University Students’ Self-Rated Health in Relation to Perceived Acoustic Environment during the COVID-19 Home Quarantine

Background: Online education became mandatory for many students during the Coronavirus disease 2019 (COVID-19) pandemic and blurred the distinction between settings where processes of stress and restoration used to take place. The lockdown also likely changed perceptions of the indoor acoustic environment (i.e., soundscape) and raised its importance. In the present study, we seek to understand how indoor soundscape related to university students’ self-rated health in Bulgaria around the time that the country was under a state of emergency declaration caused by the COVID-19 pandemic. Methods: Between 17 May and 10 June 2020, we conducted a cross-sectional online survey among 323 students (median age 21 years; 31% male) from two universities in the city of Plovdiv, Bulgaria. Self-rated health (SRH) was measured with a single-item. Participants were asked how frequently they heard different types of sounds while at home and how pleasant they considered each of those sounds to be. Restorative quality of the home (the “being away” dimension of the Perceived Restorativeness Scale) was measured with a single-item. A priori confounders and effect modifiers included sociodemographics, house-related characteristics, general sensitivity to environmental influences, and mental health. Our analysis strategy involved sequential exploratory factor analysis (EFA), multivariate linear and ordinal regressions, effect modification tests, and structural equation modeling (SEM). Results: EFA supported grouping perceived sounds into three distinct factors—mechanical, human, and nature sounds. Regression analyses revealed that greater exposure to mechanical sounds was consistently associated with worse SRH, whereas no significant associations were found for human and nature sounds. In SEM, exposure to mechanical sounds related to lower restorative quality of the home, and then to poorer SRH, whereas nature sounds correlated with higher restorative quality, and in turn with better SRH. Conclusions: These findings suggest a role of positive indoor soundscape and restorative quality for promoting self-rated health in times of social distancing.


Introduction
In 2020, the world faced an unprecedented pandemic caused by the severe acute respiratory syndrome coronavirus 2. As a reaction, governments enacted anti-COVID measures to control the pandemic [1,2]. While effective in reducing Coronavirus disease 2019 note, individual differences, such as noise sensitivity, can modify sound perception and aggravate subjectively recognizable negative reactions (e.g., dissatisfaction, annoyance, frustration, and anger) [48].
Understanding health effects of these various forms of sound in an unparalleled situation as the COVID-19 lockdown is not straightforward. Insight into how positive soundscape components (e.g., nature sounds) can offset detrimental effects of unwanted sounds (e.g., mechanical sounds) and how ambivalent human sounds fit in can inform acoustic design and enhance human experience when the capacity to avoid or control the acoustic environment is limited. However, we are not aware of earlier epidemiological research that has specifically explored the joined health impact of various forms of sound during the COVID-19 pandemic in residential homes of participants.
In the present study, we seek to understand how the indoor soundscape related to university students' self-rated health in Bulgaria around the time that the country was under a state of emergency declaration caused by the COVID-19 pandemic. For this purpose, we use a perceived soundscape approach [49,50] taking up on recent developments in indoor soundscape research [51,52]. First, we explore whether preferences for various forms of sound experienced indoors can be grouped into conceptual categories. Second, we investigate whether exposure to unwanted sounds is associated with worse self-rated health, and exposure to wanted sounds, with better self-rated health. Third, we hypothesize that these effects are carried out by constraining or promoting the feeling of "being away" in the home. This conceptual framework is depicted in Figure 1. being [47]. Additionally, of note, individual differences, such as noise sensitivity, can modify sound perception and aggravate subjectively recognizable negative reactions (e.g., dissatisfaction, annoyance, frustration, and anger) [48]. Understanding health effects of these various forms of sound in an unparalleled situation as the COVID-19 lockdown is not straightforward. Insight into how positive soundscape components (e.g., nature sounds) can offset detrimental effects of unwanted sounds (e.g., mechanical sounds) and how ambivalent human sounds fit in can inform acoustic design and enhance human experience when the capacity to avoid or control the acoustic environment is limited. However, we are not aware of earlier epidemiological research that has specifically explored the joined health impact of various forms of sound during the COVID-19 pandemic in residential homes of participants.
In the present study, we seek to understand how the indoor soundscape related to university students' self-rated health in Bulgaria around the time that the country was under a state of emergency declaration caused by the COVID-19 pandemic. For this purpose, we use a perceived soundscape approach [49,50] taking up on recent developments in indoor soundscape research [51,52]. First, we explore whether preferences for various forms of sound experienced indoors can be grouped into conceptual categories. Second, we investigate whether exposure to unwanted sounds is associated with worse self-rated health, and exposure to wanted sounds, with better self-rated health. Third, we hypothesize that these effects are carried out by constraining or promoting the feeling of "being away" in the home. This conceptual framework is depicted in Figure 1.

Study Design and Sampling
Between 17 May and 10 June 2020, we conducted a cross-sectional online survey among students from two universities in the city of Plovdiv, Bulgaria. The weeks preceding data collection overlapped both the period of more restrictive measures in Bulgaria, when recreational establishments were closed and access to public outdoor spaces such as parks was banned, and the weeks that followed immediately after, when those measures were gradually relaxed. Students in medicine, dentistry, and biology were approached by their lecturers with an invitation to participate in a survey on living conditions and mental health. Students could also forward the link to their peers. We targeted

Study Design and Sampling
Between 17 May and 10 June 2020, we conducted a cross-sectional online survey among students from two universities in the city of Plovdiv, Bulgaria. The weeks preceding data collection overlapped both the period of more restrictive measures in Bulgaria, when recreational establishments were closed and access to public outdoor spaces such as parks was banned, and the weeks that followed immediately after, when those measures were gradually relaxed. Students in medicine, dentistry, and biology were approached by their lecturers with an invitation to participate in a survey on living conditions and mental health. Students could also forward the link to their peers. We targeted students in health-related programs because the psychological wear and tear of the student occupation more than likely took a heavy toll on their mental health [53]. To be included, students had to be aged from 18 to 35 years and, to ensure that they were familiar with their neighborhood environment, had to have lived in their current home for at least six months. Further details have been reported elsewhere [10].
The design and conduct of the study followed the general principles outlined in the Declaration of Helsinki. After reading the information about the objectives of the study and instructions on filling-in the questionnaire, all respondents confirmed that they were at least 18 years old and provided informed consent in the survey form, thereby agreeing that their personal information would be processed and stored according to the General Data Protection Regulation in the European Union. The generic design of earlier studies in this series has been approved by the Ethics Committee at the institution of the principal investigator [54].

Sefl-Rated Health
Self-rated health (SRH) was measured with the question "In general, what would you say your health was in the past two weeks?". Response options included "very poor", "poor", "fair", "good", and "very good" [55]. Single-item SRH captures various healthrelevant factors and states and is considered a good proxy for general health across diverse populations [56][57][58][59].

Perceived Acoustic Environment
Participants were asked how frequently they heard the following types of sounds while at home in the past two weeks: traffic noise (e.g., cars, buses, airplanes, and trains); other mechanical sounds coming from outside (e.g., sirens, construction, and machines); nature sounds coming from outside (e.g., singing birds, flowing water, wind blowing, and rustling leaves); mechanical sounds from within the dwelling (e.g., appliances, pluming, and elevators); music from within the dwelling; and sounds from human beings within the dwelling (e.g., conversation, laughter, children at play, and footsteps). Response options were "0 = never", "1 = sometimes", "2 = half of the time", "3 = most of the time", and "4 = all the time".
In addition, we computed pleasantness-weighted exposure scores as the product of exposure frequency and corresponding pleasantness. To that end, pleasantness ratings were recoded, so that both higher and lower scores indicated greater perceived intensity (positive or negative) of the sound (i.e., "2 = very unpleasant", "1 = unpleasant", "0 = neither pleasant nor unpleasant", "1 = pleasant", and "2 = very pleasant"). Thus, a higher score reflects a more frequent exposure to a sound that holds greater emotional valence for the individual and vice versa.

Mediator
Restorative quality of the home, and specifically the feeling of "being away" (i.e., opportunity to get distance from daily stressors and routines), was hypothesized to mediate the effect of perceived sounds on SRH. To reduce questionnaire length and response burden, we used a single item [60,61] adapted from the Perceived Restorativeness Scale [62,63]. The questionnaire instructions were adapted to refer to the home environment (i.e., "At home, the time spent gives me a break from my day-to-day routine and I can get away from the things that usually demand my attention."). Responses were again given on an 11-point scale with two anchors ("0 = not at all" to "10 = completely").

Confounders
We selected a parsimonious set of variables that could confound or modify the associations between perceived sounds and health based on previous research findings.
Sociodemographic characteristics included age, gender, ethnicity (Bulgarian or not), and income. Income adequacy was measured with a single item: "Having in mind the total monthly income you can make use of, how easy is it for you to meet your expenses without depriving yourself?" Response options ranged from "0 = very difficult" to "5 = very easy".
House-related characteristics included dwelling type (i.e., apartment, house, or hostel), presence of a quiet room (far from noise sources) and soundproof windows, and crowding in the household (people-to-rooms ratio). We also asked about duration of residence and time spent at home per day.
Noise sensitivity, and by extension general sensitivity to environmental influences [48], may be independently associated with health or modify the effect of noise on health [64][65][66]. Moreover, it may influence perceptions of the acoustic environment, with more sensitive individuals being more likely to be hypervigilant [67,68]. Therefore, we collected data on sensitivity to noise, light, odors, weather conditions, and dust. Participants were asked "Compared to most people, to what extent are you sensitive to each of the following?". Response options were "0 = not at all sensitive", "1 = slightly sensitive", "2 = moderately sensitive", "3 = very sensitive", and "4 = extremely sensitive". We computed the sum of item responses, which could range from 0 to 20, with a higher summary score indicating greater environmental sensitivity. The scale's internal consistency was not outstanding (McDonald's ω = 0.65).
Since mental health may modify noise perception and its relationship with health [69], we also measured severities of depressive and anxiety symptoms over the past two weeks. The Patient Health Questionnaire 9-item (PHQ-9) measures the frequency of symptoms of depression [70]. Response options included "0 = not at all", "1 = several days", "2 = more than half of the days", or "3 = nearly every day". Scores (sum of the item responses) could range from 0 to 27 (McDonald's ω = 0.68). The Generalized Anxiety Disorder 7-item (GAD-7) scale was used to assess common symptoms of anxiety [71]. Response options were provided on the same scale. Scores (sum of the item responses) could range from 0 to 21. The seven items loaded onto one latent factor and the internal consistency in our sample was high (McDonald's ω = 0.91). The PHQ-9 and GAD-7 scales were dichotomized, where scores of 10 or above were consistent with moderate depression [72] and generalized anxiety disorder [73], respectively.
We did not have information on the exact home address but did collect the settlement where the respondent currently lived. We classified these settlements as cities (>100,000 residents), towns (10,000 to 100,000), or villages (<10,000). Finally, we retrieved data on which university the respondent studied at.

Analysis Strategy
Missing values (<10% on any given variable) were missing completely at random and imputed using the expectation-maximization algorithm [74]. All variables in multivariate analysis models were included in the imputation. The sound pleasantness ratings were not multivariate normal according to Small's omnibus test of multivariate normality (VQ3 (12) = 185.90, p < 0.001), "being away" had excessive kurtosis, and SRH was left-skewed and treated as ordered categorical. We examined general patterns of associations within the data, using Spearman correlation and Mood's median test.
As a next step, we conducted an exploratory factor analysis (EFA) to evaluate the dimensionality of the set of sound pleasantness ratings. Procedure to determine the appropriate number of factors to be extracted was parallel analysis [75,76]. The item scores were treated as ordered-categorical variables, so we chose to compute the factor analysis on the basis of polychoric interitem correlations. A factor solution was fitted using robust diagonally weighted least squares (DWLSs) extraction technique. To achieve factor simplicity we used robust promin rotation [77]. Factor loadings ≥ 0.40 were interpreted as salient [78]. Analyses were carried out in FACTOR v. 10.10.02 [79,80].
Then, we tested multivariate associations between mechanical sounds (traffic noise, outdoor, and indoor mechanical sounds), human sounds (human sounds and music), and nature sounds, on the one hand, and "being away" and SRH, on the other. We fitted a robust linear regression model for "being away" and an ordered logistic regression for SRH. We report two types of adjusted models: a main model adjusted for core confounders (age, gender, ethnicity, income, and environmental sensitivity) and a full model additionally adjusted for dwelling type, settlement type, university, presence of a quiet room, soundproof windows, crowding, duration of residence, and time spent at home. As a sensitivity analysis, these models were also fitted by regressing the outcomes on the pleasantness-weighted sound exposure scores instead of the original exposure frequency scores. Tolerance values > 0.2 [81] and variance inflation factor values < 5.0 [82] indicated no multicollinearity.
Stratified analysis and multiplicative interaction terms were computed to investigate possible modification of the associations of mechanical, human, and nature sounds with SRH. Based on earlier related work [83][84][85][86] and literature reviews [87,88], we tested the following a priori modifiers: gender, income, presence of a quiet room, soundproof windows, garden, presence of a terrace/balcony, crowding, environmental sensitivity, depression and anxiety, and time spent at home per day. Criterion for statistical consideration was relaxed to p < 0.1 to report relevant effect modification that might otherwise remain undetected [89][90][91].
This was followed up with a structural equation modeling (SEM) to test the theoretically indicated interplay between the variables ( Figure 1). We treated SRH as an ordered categorical variable and employed a DWLS estimator [92] with bootstrap-generated (5000 draws) standard errors and confidence intervals for all paths [78,93]. Model specification was guided by theory and bivariate correlations in the dataset. Covariances between the three sound types were specified a priori. Goodness-of-fit was evaluated using indices of acceptable model fit provided in Hu and Bentler [94]: a non-significant χ 2 (p > 0.05); a comparative fit index (CFI) ≥ 0.95; a root mean square error of approximation (RMSEA) ≤ 0.06 with a 90% CI ≤ 0.06; and a standardized root mean square residual (SRMSR) ≤ 0.08. Modification indices were used to identify sources of misfit and, if the suggested modification was supported by theory, the initial model was revised accordingly. Confounding paths with at least marginal statistical significance (p < 0.1) were retained in the final solution. An indirect effect (i.e., a product of coefficients for the constituent paths) that significantly exceeded zero was taken as evidence of mediation [95,96]. While we use terminology accepted in mediation modeling to denote the overall (total effect), direct (direct effect), and indirect (indirect effect) relationships in the SEM, the word "effect" should not be taken to indicate claims of causality.
Data were processed Stata v. 14 (StataCorp. 2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP.) and the lavaan v. 0.6-7 package [97] for R v. 4.0.3. (R Core Team (2020). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.). A p-value of < 0.05 was considered statistically significant except as noted above.

Participant Characteristics
The majority of participants was in their late teens/early twenties, more often female, and Bulgarian. Most participants reported good or very good SRH. Detailed participant characteristics are shown in Table 1.

Exploratory Factor Analysis of Sounds Pleasantness
Several well-recognized criteria for adequacy of the polychoric correlation matrix were used. Bartlett's test of sphericity was significant (χ 2 (15) = 628.10, p < 0.001). The interitem correlation matrix had "fair" sample adequacy (Kaiser-Meyer-Olkin = 0.70; 95% CI: 0.66, 0.76). Parallel analysis indicated that two factors could be extracted to explain the correlations among the six items, and they were labeled according to their content. Pleasantness of traffic noise, outdoor, and indoor mechanical sounds loaded highly on one latent factor, which we labeled "Mechanical sounds". Pleasantness of human sounds, music, and nature sounds loaded on a second factor, which we labeled "Non-mechanical sounds". These two factors were moderately correlated (r = 0.37). The initial eigenvalues showed that the first factor explained 43% and the second factor 25% of the variance in the data. Interitem correlations, factor loading matrix, and extraction communalities for the final two-factor solution are presented in Table 2. Most items had high primary loadings and trivial cross-loadings. However, compared with pleasantness of human sounds and music, pleasantness of nature sounds was weakly associated with the factor "Non-mechanical sounds" (<0.40) and had low communality (<0.30). This supported the treatment of nature sounds as semantically different from anthropogenic sounds, as in earlier soundscape research [98][99][100]. Based on these results, we constructed three exposure variables for subsequent analyses: mechanical sounds (traffic noise, outdoor, and indoor mechanical sounds), human sounds (human sounds and music), and nature sounds.

Bivariate Associations
Correlations between the variables are shown in Table 3. SRH was associated with less frequent exposure to mechanical sounds but not with other sounds. Male participants, those with low environmental sensitivity, those who experienced stronger feelings of "being away", who had a garden, and a quiet room or soundproof windows reported better SRH. Depression and anxiety symptoms were consistent with worse SRH.
"Being away" was higher in older age, in participants having a garden, access to a quiet room, and soundproof windows. "Being away" was inversely related to depression and anxiety symptoms.
Higher environmental sensitivity was related to more frequent perceived exposure to all sound types. Mechanical sounds were negatively associated with "being away", while nature sounds contributed to "being away". Participants having a domestic garden reported lower exposure to mechanical and more frequent exposure to nature sounds. On the other hand, crowding related to mechanical and human sounds. Exposure to mechanical sounds was lower in those having a quiet room and those who did not report pronounced depressive and anxiety symptoms. From Table 4, participants living in a village were less frequently exposed to mechanical and human sounds. Apartment dwellers reported lower exposure to nature sounds, and house dwellers, to mechanical sounds.

Multivariate Associations between Perceived Sounds Exposure and Self-Rated Health
In the main model, nature sounds were positively associated with "being away" (Table 5). "Being away" was also associated with older age, higher income adequacy, and lower environmental sensitivity. These findings persisted in the full model.
Only mechanical sounds were associated with worse SRH. The estimates associated with human and nature sounds were non-significant (Table 5). In addition, lower environmental sensitivity and higher income related to better SRH. In the full model, controlling for additional covariates did not materially change the effect sizes for perceived sounds sound, but having soundproof windows was independently associated with better SRH.
These patterns of association were also observed in the sensitivity analysis, in which pleasantness-weighted sound exposure scores were used. Respective effect estimates in the main model were OR = 0.91 (95% CI: 0.88, 0.95) for mechanical sounds, OR = 1.01 (95% CI: 0.95, 1.08) for human sounds, and OR = 1.03 (95% CI: 0.96, 1.12) for nature sounds.   Table 6 shows the results of stratified analysis. Few of the tested interactions were statistically significant at the p < 0.10 level. Mechanical sounds were more strongly associated with worse SRH in less crowded household and when participants experienced depression. Human sounds were associated with better SRH in participants without soundproof windows installed, while the effect size itself was statistically significant only in men. The effect of nature sounds went in a positive direction when participants did not have a domestic garden and in crowded households. Abbreviations: MUP-Medical University of Plovdiv, PU-Plovdiv University. Coefficients are unstandardized linear regression coefficients (β) and ordered odds ratios (ORs) with their confidence interval (95% CI). 1 Point estimate is statistically significant at the p < 0.05 level (two-tailed). Abbreviations: p-int.-significance of interaction term (perceived sounds × effect modifier). Coefficients are ordered odds ratios with their confidence interval (95% CI). All models are adjusted for gender, age, ethnicity, income adequacy, and environmental sensitivity (when the respective covariate is not tested as a modifier). 1 Interaction term is statistically significant at p < 0.1 level (two-tailed).

Structural Equation Model of the Effect of Perceived Sounds Exposure, as Mediated by "Being Away", on Self-Rated Health
Goodness-of-fit levels for the initial model were not outstanding: χ 2 (22) = 40.92, p = 0.008; CFI = 0.90; RMSEA = 0.05 (90% CI: 0.03, 0.08); and SRMR = 0.05. Therefore, the model was modified by removing paths associated with a p-value larger than 0.10. Ethnicity was dropped from the model.
Goodness-of-fit statistics associated with the modified model indicated good consistency with the data: χ 2 (19) = 18.86, p = 0.466; CFI = 1.00; RMSEA = 0.00 (90% CI: 0.00, 0.05); and SRMR = 0.09. The model explained 22% of the variance in SRH and 14% in "being away". As can be seen in Table 7 and Figure 2, only mechanical sounds were associated with a significant total effect on SRH. On the other hand, both mechanical and nature sounds had indirect effects on SRH through "being away". These indirect pathways operated in opposite directions-mechanical sounds related to weaker feelings of "being away", and then to worse SRH; conversely, nature sounds were associated with stronger feelings of "being away", and in turn with better SRH. ated with a significant total effect on SRH. On the other hand, both mechanical and nature sounds had indirect effects on SRH through "being away". These indirect pathways operated in opposite directions-mechanical sounds related to weaker feelings of "being away", and then to worse SRH; conversely, nature sounds were associated with stronger feelings of "being away", and in turn with better SRH. Table 7. Associations between perceived sounds, "being away" and self-rated health in the structural equation model.   paths linking perceived sounds to feelings of "being away" and self-rated health. Note: Unstandardized regression weights with their significance level are given for each path. R 2 shows proportion of variance explained in endogenous variables. Coefficients marked with an asterisk (*) are statistically significant at p < 0.05. Control variables are shown in grey. Covariances, associations between control variables, and errors terms are not displayed to enhance readability.

General Findings
This study explored associations between different forms of sound experienced indoors and self-rated health during the COVID-19 quarantine in Bulgaria when students spent home almost all of their time. In the multivariate regression analyses, greater exposure to mechanical sounds was consistently associated with worse SRH, whereas no significant associations were found for human and nature sounds. However, SEM revealed interesting mechanistic insights. While a total effect was only observed for mechanical sounds, supporting regression-based findings, both mechanical and nature sounds were found to indirectly relate to SRH through "being away", that is, the perceived restorative capacity of the home environment. Exposure to mechanical sounds related to lower restorative quality of the home, and then to poorer SRH, whereas nature sounds correlated with higher restorative quality, and in turn with better SRH.
Exploration of the dimensionality of perceived pleasantness via EFA supported grouping individual sound exposures into three distinct factors-mechanical, human, and nature sounds. This factorial structure is consistent with categorization of sounds in soundscape studies [98][99][100] and also reflects differential affective response to sounds and underlying neural networks [101,102]. Earlier research has identified distinct neural activation patterns corresponding to perception of these conceptual categories of sounds, with human sounds yielding activation preferentially in the bilateral posterior superior temporal sulci, animal sounds-in the bilateral posterior insulae, and mechanical sounds-in the anterior superior temporal gyri and parahippocampal cortices [101]. Albeit distinct in terms of perceived pleasantness, human and nature sounds stood closer together relative to mechanical sounds, in line with the broader cognitive differentiation between "living" and "non-living" sounds [101,103].
Our findings about mechanical sounds align with a large body of evidence linking traffic noise to a variety of ill-health outcomes [104]. Fewer studies used health related quality of life [105][106][107] or poor self-rated health, as operationalized here [108,109]. Even less is known about how perceived soundscape affects self-rated health in residential homes [87,110], as most studies of this type were conducted in public places like parks or shopping areas, where people spend only a small amount of time compared with the time spent home. While there is no doubt that unwanted sounds, such as traffic noise, can directly affect health and quality of life, it often escapes the spotlight that noise can in addition constrain psychophysiological stress recovery in a setting [111]. This indirect effect can lead to worse adaptation to other concurrent environmental stressors [6], which in turn may be further compounded with the already constrained restoration of student participants attending online lessons at home (cf. [7][8][9][10]).
Nature sounds did not directly result in better SRH, but the presence of a total effect is not a prerequisite for a meaningful indirect effect to exist [96]. Nature sounds seemed to work through restorative quality, which is consistent with a previously reported contribution of bird songs [37] and other nature sounds [112,113] to stress recovery and perceived attention. Stress reduction theory holds that multimodal sensory input from nature can evoke positive emotions and block psychophysiological stress as a result of evolutionary embedded preference for natural attributes, such as sounds and visual configurations [114,115]. In addition, according to Attention restoration theory, natural environments readily enable recovery of fatigued attentional resources needed to cope with environmental stressors [116,117]. Auditory natural stimuli are a key component of nature experience and may even drive stress recovery beyond visual exposure [112]. Hence, nature sounds penetrating the dwelling may provide rich information content and act as sensory cues retrieving relaxation conditioned on the positive affect and tranquility experienced in natural settings (cf. [118,119]). An alternative explanation of our findings could be wanted sounds enhancing indoor soundscape pleasantness and eventfulness through informational masking of unwanted sounds [120][121][122]. However, the masking effect is greater when the competing stimuli are the same frequency, and most nature sounds tend to be too high-pitched compared with low frequency components of mechanical sounds [123,124].
In the stratified analysis, there was some indication of potentially beneficial associations, but no clear pattern was observed for human sounds. We are mindful that a lack of specificity in our items could have introduced bias, especially with regard to ambivalent human sounds, which may activate processes of either stress or restoration, depending on their affective content [102]. For example, a baby's cry will have lower valence and instigate greater arousal than a baby's laugh [102]. It could be that, since we did not differentiate between sound sources, positive (e.g., voices of loved ones) and negative (e.g., neighbor sounds) semantic components were both captured by the items and cancelled each other out. On the one hand, noises coming from neighbors can be especially annoying while students are trapped in their dwelling and need peace and quiet to focus on their studies. Conversely, presence of human sounds can reinforce feelings of social embeddedness, which is an essential resource in times of isolation from one's usual social network (cf. [125]).
Overall, our findings support the importance of indoor soundscape for shaping perceived health through the extent to which the home environment is conducive to psychological restoration. A restorative environment, such as one supplying nature sounds and relatively free of mechanical sounds, could contribute to personal coping resources needed to offset stress [6,126].

Secondary Findings
Among the other factors investigated here, environmental sensitivity emerged as highly predictive of worse SRH.
Although not yet fully understood, noise sensitivity is a psychometric construct with underlying neurobiological and genetic components [127], which represents individual's heightened reactivity to auditory stimuli in general [128]. It partly overlaps with other types of environmental sensitivity [63,128] and has been found to independently contribute to psychological and physical ill-health [129,130], and modify the effects of noise on health [131][132][133]. However, overall results are inconsistent [134], which may in part explain why we observed no effect modification by environmental sensitivity. Income adequacy was also predictive of "being away" and SRH, which is not surprising since perceived income affects the probability of reporting a better health more than objective income [135].
By and large, evidence of effect modification in our study was sparse. The effect of mechanical sounds was robust across various individual and contextual factors, with statistical evidence of effect modification by crowding and depression. That is, low crowding and presence of depressive symptoms seemed to compound the negative effect. While crowding is usually considered detrimental to health and well-being [136], having more people in the household in a time of loss of other immediate social interaction may be protective or alter processing of mechanical sounds by providing distraction. As for depressed individuals, they may have a more pronounced annoyance reaction to noise [69]. Concerning human sounds, we only saw effect modification by having soundproof windows, but we can offer no satisfactory explanation. This finding could be the result of unaccounted for confounding or merely a statistical artifact. We suspect that a lack of specificity in the soundscape items and referring to sounds inside the dwelling in general could have obscured effect modification by access to a quiet room. In participants having a domestic garden, nature sounds did not appear protective, arguably because they may not have felt mobility restrictions as much as those without gardens and could spend unrestricted amounts of time outdoors, which would itself drive the positive health effect (as suggested by bivariate correlations). Conversely, for those who lacked access to an actual natural setting, nature sounds probably served as a surrogate. The finding that in crowded households nature sounds were associated with better SRH is difficult to explain and may again have to do with the intertwining of contextual factors and soundscape perception. However, our data were not suitable for exploring higher order interactions.

Strengths and Limitations
To our knowledge, the current study was the first to explore health effects of various forms of perceived sounds while people were forced to spend unprecedented amount of time in their homes. We also tested for the first time indirect pathways linking soundscape to health in residential environments through feelings of "being away", one important component of a person's environmental restorative quality at home. We collected data on both exposure frequency and pleasantness. This allowed us to explore the underpinning dimensionality of sound preference, which guided the differentiation between wanted and unwanted soundscape components. Finally, we accounted for multiple contextual (i.e., income and housing characteristic) and individual factors (i.e., environmental sensitivity and mental health), which could have confounded or modified the effects of soundscape on self-rated health.
Several limitations should also be acknowledged. Our sample size was modest and, therefore, especially our ordinal regression models could have lacked sufficient power to detect small total effects of human and nature sounds. In this vein, ratings of perceived soundscape and "being away" were narrowly distributed and little variation in these variables could have attenuated or even concealed potentially meaningful associations despite our use of robust estimation techniques. We used a convenience sample of students in health-related programs, which was restrictive as to participants' economic and nonprofessional activity, precluding generalization of findings to experiences of all youth during the COVID-19 epidemic. Moreover, an actual response rate could not be calculated given that students were also allowed to forward the survey link to their peers.
Next, a cross-sectional design does not allow making causal claims about the associations observed. It is possible that participants' health status determined their environmental awareness and salience of sounds. We attempted to control for this by introducing environmental sensitivity and mental health into the models. Reassuringly, the associations of interest were robust to confounding/effect modification by those factors. Furthermore, cross-sectional analyses of mediation may yield findings even in the absence of an indirect effect [137,138]. Still, our predictor, mediator, and outcome variables, even if measured at one point in time, satisfied the conceptual timing criterion (i.e., conceptual time-ordering) required for mediation [139].
Further, we only collected data on perceived health and soundscape. SRH has been found to agree reasonably well with physician-diagnosed health status and be a good predictor of adverse health outcomes in earlier studies [56][57][58][59]. However, we could not ascertain the predictive validity of SRH in our sample and we acknowledge that biomedical data used to establish the validity of SRH may vary over time, social, and age strata. In addition, we chose to focus on perceptual soundscape measures, owing to a lack of data on participant's residential address and the large uncertainty in expected sound exposures at the receiver's home. Out of privacy concerns, we did not collect information on participant's residential address, which thwarted linkage to existing noise maps. Even if such data were available, we could not rely on the official noise maps of Plovdiv, because of the profound and spatially variable changes observed in traffic flow and structure during the lockdown [23,140] and the fact that only half of the participants resided in Plovdiv in the period of interest. Moreover, typical average noise level measurements would not only underestimate participants' actual exposure to all sources at all sites both indoors and outdoors, but also carrying out measurements at participants' homes was not feasible during the state of emergency. We acknowledge that using psychoacoustic noise indicators would have allowed for better discrimination of sound pleasantness and other perceptual dimensions [30,52] and raised our confidence in the associations observed.
Finally, while we collected data on additional housing and lifestyle factors (e.g., exposure to green spaces, presence of pets in the household, and physical activity) [10], we deliberately used only a subset of these variables for the current study. Given that all these data were self-reported, there is a high potential for common-source bias and/or reverse causation, which would thwart interpretation of a more complex structural model. We also deliberately focused on a more narrowly defined model since the nature and quality of our data did not allow for strong inferences about more complex structures where "being away" would assume the lead role as a central link mediating competitive indirect pathways leading to SRH.

Conclusions
Sounds perceived by university students in their home during their COVID-19 quarantine were conceptually grouped into three distinct factors: mechanical, human, and nature sounds. Greater exposure to mechanical sounds was consistently associated with both lower restorative quality of the home and worse self-rated health. Apart from these observed direct effects, exposure to mechanical sounds related to lower perceived restorative quality of the home, and then to worse self-rated health, whereas nature sounds were associated with stronger feelings of escape, and in turn with better self-rated health. No association was found with human sounds. These findings suggest a role of positive indoor soundscape and restorative quality for promoting self-rated health in times of social distancing. Funding: A.M.D.'s work on this publication was partially supported by the National program "Young Scientists and Postdoctoral Candidates" of the Ministry of Education and Science, Bulgaria. The information and views set out in this paper are those of the authors and do not necessarily reflect the official opinion of the funder.

Institutional Review Board Statement:
The study was conducted according to the guidelines of the Declaration of Helsinki. The generic design of earlier studies in this series has been approved by the Ethics Committee at the institution of the principal investigator.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on reasonable request from the corresponding author. The data are not publicly available as per participants' informed consent conditions.