. Silhouettes: A graphical aid to the interpretation and validation of cluster analy-

sis. Journal of computational and applied mathematics, 20 , 53–65, (1987).

worm 2022