Principal dimensions of perceptual attributes in indoor public spaces Semiha Yilmazer 1 Ray W. Herrick Laboratories Purdue University 177, S. Russell Street, West Lafayette, IN, 47907, USA Bilkent University Department of Interior Architecture and Environmental Design, 06800, Ankara Turkey Volkan Acun 2 Bilkent University Department of Interior Architecture and Environmental Design, 06800, Ankara Turkey Donya Dalirnaghadeh 3 Bilkent University Department of Interior Architecture and Environmental Design, 06800, Ankara Turkey Ela Fasllija 4 Bilkent University Department of Interior Architecture and Environmental Design, 06800, Ankara Turkey Zekiye Şahin 5 Bilkent University Department of Interior Architecture and Environmental Design, 06800, Ankara Turkey Elif Mercan 6 Bilkent University Department of Interior Architecture and Environmental Design, 06800, Ankara Turkey

ABSTRACT This study aims to analyze the principal dimensions of perceptual attributes in indoor public spaces. Healthcare, working, cultural, educational, leisure, worship, and transportation spaces (e.g., bus, train, metro stations) were chosen as public spaces. A listening test was performed with university

1 syilmaze@purdue.edu semiha@bilkent.edu.tr 2 volkan.acun@bilkent.edu.tr 3 ddalirnaghadeh@bilkent.edu.tr 4 e.fasllija@bilkent.edu.tr 5 zekiye.sahin@bilkent.edu.tr 6 elif.mercan@bilkent.edu.tr

tt inter-ncie’” SCOTTISH evENT CAMPUS 3) O , o iLASGOW

students and faculty members (n=32), where they were asked to evaluate the 21 binaural recordings of indoor public soundscapes based on 30 adjective pairs. Principal component analysis shows three prominent perceptual dimensions: Pleasantness, Eventfulness and Clarity, explaining respectively 28.3%, 18.9% and 11.5% of the total variance within the data. The findings were consistent with the literature, suggesting that Pleasantness and Eventfulness can generalize to different soundscapes and be used as attribute scales to measure perception. Moreover, the third principal dimension of Clarity was exclusive to this research and indicated that underlying dimensions of indoor soundscapes could differ based on the function of the space. 1. INTRODUCTION

The soundscape approach is concerned with the perception, interpretation, and meanings associated with sounds. Over the years, research in the soundscape area gained significant momentum. In an attempt for standardization, the working group of ISO/TC 43/SC1 has published a series of ISO standards regarding the definition, data collection, and data analysis methods for soundscape research 1-3 . While measuring the perception was a challenging task on its own, language poses another barrier. People's perception and interpretation of their surroundings highly depend on the language they use, meaning that perception of soundscapes will be influenced by the language they use 4 .

The ISO 12913-2:2018 2 has suggested different means of data collection protocols such as guided interviews, soundwalks, and in situ questionnaire surveys. These protocols include perceptual attributes of soundscape perception such as Pleasantness and Eventfulness and are mostly based on the works of Axelsson et al. 5,6 . However, considering that these data collection protocols will be used by people worldwide, their applicability to different languages must be considered.

The other point we need to consider is the validity of the standards for indoor soundscapes. The standards, and the previous research that paved the way for them, were developed based on the urban soundscapes. As explained by Torresin et al., 7 the acoustical characteristics of indoor environments are different than outdoors, such as people have less control over the acoustic environments, the contexts can significantly differ due to the greater variety of tasks, volumetric variety within the indoor environments, and the presence of both external and internal sound sources within the environment.

With this regard, this paper presents the preliminary results of an indoor soundscape study that focuses on addressing the two issues mentioned above. This study aims to identify the principal dimensions of indoor public spaces that are associated with Turkish attribute scales.

2. METHODS

For this study, we conducted listening tests with participants in a listening room and obtained their subjective evaluation of the recordings through adjective pairs on 5-point scales. A detailed description of our experimental procedure can be found in this section.

2.1. Soundscape Recordings

The acoustic stimuli of the studies consisted of soundscape recordings of seven different types of indoor public spaces. These can be listed as an open-plan office, a café, a hospital, a metro station, ,a mosque, a library and two museums. Since the characteristics of indoor soundscapes are significantly associated with the changes in the user activity, we took recordings at different times of the day and under different crowdedness levels. All environments are recorded in situ and used to create an indoor public soundscapes database.

We selected three thirty-second-long recording samples for each space type, adding up to twenty- one recordings. Each participant listened to seven recordings in random order. We randomized the

listening order to ensure that each sample was listened to an equal number of times at the end of the study.

2.2. Questionnaire

The questionnaire consisted of three parts and was in Turkish. Before starting the survey, we conducted an audiometry test for each participant. The first part of the questionnaire was concerned with demographic information such as Age and Gender. The listening tests follow up this part. Participants listened through 7 indoor public soundscapes. Two question sets followed each recording. The first set of questions was regarding the sound sources categories they heard in the recording . The second question set included 30 adjective pairs (Table 1). These pairs were based on the research conducted by Ozcevik et al. 8 . A five-point Likert scale is used for the adjective pairs.

Table 1: The adjective pairs used in the questionnaire survey 8 .

Adjective Pairs English Turkish Loud-Quite Gürültülü-Sessiz Unpleasant-Pleasant Mem.Ver.Değil-Memnuniyet Verici Disturbing-Comfortable Rahatsız edici-Rahatlatıcı Stressing-Relaxing Stres Yaratıcı-Dinlendirici Artificial-Natural Yapay-Doğal Agitating-Calming Heyecanlandırıcı-Yatıştırıcı Boring-Exciting Sıkıcı-İlgi Çekici Not Preferred-Preferred Tercih Etmem-Tercih Ederim Open-Enveloping Açık-Sarmalayıcı Discordant-Harmonic Ahenksiz-Ahenkli Hard-Soft Sert-Yumuşak Not Sharp-Sharp Keskin Değil-Keskin Crowded-Uncrowded Kalabalık-Tenha Disorganized-Organized Düzensiz-Düzenli Far Away-Nearby Uzak Plan Ses-Yakın Plan Ses Discontinuous-Continuous Devamsız-Devamlı Steady-Unsteady Monoton-Degişken Deserted-Lively Terk Edilmiş-Yaşayan Empty-Joyful Durgun-Nesşeli Gloomy-Exciting İç Karartıcı-Coşturucu Weak-Strong Zayıf-Güçlü Soft-Loud Yavaş-Hızlı Dark-Light Boğucu-Ferah Muffled-Shrill Boğuk-Net Dull-Sharp Donuk-Keskin Heavy-Light Ağır-Hafif Rough-Smooth Pürüzlü-Pürüzsüz Unclear-Distinct Karışık-Ayırtedilebilir Calming-Eventful Sakin-Hareketli Common-Strange Alışılmış-Farklı

2.3. Data Collection and Analysis

A total of 32 individuals, twenty four women and eight men (M age = 24; SD = 3.6 years, age range 19-36), volunteered to participate in the study. All participants had less than 15dB hearing loss for the tested frequency ranges (125 Hz – 8kHz). The majority of the participants were university students and academic faculty of Bilkent University. All participants were native Turkish speakers.

The questionnaire survey was conducted in a listening room located within the Bilkent University campus. All listeners were presented with a random order of soundscape samples. They were told to evaluate the recordings after listening to them. They could listen to a recording as many times as needed but only move to the next recording after evaluating the present one. The experiments lasted 30 minutes on average, including the audiometry test. Each listener evaluated 7 of the 21 recordings. Considering that preliminary results included the results gathered from 32 participants, each recording was evaluated by a minimum number of 10 participants [(32x7)/21=10.6].

The data gathered from the survey was analyzed with R Studio. We conducted Principal Component Analysis (PCA) to analyze the principal dimensions contributing to Turkish speakers' perceived affective response towards indoor public spaces. In order to perform the PCA, the mean values for 30 adjective pairs are calculated for each of the 21 soundscape recordings, resulting in a 21 x 30 data frame. This is followed by correlating all 30 adjective pairs to create a 30 x 30 intercorrelation matrix. The Kaiser-Meyer-Olkin (KMO) Test for Sampling Adequacy and Bartlett's Test of Sphericity are utilized to ensure that the dataset is adequate for a PCA. The KMO score for the data is above the cut off value of 0.70, with a value of 0.89, confirming that the data set is adequate for a PCA model. Bartlett's Test of Sphericity yielded a statistically significant result (χ2 = 3345.412, p <0.000), confirming that the correlation matrix is not an identity matrix and is fit for PCA.

3. RESULTS AND DISCUSSION

After obtaining a satisfactory model adequacy score, we conducted PCA on the data frame. PCA is conducted using the Oblimax rotation. The model yielded a Root Mean Squares of Residuals (RMSR) score of 0.06 (χ2 = 412.2, p <0.000), which is acceptable as it is closer to zero and indicates a good model fit.

Five components have satisfied Kaiser's criterion (eigenvalue > 1.0). However, after going over the scree plot and based on components' total contribution to the explained variance, we eliminated two components with marginal contribution (1-4%). The three components (Table 2) we used for the model explained 23, 19, and 17% of the total variance of the data set, adding up to 59%.

Table 2: The extracted principal components and the weights of the factors that contribute to each of them.

Component 1 (28.3%) Component 2 (18.9%) Component 3 (11.5%)

Hard-Soft 0.79 Empty-Joyful 0.82 Disorganized-Organized 0.79 Stressing-Relaxing 0.78 Gloomy-Exciting 0.80 Muffled-Shrill 0.79 Disturbing-Comfortable 0.75 Boring-Exciting 0.79 Unclear-Distinct 0.75 Loud-Quite 0.71 Open-Enveloping 0.75 Far Away-Nearby 0.66 Unpleasant-Pleasant 0.68 Steady-Unsteady 0.60 Rough-Smooth 0.65 Heavy-Light 0.67 Deserted-Lively 0.57 Discontinuous-Continuous 0.61 Agitating-Calming 0.61 Soft-Loud 0.56 Not Sharp-Sharp 0.39

Calming-Eventful 0.55 Common-Strange 0.32

Five components have satisfied the Kaiser's criterion (eigenvalue > 1.0). However, after going over the scree plot and based on components' total contribution to the explained variance, we eliminated two components with marginal contribution (1-4%). The three components (Figure 1, Figure 2) we used for the model explained 28.3, 18.9, and 11.5 % of the total variance of the data set, adding up to 58.7%.

Figure 1: Loading of the adjective pairs in Component 1 and Component 2. Variable vectors’ distance indicates their loading score. The legend indicates vectors’ contribution to the components.

Figure 2: Loading of the adjective pairs in Component 3 and Component 2. Variable vectors’ distance indicates their loading score. The legend indicates vectors’ contribution to the components.

ma 115%)

According to the loadings (>.299) (Table 2), Component 1 is explained by Hard-Soft, Stressing- Relaxing, Disturbing-Comfortable, Loud-Quiet, Unpleasant-Pleasant, Heavy-Light , and Agitating- Calming , which can be labelled as Pleasantness . Component 2 is explained by Empty-Joyful, Gloomy-Exciting, Boring-Exciting, Open-Enveloping, Steady-Unsteady, Deserted-Lively, Soft-Loud, Common-Strange, and Calming-Eventful . Among these pairs, the contribution of the variable " Common-Strange " is very negligible and can be dropped from the component. The other adjective pairs are mainly associated with the activity taking place in the environment and, therefore, can be labelled as Eventfulness . Component 3 consists of Disorganized-Organized, Muffled-Shrill, Unclear- Distinct, Far away-Nearby, Rough-Smooth, Discontinuous-Continuous , and Not Sharp-Sharp , which are mainly associated with the intelligibility of sounds within the environment and labelled as Clarity . We measured the reliability of these components with Cronbach's Alpha test. All three components yielded scores above the cutoff value of 0.70, which can be listed as 0.89, 0.84, and 0.78.

The factor loadings of Component two indicates that the adjective pairs of Gloomy-Exciting and Boring-Exciting are among the variable with the highest contribution to the component. While the scale of Exciting seems to be repeating, this is caused due to its English translation. The original adjective pairs for Gloomy-Exciting (Turkish: İç karartıcı – Coşturucu) and (Turkish: Sıkıcı-İlgi Çekici) refer to different affective quality of an environment which does not easily translate to English. This is an example of a situation where cultural differences need to be considered to avoid confusions that might occur while translating into different languages.

When compared with the existing literature, the components produced by the analysis are consistent. The previous study conducted by Axelsson et al., 5 labelled the first two components as Pleasantness and Eventfulness, which supports our findings. The majority of the scales contributing to our first two components are similar to those found in their research; therefore, we choose to retain the same names. On the other hand, Component 3 is very different. While their third component is associated with familiarity with the environment, the one we found is related to the clarity of the sound objects within the environment. While it can be argued that we did not have an adjective pair such as Familiar-Unfamiliar , we did include Common-Strange . This pair is listed under Component 2 Eventfulness, with an insignificant component loading of 0.32 and can be dropped from the component altogether. This begs the question of whether this pair would form a fourth component if there were more adjective pairs related to familiarity included in the questionnaire.

Torresin and colleagues 7 develop the other principal component model relevant to our research. Their model is also concerned with indoor soundscapes, but their research was concerned with residential buildings while we focused on public spaces. The scales that contribute to our components 1 and 2 are also consistent with this research. For example, the adjective pairs of Component 1, Unpleasant-Pleasant, Disturbing-Comfortable, and Stressing-Relaxing , represent authors Component 1, Comfort. In addition to this, items of Component 2 are also similar, as they are both related to the activity and content of the space and include identical scales such as, Lively and Empty .

As a further contribution, our Component 3 is entirely unique. The different indoor spaces used in the research could support this finding. Residential buildings are private spaces without much sound interference or the presence of a high amount of ambient sound. Most of the recordings used in our research included crowded spaces with a large amount of ambient sounds, mainly originating from human activities (e.g. intelligible and unintelligible speech, laughing, coughing, footsteps etc.). This could potentially elevate the importance of Clarity and make it a factor influencing the perception of indoor public spaces.

4. CONCLUSIONS

This research investigated the principal dimensions of indoor public spaces that are associated with Turkish attribute scales. The analysis revealed three principal dimensions: Pleasantness, Eventfulness, and Clarity. The results were consistent with the previously conducted research as the scales contributing to the two primary dimensions were similar. The third dimension, on the other hand, has been different. Considering that a greater variety of tasks are performed in indoor spaces, their soundscape characteristics will be different from outdoor spaces and different from the indoor context of other functions, such as private and public.

5. REFERENCES

1. ISO. ISO 12913-1 Acoustics —Soundscape — Part 1: Definition and conceptual framework.

(2014). 2. ISO. ISO/TS 12913-2:2018 Acoustics — Soundscape — Part 2: Data collection and reporting

requirements. (2018). 3. ISO. ISO/TS 12913-3:2019 - Acoustics — Soundscape — Part 3: Data analysis. (2019). 4. Nagahata, K. (2018). Linguistic issues we must resolve before the standardization of soundscape

research. Proceedings of Euronoise 2018 , Crete, Greece, May 2018. 5. Axelsson, Ö., Nilsson, M. E., & Berglund, B. A principal components model of soundscape

perception. The Journal of the Acoustical Society of America , 128 (5) , 2836–2846 (2010). 6. Axelsson, Ö., Nilsson, M. E., & Berglund, B. The Swedish soundscape-quality protocol. The

Journal of the Acoustical Society of America , 131 (4) , (2012). 7. Torresin, S., Albatici, R., Aletta, F., Babich, F., Oberman, T., Siboni, S., & Kang, J. Indoor

soundscape assessment: A principal components model of acoustic perception in residential buildings. Building and Environment , 182 , (2020). 8. Ozcevik Bilen, A., & Yuksel Can, Z. An applied soundscape approach for acoustic evaluation –

compatibility with ISO 12913. Applied Acoustics , 180 , (2021).