Exposure to noise—or unwanted sound—is considered a major public health issue in the United States and internationally. Previous work has shown that even acute noise exposure can influence physiological response in humans and that individuals differ markedly in their susceptibility to noise. Recent research also suggests that specific acoustic properties of noise may have distinct effects on human physiological response. Much of the existing research on physiological response to noise consists of laboratory studies using very simple acoustic stimuli—like white noise or tone bursts—or field studies of longer-term workplace noise exposure that may neglect acoustic properties of the noise entirely. By using laboratory exposure to realistic heating, ventilation, and air conditioning (HVAC) noise, the current study explores the interaction between acoustic properties of annoying noise and individual response to working in occupational noise. This study assessed autonomic response to two acoustically distinct noises while participants performed cognitively demanding work. Results showed that the two HVAC noises affected physiological arousal in different ways. Individual differences in physiological response to noise as a function of noise sensitivity were also observed. Further research is necessary to link specific acoustic characteristics with differential physiological responses in humans.
I. INTRODUCTION
Noise, or unwanted sound, is a significant source of annoyance and distress (Basner et al., 2015) and is increasingly considered a major public health issue in the United States (Hammer et al., 2014) and internationally (Theakston, 2011). Workplace and environmental noise impair cognitive performance (Banbury and Berry, 2005) and increase fatigue (Jahncke et al., 2011), absenteeism (Clausen et al., 2013), and susceptibility to chronic health problems (Basner et al., 2014; Kristiansen, 2010). Background noise may be particularly troublesome for individuals who use hearing assistive technologies and/or who have heightened noise sensitivity or hyperacusis (Jahncke and Halin, 2012; Shepherd et al., 2010). In this preliminary study, we examine affective physiological responses in individuals working in silence and in two acoustically distinct background noises to develop hypotheses to guide future work exploring the interaction between acoustic properties of annoying noise and individual differences in health-related consequences of occupational exposure to noise.
A. Physiological response to noise
External stressors such as noise engage either the hypothalamic-pituitary-adrenal (HPA) and/or the sympathetic-adrenal-medullary (SAM) stress response systems. HPA activity is particularly associated with psychosocial stressors, while stressors that demand application of physical and/or cognitive effort tend to involve arousal of the sympathetic nervous system and/or suppression of parasympathetic activity (Skoluda et al., 2015). In the case of occupational noise exposure, it seems quite likely that the physiological impact of noise may interact with psychosocial aspects of the workplace (interactions between work environment, job content, organizational conditions and a worker's capacities, needs, expectations, customs and culture, and personal extra-job conditions) (Joint ILO/WHO Committee on Occupational Health, 1986) leading to an involvement of both the HPA and SAM stress response systems (Love et al., 2019). Noise exposure that causes chronic arousal of these stress response systems has a profound negative effect on health, increasing a person's risk for cardiovascular disease, hypertension, and stroke (Passchier-Vermeer and Passchier, 2000).
Perhaps because of the observed relationship between noise exposure and cardiovascular health (Babisch, 2011; Davies et al., 2005; Davies and Van Kamp, 2012; Passchier-Vermeer and Passchier, 2000; Stansfeld and Matheson, 2003; Theakston, 2011), much of the experimental and observational research on physiological responses to noise focuses on measures associated with blood pressure over relatively long-term windows of exposure on the order of workdays or longer (e.g., Chang et al., 2003; Chang et al., 2015; Lusk et al., 2004). Here, we focus on autonomic (SAM axis) measures in part due to their association with cognitive effort, but also because there is evidence that noise exposure activates the SAM axis. For example, Kryter and Poza (1980) found a noise-related increase in peripheral capillary vasoconstriction (a decrease in a measure similar to the sympathetically governed blood volume pulse amplitude (BVPA), we discuss below) during exposure to noise (see also Ray et al., 1984). Subsequent research lends stronger support to the idea that acute noise exposure does affect a broader range of autonomic, especially cardiovascular, responses (see Lusk et al., 2004 for a review and additional supporting data; Chang et al., 2015; Kraus et al., 2013; Kristiansen et al., 2009). Thus, our measures include sympathetically governed responses like electrodermal activity (EDA) or BVPA and cardiovascular responses (heart period, heart rate variability).
1. EDA
EDA reflects sympathetically controlled eccrine sweat gland activity (Andreassi, 2007; Boucsein, 2012; Dawson et al., 2007). Here, we used two EDA measures: skin conductance level (SCL) and skin conductance response (SCR). SCL and SCR are correlates of sympathetic arousal and are associated with a wide range of emotions with both positive and negative valences (Kreibig, 2010) as well as with demand on working memory (Mehler et al., 2012), sustained attention (Davies and Krkovic, 1965), and task engagement (Dawson et al., 2007). SCL reflects the background/tonic level of electrodrodermal activity, while SCRs reflect rapid/phasic changes in conductance. SCRs can be further subcategorized into non-specific skin conductance responses (NS-SCRs), which occur in the absence of an identifiable eliciting event and event-related skin conductance responses (ER-SCR), which can be reasonably attributed to a specific eliciting stimulus.
2. Heart period
Changes in heart rate are modulated by both sympathetic and parasympathetic activity (Berntson et al., 2007). Many studies have shown increased heart rate (decreased heart period) in demanding task conditions (Backs and Seljos, 1994; Carroll et al., 1986; Carroll et al., 2009; Kennedy and Scholey, 2000; Turner and Carroll, 1985). Results from previous research on the effects of occupational noise exposure on heart rate are mixed, with some studies showing no effect (Lusk et al., 2002; Ray et al., 1984), while others suggest at least a limited response to specific types of noise, including sudden, sharp sounds (Lusk et al., 2004) or long-term exposure to high intensity sound (Goyal et al., 2010). Here, we report measures of heart period, not heart rate, because measures of change in heart period are more linearly related to changes in autonomic activity than are measures of change in heart rate, and therefore heart period has been argued to be more appropriate when comparing changes in cardiac autonomic activity across individuals (Berntson et al., 1995).
3. Heart rate variability
The human heart does not beat at a perfectly steady rate, but rather exhibits some degree of beat-to-beat variability in duration. For example, respiratory sinus arrhythmia (RSA) is a measure derived by comparing the heart period during inspiration and expiration (Grossman et al., 1990). Under normal circumstances, the heart period is slightly shorter during inspiration and slightly longer during exhalation. Differences between individuals in resting RSA may be associated with trait characteristics including cognitive flexibility (Thayer et al., 2009) and emotional state (Oveis et al., 2009) and some studies have also shown task-dependent changes in RSA such that decreased RSA is associated with greater cognitive task demands (Muth et al., 2012; Overbeek et al., 2014). There are many other methods for measuring heart rate variability (Berntson et al., 2007) but we focus here on RSA calculated from a combination of thoracic strain-gauge measurement of respiration and electrocardiography (ECG) as these methods arguably provide a more reliable (though far from perfect) assessment of heart rate variability (Quintana and Heathers, 2016).
4. BVPA
BVPA is a measure of change in capillary dilation between systole and diastole, reflecting blood vessel wall resistance. BVPA is governed purely by the sympathetic nervous system, with heightened sympathetic arousal leading to decreased capillary dilation. BVPA has been shown to decrease in response to noise exposure (Ray et al., 1984), and in response to increasing demands of cognitive tasks such as the Stroop task (Tulen et al., 1989), mental arithmetic (Goldstein and Edelberg, 1997), and working memory load (Iani et al., 2004). Decrease in pulse volume has been linked specifically to the increased investment of mental effort in a task such that pulse amplitude decreases parametrically with increase in working memory load (Iani et al., 2004). However, decrease in BVPA is also associated with mostly negative emotions (fear, disgust, anger, sadness, and anxiety) (Kreibig, 2010), and thus may be associated with noise-related annoyance or frustration as well (cf. suggestion by Francis et al., 2016).
5. Facial electromyography (EMG)
Facial EMG is a method for quantifying the electrical activity in facial muscles and is one of the predominant physiological measures for assessing affective valence (Cacioppo et al., 1986; Potter and Bolls, 2012). Here, we examine activity in corrugator supercilii, sometimes termed the “frown muscle,” which is located above the inner corner of each eye. It serves to furrow or wrinkle the center of the brow, pulling the eyebrows together. Although this response may occur in bright light to assist in shielding the eyes, its primary role seems to be related to emotion, especially negative emotions such as anger and discomfort (Larsen et al., 2003). Corrugator supercilli activity behaves similarly in response to acoustic stimuli as with visual stimuli—activity increases with unpleasant sounds compared to neutral sounds (Bradley and Lang, 2000).
B. Properties of annoying noise
Extant studies have used broad spectrum white noise (Keith et al., 2019; Lee et al., 2010), the International Collegium of Rehabilitative Audiology Noise (ICRA-noise) (Shoushtarian et al., 2019; see Dreschler et al., 2001 for description of stimuli), traffic noise, and speech (Sim et al., 2015) at various intensities to assess the effect of noise level on autonomic arousal, showing that even low-level noise elicits an instant and significant activation of the sympathetic nervous system. The current study uses realistic sounds recorded from heating, ventilation, and air conditioning (HVAC) equipment that are similar in level but differ in other psychoacoustic properties (see Sec. II B for description).
Psychoacoustic properties of sound that have been linked to affective responses in humans discussed in the previous section include loudness (e.g., as calculated according to a revised version of the Zwicker model, DIN 45631; see Zwicker and Fastl, 1999), sharpness (high frequency emphasis; Carr and Davies, 2017; von Bismarck, 1974), roughness (quasi-periodic variability in amplitude and frequency in the 15–300 Hz range; Daniel and Weber, 1997), fluctuation (quasi-periodic variability of amplitude and frequency in the 1–20 Hz range; Zwicker and Fastl, 1999), and tonality (related to harmonic-to-noise ratio; Bienvenue et al., 1991). Research by Västfjäll et al. (2003) and Västfjäll (2012) shows that these acoustic properties load onto different components of a two-dimensional affective space structured in terms of valence (e.g., pleasant to unpleasant) and arousal (e.g., low activation to high activation), a structure that is commonly accepted in the study of human emotional response to visual and auditory stimuli (Cacioppo and Berntson, 1994; Lang et al., 1990). Loudness and roughness are most clearly associated with valence (Västfjäll et al., 2003; Västfjäll, 2012), while sharpness and tonality relate better to arousal (Västfjäll, 2012). In addition, Persson Waye and Ohrstrom (2002) found that the strength of low frequency (20–200 Hz) components caused by certain industrial sound sources such as HVAC systems and wind turbines (Baliatsas et al., 2016; Berglund et al., 1996; Schäffer et al., 2017) may contribute strongly to a sense of noise annoyance, though it is not yet known whether this occurs via an influence on valence, arousal, or both. Though distinct acoustic properties of noise may differentially affect physiological responses, distinguishing between valence and arousal is beyond the scope of the present work. Moreover, because we are dealing with affective responses, it is also likely that individual differences in responsivity to noise will play an important role in the degree or nature of the physiological changes measured.
C. Individual differences in noise sensitivity
For the present purpose of discussing occupational noise exposure, we must distinguish between noise-sensitivity as a general trait and clinical conditions such as recruitment, hyperacusis, and misophonia. Loudness recruitment—though also categorized as a disorder of sound tolerance—is primarily a phenomenon associated with sensorineural hearing loss. Audiometric thresholds for individuals included in this analysis fell within normal limits; therefore, the question of loudness recruitment likely falls outside of the scope of the current study. Hyperacusis may refer to multiple phenomena that likely have distinct etiologies (e.g., loudness hyperacusis vs annoyance hyperacusis), but the overarching sense is one of a pathological degree of discomfort from particular sounds, or types of sounds (Paulin et al., 2016; Tyler et al., 2014). Similarly, misophonia is characterized by extreme, negative emotional responses (irritation, anger, disgust, anxiety) to specific sounds (Taylor, 2017) and is often co-morbid with hyperacusis (Duddy and Oeding, 2014).
In contrast to both hyperacusis and misophonia, noise sensitivity is typically discussed in terms similar to that of a personality trait as ranging along a continuum in the entire population. For example, Shepherd et al. (2015) describe noise sensitivity as a stable trait that is manifest in roughly 20%–40% of individuals, with about 12% exhibiting high noise sensitivity. Noise sensitivity is also associated with a variety of chronic clinical conditions including autism and traumatic brain injury (Dischinger et al., 2009; Stiegler and Davis, 2010). Noise-sensitive individuals appear to have normal auditory acuity (Ellermeier et al., 2001) but are more likely to have poor health-related quality of life (Shepherd et al., 2010) and to miss work, especially if they have complex, mentally demanding jobs (Fried et al., 2002). Noise-sensitive people also experience greater interference with job performance from workplace noise, especially when performing mentally demanding tasks (Belojević et al., 1992), and find working in noise more annoying and more mentally straining than do less noise-sensitive people (Sandrock et al., 2009). Noise sensitivity may also be related to specific personality traits, especially the introversion-extroversion axis (Shepherd et al., 2015) and neuroticism (Belojević et al., 1997; Öhrström et al., 1988), though recent research by Heinonen-Guzejev et al. (2018) suggests rather a biological origin.
Regardless of the specific psychological or physiological basis for noise sensitivity, there is a broad consensus that noise-sensitive individuals should manifest greater physiological responses to workplace noise than do less noise-sensitive people, especially if the noise interferes with cognitively demanding work. Indeed, one hypothesis for how workplace noise may induce physiological stress is that noise introduces repeated distraction or interruption from achieving work-related goals (van de Poll and Sörqvist, 2016) which is stressful (Baethge et al., 2015). However, no clear description of differences in physiological response of noise-sensitive vs noise-insensitive people has yet emerged (Notbohm et al., 2013).
D. Summary
In the present study, we examine physiological variables related to autonomic nervous system arousal collected while participants performed a cognitively demanding task in silence and in two types of background sounds that might realistically be encountered in a workplace setting. Two HVAC noises were selected that were similar in level [70.9 dB(A) and 71 dB(A)] but that differed in terms of a variety of other annoyance-related variables (e.g., roughness, tonality, sharpness, fluctuation strength). This analysis focuses on differences in observed affective response both between noise groups (silence, C1, C2) and within groups at the level of the individual. The goal of this study was to investigate the influence of naturalistic, workplace noise on physiological and psychological markers of stress, and to explore how individual differences contribute to variability in observed health-related consequences of occupational noise exposure.
II. METHODS
A. Participants
Sixty-one participants were recruited for this study from Purdue University and the surrounding community using ads placed on campus and in a university electronic newsletter. Of these, datasets from ten individuals were excluded from analysis for the following reasons: anxiety scores (2) or hearing thresholds (2) above acceptable range, equipment failure or operator error recording physiological data (6). Of the remaining 51 participants (mean age 23.1 years, range 18–39; 29 cisgender women, 22 cisgender men), 36 reported speaking a North American variety of English as their native language, while 15 did not. Forty-six reported being right-handed, four left-handed, and one ambidextrous.
All participants passed standard screening assessments of anxiety (Geriatric Anxiety Index; Pachana, et al., 2007; group mean = 1.9; range = 0–7) and depression (Geriatric Depression Scale short form; Yesavage and Sheikh, 1986; group mean 1.1, range 0–6). Hearing screenings were conducted according to standard practice in the audiology clinic (GSI 18 travelling screening audiometer, pure tone thresholds at octave intervals from 250 to 8000 Hz < 25 dB sound pressure level, SPL). Current tobacco users were excluded from participation. Participants were asked to refrain from consuming any caffeine within 2 h of the start of the study, and none reported currently being under the influence of any drug or alcohol (see supplementary material1 for a detailed description of participant properties).
B. Materials and stimuli
1. Working memory task
The cognitive task was a standard N-back task in which participants saw a series of letters presented one at a time on a computer screen and were asked to respond whenever a given stimulus matched a specific target condition. Stimuli consisted of the written capitalized letters A, B, C, D, E, F, G, H, I, J. In the 0-back version of the task, the target is a designated stimulus (e.g., the letter “E”). In the 1-back condition, a target is any stimulus that matches the stimulus immediately before it (i.e.,1 stimulus back in time), while in the 2- and 3-back conditions targets are those stimuli identical to stimuli two- and three-steps back in time, respectively. In each block, stimuli were presented at a uniform rate for 500 ms with a 1000 ms inter-stimulus interval (ISI). Participants responded by pressing a button whenever a target appeared using their dominant index finger on a response box (Cedrus RB-730) resting on a TV table placed over their lap. Responses made fewer than 100 ms or more than 1500 ms after the onset of the stimulus were treated as non-responses (i.e., counted as misses for target stimuli, or as correct rejections for non-targets).
2. Individual difference measures (Table I)
Hearing acuity. To quantify hearing acuity, pure tone averages (PTA) were calculated for each participant following best practices established by the American Speech-Language-Hearing Association (1978)—the mean of thresholds at 500, 1000, and 2000 Hz were calculated, and the best (lowest) value for the two ears was used. This allowed us to convert an approximation of the audiogram into a single number.
Personality. “Big 5” personality traits (John and Srivastava, 1999) were quantified using the validated Big Five Inventory-10 (BFI-10) (Rammstedt and John, 2007). This is a self-assessment questionnaire consisting of questions such as, “I am someone who is outgoing and sociable,” to which participants respond using a 5-point scale (strongly agree = 5, agree = 4, neutral = 3, disagree = 2, strongly disagree = 1). Scores are averaged over the two questions associated with each subscale: extraversion, agreeableness, conscientiousness, neuroticism, and openness. We recognize that the short-form assessment may be less sensitive than the more extensive original long-form assessment; however, for purposes of comparison with previous studies and following studies validating the short BFI, it was chosen for use here (Rammstedt and John, 2007).
Noise sensitivity. We used the noise-sensitivity-questionnaire (NoiSeQ) (Schütte et al., 2007) as a measure of noise sensitivity. The NoiSeQ is a 35-item, self-assessment questionnaire consisting of five 7-question subscales covering sensitivity to noise at home, during leisure, in communication, at work, and during sleep. Items consist of a statement such as “When people around me are noisy, I have trouble getting my work done” that are rated on a 5-point scale (strongly agree = 5, agree = 4, neutral = 3, disagree = 2, strongly disagree = 1). The different subscales are, at least in principle, potentially orthogonal.
Working memory task performance. Performance in each block was quantified as the sensitivity index d' using the formula: d' = z(hits) – z(false alarms) (Macmillan and Creelman, 2004), where hits refers to the proportion of target trials on which the participant responded correctly, and false alarms refers to the proportion of non-target trials on which the participant failed to withhold their response (i.e., responded incorrectly). Perfect scores were adjusted using the substitution of 1 – 1/(2n) for hits = (1.00), and 1/(2n) for false alarms = (0.00), where n is the total number of targets (for hits) or non-targets (for FAs) in a block (Macmillan and Creelman, 2004). Scores for the two blocks of each N-back level (i.e., 0-back, 1-back, etc.) were combined (i.e., the two blocks were treated as a single test sequence, with a total of 100 trials of which 20 were targets and 80 were non-targets), resulting in four scores for each participant in each noise condition, one for each N-back level from 0- to 3-back.
Annoyance ratings. After each noise condition, participants also completed the NASA task load index questionnaire (NASA TLX; Hart and Staveland, 1988). This is a six-item instrument in which participants use a 20-point equal-appearing interval scale to rate their estimation of the mental, physical, and temporal demand of a task, as well as the effort and frustration they experienced while performing it and their sense of their performance on the task.
Summary table of study measures and associated tests.
Property . | Associated Test/Measure . | Abbreviation . |
---|---|---|
Noise sensitivity | Noise-sensitivity-questionnaire | NoiSeQ |
Hearing acuity | Pure tone average | PTA |
Annoyance (mental, physical, and temporal demand; performance, effort, frustration) | NASA TLX questionnaire | NASA TLX |
Working memory capacity | d' on N-back task | WMC |
Personality traits | Big Five inventory | BFI |
Electromyographic (EMG) | Corrugator supercillii | - |
Cardiac (ECG) | Heart period; heart rate variability (respiratory sinus arrhythmia) | RSA |
Vascular (PPG) | Blood volume pulse amplitude | BVPA |
Electrodermal (EDA) | Skin conductance level; skin conductance response; skin conductance response rate | SCL SCR SCRR |
Property . | Associated Test/Measure . | Abbreviation . |
---|---|---|
Noise sensitivity | Noise-sensitivity-questionnaire | NoiSeQ |
Hearing acuity | Pure tone average | PTA |
Annoyance (mental, physical, and temporal demand; performance, effort, frustration) | NASA TLX questionnaire | NASA TLX |
Working memory capacity | d' on N-back task | WMC |
Personality traits | Big Five inventory | BFI |
Electromyographic (EMG) | Corrugator supercillii | - |
Cardiac (ECG) | Heart period; heart rate variability (respiratory sinus arrhythmia) | RSA |
Vascular (PPG) | Blood volume pulse amplitude | BVPA |
Electrodermal (EDA) | Skin conductance level; skin conductance response; skin conductance response rate | SCL SCR SCRR |
3. Baseline video
To allow for the collection of individual baseline physiological values, participants were asked to sit quietly in the sound booth and watch a 3-min segment of a relaxation video showing waves rolling in on a sandy beach (Isis Visuals, Inc.). The baseline video was shown as the first and last component of the experimental session.
4. Background noises
Auditory stimuli consisted of two different 20 s recordings of operating heating, ventilation, air conditioning, and refrigeration (HVAC&R) equipment that were further modified for use in a study of conscious assessment of noise annoyance (Sung et al., 2018). The two noises were selected because they are comparably loud with similar average intensity while being otherwise acoustically distinct according to a variety of properties associated with noise annoyance (see Table II and Fig. 1). Annoyance ratings were collected by asking participants to rate the sound from “not at all annoying” to “extremely annoying” (numbers on the annoyance scale correspond to: 2 = “not at all annoying,” 3.5 = “slightly annoying,” 5 = “moderately annoying,” 6.5 = “very annoying,” and 8 = “extremely annoying”). Note that the data presented in Table II are based on listener assessments of 6 s of sound. They therefore may not reflect the actual annoyance imparted by these sounds in the present experiment. To assess differences in noise annoyance in the present experiment, we used the NASA TLX dimensions of effort and frustration (see materials section: NASA TLX). While original sound levels from the Sung (2018) study are reported in dB(A), we also measured the sound level of each noise as presented to participants in units of A-weighted equivalent continuous sound level (LAeq). These sounds were presented at an LAeq of 59.5 (sound C1) and 54.5 (sound C2) as measured over a 60-s period at the location of the listener's head using a Brüel & Kjaer model 2230 sound level meter. LAeq values reflect the level at which a steady state noise would have to be presented to produce the same total energy over the same amount of time as the measured dynamic sound. It is a commonly accepted method of reporting sound level for sources, like HVAC systems, whose mechanical components introduce variation in momentary sound level over time. Leq is the standard for noise exposure research thanks to its usefulness in assessing noise dose in workplace noise exposure (Brüel & Kjærr, 2021). We selected only two noises for this preliminary study, but future research should be designed to parametrically vary each of these dimensions independently, as needed.
Acoustic characteristics and average subjective annoyance ratings associated with silence (C0), sound C1, and sound C2, as reported in Sung et al. (2018). Note: Sharpness is calculated using the Aures method (Aures, 1985).
. | Loudness . | dB(A) . | Roughness . | Tonality . | Fluctuation strength . | Aures* sharpness . | Average annoyance rating . |
---|---|---|---|---|---|---|---|
C1 | 28.1 | 71 | 5.44 | 0.17 | 0.012 | 1.54 | 4.81 |
C2 | 27.1 | 70.9 | 2.51 | 0.39 | 0.008 | 3.24 | 6.80 |
. | Loudness . | dB(A) . | Roughness . | Tonality . | Fluctuation strength . | Aures* sharpness . | Average annoyance rating . |
---|---|---|---|---|---|---|---|
C1 | 28.1 | 71 | 5.44 | 0.17 | 0.012 | 1.54 | 4.81 |
C2 | 27.1 | 70.9 | 2.51 | 0.39 | 0.008 | 3.24 | 6.80 |
Narrowband spectrograms and waveforms (top and bottom left) and spectra (top and bottom right) of sound C1 (left) and sound C2 (right).
Narrowband spectrograms and waveforms (top and bottom left) and spectra (top and bottom right) of sound C1 (left) and sound C2 (right).
C. Procedure
1. Testing environment
Participants were comfortably seated in a large, single-walled sound booth (IAC), facing a 50 in. television screen suspended on the wall of the booth approximately 2 m directly in front and slightly above head height of the participant. Visual stimuli were presented on the screen in approximately 20-point type using Tahoma font, in black on a light gray background. Audio stimuli were presented via a Hafler M5 reference speaker placed at floor level directly behind the participant. Participants placed their dominant hand on a 7-button response box (Cedrus RB-730) set on a TV tray stand positioned over their lap.
2. Block design
The three noise conditions (C1, C2, and silence) were counterbalanced to control for order effects. Participants completed 8 blocks of N-back trials within each noise condition—two each of 0-, 1-, 2-, and 3-back trials presented in this order. Each N-back block, which included 50 stimuli—40 non-targets and ten targets—took roughly 76 s. This varied slightly between participants, as they advanced through the instructions between blocks at their own pace. The noise continued throughout the series of N-back blocks and intervening instructions. The noise stopped following the final 3-back trial, and participants were instructed to record their annoyance ratings on paper (∼5 min) before pressing a button to initiate the next noise and set of N-back trials. This allowed for a physiological recovery interval between each noise condition that was approximately 5 min long, but that varied slightly per individual based on reading speed and how long they took to fill out the NASA TLX.
3. Psychophysiological measurements (Table I)
Electrodermal activity (EDA), ECG, photoplethysmography (PPG), and facial electromyography (EMG) data were simultaneously collected using corresponding Biopac modules set according to the manufacturer's recommendations (see supplementary material1 for a detailed description of data collection methods).
D. Analyses
1. Preprocessing of data
Behavioral variables (working memory task score, noise sensitivity score) were scaled and centered before analysis to help with model convergence and comparability of coefficients across predictors in multilevel modeling as suggested by Bates et al. (2015) and Eager and Roy (2017). Working memory task scores were scaled such that the scores ranged from –1.70 to 2.07 and centered such that the variable's mean was equal to zero. Noise sensitivity scores were likewise scaled such that the scores ranged from –2.06 to 1.26 and centered such that the variable's mean was equal to zero.
Following data collection, physiological signals were pre-processed following standard procedure to ensure signal quality (see supplementary material1 for detailed description). Relevant measures—SCL, heart period, heart rate variability, and BVPA—were then derived from the original signal following standard protocol (see supplementary material1 for a detailed description). The SCR rate measure used was derived by converting the total number of SCRs (both event-related and spontaneous/non-specific) per test block into a rate measure, SCR rate (SCRs per second). Prior to statistical analysis, all physiological variables were normalized against values calculated during the two baseline periods conducted before and after the experimental blocks. Values collected during the middle minute of each 3-min video were averaged to obtain a baseline value for each physiological measurement. Here, we use baseline simply to normalize physiological values across individuals, and not to assess individual differences in baseline autonomic measures. The assessment of trait-like “resting state” values would likely require considerably longer baseline durations. Although some studies have begun to use shorter periods for resting state measurements, results are not always consistent with those derived from longer periods (Kim et al., 2021; Shaffer et al., 2020). Average physiological values collected during experimental blocks (76 s) were then divided by the participant's baseline value, resulting in a ratio indicating change from baseline. The logarithm (base 10) of this ratio was then used as the dependent measure in all analyses of physiological variables.
2. Multi-level mixed effects modelling
To identify relationships between background noise type and physiological and behavioral responses associated with effort or stress, we employed multi-level mixed effects modeling (lmer) implemented in lme4 (v. 1.1–21) (Bates et al., 2015) within the R (v. 3.6.3) programming environment (R Development Core Team, 2018) (see supplementary material1 for model information). We performed analysis of deviance testing on the fitted model using type II Wald F tests with Kenward Roger degrees of freedom via the car package (v. 3.0–7, Fox and Weisberg, 2019). Participants were included within the random effect structure for all dependent measures. Level of working memory demand (level of the N-back task, i.e., 1-back, 2-back, or 3-back) was also included in the model for physiological and working memory performance measures. This factor was not included for self-rated measures of effort (NASA TLX scores), however, because task load index (TLX) scores were only collected after each noise exposure, not after each level of the N-back task. We calculated post hoc analyses by comparing estimated marginal means (EMMs) with the commands emmeans (for means comparisons) and emtrends (for comparing slopes of trends) within the emmeans package (v. 1.4.7) (Lenth et al., 2020) using Tukey adjustment for multiple comparisons.
III. RESULTS
A. Personality and noise sensitivity
A multiple linear regression of the five BFI scores against noise sensitivity showed no significant relationships, suggesting that none of the measured personality factors are a significant predictor of noise sensitivity in the present dataset (see supplementary material1 for summary data). An assessment of the relative importance of the five measures using the relaimpo (v. 2.2–3) package (Grömping, 2006) for comparing relative contributions of regressors in linear models (using the recommended lmg method) showed that the five BFI factors only accounted for 8.73% of the variance in noise sensitivity. Among the five factors, openness contributes the largest fraction of this total (42.78%), followed by extroversion (24.67%), conscientiousness (17.42%), agreeableness (13.77%), and neuroticism (1.35%).
B. Working memory task performance
Participants scored an average d' of 2.47 across all blocks (standard deviation, SD = 1.13; range = 0.828–3.89), but scores differed significantly according to N-back level (see supplementary material1 for summary data). Results of multilevel modeling using scaled and centered values showed significant contributions from N-back level, F(2, 816) = 588.81, p < 0.0001, and a significant interaction between N-back level and noise sensitivity, F(2, 816) = 3.70, p = 0.003, as shown in Fig. 2. The relationship between noise sensitivity and d' is such that performance decreases with increasing noise sensitivity, but only in the 1- and 2-back conditions. Analysis of the estimated marginal means of linear trends also shows a significant difference (p < 0.05 with Tukey correction) in the slope of the linear relationship between scaled noise sensitivity score and scaled performance between the 1- and 3-back conditions and between the 2- and 3-back conditions. As expected, Tukey-corrected post hoc analysis of estimated marginal means confirmed a significant (p < 0.001) difference in mean performance between all three N-back levels (i.e., 1-back > 2-back > 3-back).
Working memory task performance (scaled d') as a function of scaled composite NoiSeQ score, by N-back level. Each dot represents an individual's performance in a given condition. Lines show linear regression.
Working memory task performance (scaled d') as a function of scaled composite NoiSeQ score, by N-back level. Each dot represents an individual's performance in a given condition. Lines show linear regression.
C. TLX ratings
Overall, participants showed little difference in responses to the NASA TLX questionnaire across different background noise conditions (see supplementary material1 for summary data). Separate multilevel modeling of the relationship between each TLX score (mental demand, physical demand, temporal demand, effort, frustration, and performance) revealed a significant effect of noise type, but only for measures of effort, F(2, 96) = 3.41, p = 0.037; frustration, F(2, 96) = 4.44, p = 0.014 and temporal demand, F(2, 95.05) = 3.29, p = 0.042. Examination of post hoc analyses in each case showed the following: for effort, there was a significant difference only between the silence < noise C1 conditions (p = 0.030); for frustration, there was a significant difference between silence < noise C1 (p = 0.049) and between silence < noise C2 (p = 0.02); and for Temporal Demand there was a significant difference only between the silence < noise C2 conditions (p = 0.033).
D. Physiological measures
Individual datasets for a particular physiological measure were excluded from analysis if normalized individual scores fell outside the range of ±3 SD from the group mean. A total of 11 participants exhibited at least one physiological measure falling outside of this range. This did not preclude an individual's other physiological measure datasets from being included in other analyses.
1. SCL
Four participants (S102, S110, S135, and S140) were removed from this analysis for exhibiting normalized scores outside the range of ±3 SD from the group mean. The remaining 47 participants showed a mean SCL of 5.66 μS (SD = 4.30, range = 0.80–26.17; see supplementary material1 for summary data). Results of multilevel modeling using scaled and centered values of SCL showed one significant main effect of block, F(1, 748) = 9.20, p = 0.002, and no other significant effects or interactions (p > 0.10 for all tests). Further analysis confirmed that the effect of block is driven by a small but reliable overall decrease in SCL from block 1 to block 2, commensurate with habituation to task.
2. SCR rate
Two participants (S111, S242) were removed from this analysis for exhibiting normalized scores outside the range of ±3 SD from the group mean. The group mean rate of spontaneous and event-related SCR was 0.029 SCRs per second (SD = 0.030, range = 0–0.17; see supplementary material1 for summary data of log ratio scores). Multilevel modeling of the log ratio rates showed a significant effect of N-back level, F(2, 782) = 5.15, p =0.005, a significant effect of block, F(1, 782) = 6.86, p = 0.008, and a significant interaction between noise type and noise sensitivity, F(2, 782) = 4.04, p = 0.018. Analysis of estimated marginal means confirms that the effect of N-back level is driven mainly by a greater value of Skin Conductance Response Rate (SCRR) in the 1- and 2-back conditions as compared to the 3-back (1 vs 3, p = 0.015; 2 vs 3, p = 0.015, both Tukey adjusted). The effect of block is a result of a decrease in SCRR from the first to the second block of trials. As shown in Fig. 3, the interaction between noise type and noise sensitivity appears to manifest in terms of a general tendency for change in SCRR from baseline to decrease as noise sensitivity score increases. Analysis of estimated marginal means of linear trends found that the slopes associated with noises C1 and C2 were significantly different (Tukey adjusted p = 0.027), while those associated with silence and C2 were marginally so (Tukey adjusted p = 0.05).
Scaled rate of SCRs as a function of scaled composite NoiSeQ Score. Each dot represents a single individual's performance in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
Scaled rate of SCRs as a function of scaled composite NoiSeQ Score. Each dot represents a single individual's performance in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
3. Heart period
Three participants were excluded from this analysis for exhibiting scores more than three SDs outside the group mean (S133, S135, S156). Mean heart period was 0.810 s (SD = 0.132, range 0.400–1.259), or about 74 beats per minute, with a range from 48 to 150 BPM (see supplementary material1 for summary data of log ratio scores). Results of multilevel modeling of the log ratio scores show a significant effect of both noise type, F(2, 782) = 10.97, p < 0.0001, and of N-back Level, F(1, 782) = 16.90, p < 0.0001. Examination of estimated marginal means suggests that RR intervals, the interval between two consecutive R peaks in the ECG signal, were lower in the experimental conditions than during the baseline (resulting in mostly negative log ratios), corresponding to a higher heart rate during the task than at rest, as expected. The significant effect of noise type is due to both noise conditions exhibiting a significantly more negative log ratio than the silence condition (Tukey adjusted p < 0.001 for both), suggesting that in the noise conditions listeners showed a stronger decrease in heart period (i.e., greater acceleration of heart rate) relative to baseline than they did in the silent condition (Fig. 4). Similarly, the effect of N-back Level is manifest primarily in terms of overall more negative values (greater increase in heart rate over baseline) in the 2-back and 3-back conditions as compared to the 1-back condition (Tukey adjusted p < 0.0001 for both) and silence. Thus, participants showed the expected task-related decrease in heart period, and this effect was stronger in the more cognitively demanding task conditions as well as in noise compared to the silence conditions.
Scaled heart period as a function of Noise type. Boxes extend between the 25th and 75th percentile. Whiskers indicate the range 1.5 times the inter-quartile range above and below these boundaries. Dots indicate individual outliers beyond that range. The central bar reflects the median of the distribution. Square dots indicate group means.
Scaled heart period as a function of Noise type. Boxes extend between the 25th and 75th percentile. Whiskers indicate the range 1.5 times the inter-quartile range above and below these boundaries. Dots indicate individual outliers beyond that range. The central bar reflects the median of the distribution. Square dots indicate group means.
4. Heart rate variability
Three participants (S115, S133, S156) were excluded from this analysis for exhibiting normalized RSA scores outside of three SDs from the group mean. Mean RSA was 78.1 ms (SD = 74.6, range 6.4–681.6; see supplementary material1 for summary data of log ratio scores). Results of multilevel modeling of RSA (Fig. 7) showed a significant effect of noise type, F(2, 782) = 6.04, p = 0.002, N-back Level, F(2, 782) = 11.42, p < 0.0001, and a significant interaction between noise type and noise sensitivity, F(2, 782) = 4.38, p = 0.013. Participants exhibited significantly lower RSA scores in the 2- and 3-back conditions than in the 1-back (Tukey adjusted p < 0.001 for both pairs), as expected (recall that a decrease in RSA is associated with greater cognitive task demands). Furthermore, noise C1 incurred the greatest decrease in RSA, and this value was significantly lower than that for either silence (Tukey adjusted p = 0.002) or noise C2 (Tukey adjusted p = 0.021) [Fig. 5(a)]. Finally, as shown in the graph of the interaction between noise type and noise sensitivity [Fig. 5(b)], change in RSA from baseline tended to rise with increasing noise sensitivity. Post hoc trend analysis showed that the slopes of the effect of noise sensitivity on RSA differed between the C1 condition and silence (Tukey adjusted p = 0.009), suggesting that noise sensitivity had the strongest effect on RSA during exposure to noise C1.
(a) Scaled heart rate variability as a function of Noise type. Boxes extend between the 25th and 75th percentile. Whiskers indicate the range 1.5 times the inter-quartile range above and below these boundaries. Dots indicate individual outliers beyond that range. The central bar reflects the median of the distribution. Square dots indicate group means. (b) Scaled heart rate variability (quantified as respiratory sinus arrythmia, or RSA) as a function of scaled composite NoiSeQ score and noise type. Each dot represents a single individual's performance in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
(a) Scaled heart rate variability as a function of Noise type. Boxes extend between the 25th and 75th percentile. Whiskers indicate the range 1.5 times the inter-quartile range above and below these boundaries. Dots indicate individual outliers beyond that range. The central bar reflects the median of the distribution. Square dots indicate group means. (b) Scaled heart rate variability (quantified as respiratory sinus arrythmia, or RSA) as a function of scaled composite NoiSeQ score and noise type. Each dot represents a single individual's performance in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
5. BVPA
One participant was excluded from this analysis for exhibiting scores more than three SD outside the group mean (S243). Mean BVPA was 0.355 (SD = 0.214, range 0.074–0.962; see supplementary material1 for summary data of log ratio scores). Results of multilevel modeling of log ratio scores did show a significant effect of noise type, F(2, 811) = 4.44, p = 0.012, and a significant interaction between noise type and noise sensitivity, F(2, 811) = 3.97, p = 0.019. Post hoc analysis of the effect of noise type showed only one significant pairwise difference, such that C2 differed significantly from silence (Tukey adjusted p = 0.009). Analysis of the interaction between noise type and noise sensitivity showed that the linear effects of noise sensitivity on BVPA for both C1 and C2 are significantly different from that for silence (Tukey adjusted p = 0.042 and 0.037, respectively), suggesting that as noise sensitivity increases, mean pulse volume decreases (i.e., arousal increases) in silence, but this trend is not seen during exposure to noise C1 or C2 (see Fig. 6).
Scaled BVPA as a function of scaled composite NoiSeQ Score, distinguished by Noise type. Each dot represents a single individual's scaled BVPA response in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
Scaled BVPA as a function of scaled composite NoiSeQ Score, distinguished by Noise type. Each dot represents a single individual's scaled BVPA response in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
6. Corrugator EMG
One participant was excluded from this analysis for exhibiting scores more than three SD outside the group mean (S116). Mean rectified corrugator amplitude was 0.030 mV (SD = 0.049, range 0.005–0.482; see supplementary material1 for summary data of log ratio scores). Results of multilevel modeling of corrugator EMG log ratio scores showed a significant effect of N-back level, F(2, 799)= 5.79, p = 0.003, and a significant interaction between noise type and noise sensitivity, F(2, 799) = 5.04, p = 0.006. Examination of analysis of estimated marginal means in the effect of N-back level shows a significant difference only between levels 1 and 3 (Tukey adjusted p-value = 0.002). Analysis of estimated marginal means of linear trends found that the slopes of the relationship between noise sensitivity and scaled corrugator EMG are significantly different between noise C1 and both the silence (Tukey adjusted p = 0.007) and C2 (adjusted Tukey p = 0.046) conditions. As shown in Fig. 7, this result suggests that the effects of noise C1 on corrugator EMG response varied more as a function of noise sensitivity than did the effects of either silence or noise C2.
Scaled corrugator EMG activity as a function of scaled composite NoiSeQ score, distinguished by Noise type. Each dot represents a single individual's scaled EMG response in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
Scaled corrugator EMG activity as a function of scaled composite NoiSeQ score, distinguished by Noise type. Each dot represents a single individual's scaled EMG response in a given Noise type, shape-coded by Noise type. Lines show linear regression calculated for each Noise type (solid line for silence, dashed line for C1, dotted line for C2).
IV. DISCUSSION
A. Personality and noise sensitivity
In the present sample, we found no significant relationship between BFI measures of personality traits and responses on the NoiSeQ self-report survey of sensitivity to noise. This contrasts with previous studies that have generally showed a significant relationship between NoiSeQ scores and the introversion-extroversion axis of the BFI (Shepherd et al., 2015). However, it must be noted that the present sample differs from that of Shepherd et al. (2015) in several key areas: the current study has (1) a smaller sample size, (2) a narrower range of participant ages, and (3) reported ranges of noise sensitivity scores, whereas the Shepherd study reported none (see supplementary material1 Table S14 for a detailed summary). Thus, it is difficult to draw any useful conclusions from a simple failure to disprove a null hypothesis. Further research on this topic should ensure that participants exhibit a sufficiently broad range of all traits examined and may be advised to employ more sophisticated statistical methods that would allow stronger conclusions to be drawn.
B. Working memory task performance
The manipulation of N-back Level was designed to increase working memory demand in a parametric way. Although previous research suggested that the negative effects of background noise should be exaggerated in conditions involving greater cognitive task demand (Belojević et al., 1992), we did not know a priori what level of this task would prove to be sufficiently demanding to show such effects. Behavioral measures of task performance showed no effects of noise type, possibly because the task imposed relatively high cognitive demand, meaning that performance was probably dominated by task-determined differences (i.e., N-back level). Furthermore, the background noises used were not particularly noxious. Thus, it is possible that for noise to negatively affect performance, a more annoying or disturbing sound would be required. Nevertheless, there was a significant interaction between performance and noise sensitivity, such that individuals who rated themselves as more sensitive to noise tended to perform more poorly in the 1- and 2-back conditions than did those with lower sensitivity. From the current results, we can draw four major suggestions for future research on the interaction between noise sensitivity and cognitive task performance. First, the research should specifically investigate the possibility that noise sensitivity scores reflect aspects of personality traits that may be linked to differences in working memory task performance. Second, a greater variety of tasks beyond working memory tasks should be employed to determine whether this relationship holds across varieties of cognitive demand, or whether it is constrained to working memory tasks or N-back tasks in particular. Third, parametrically varying levels of cognitive demand should be incorporated with the goal of identifying ranges of moderate-to-high demand (but not so high that participants start to give up). Finally, efforts should be made to employ a larger number of participants with a much wider range of noise sensitivity values.
C. TLX ratings
The current study design was such that NASA TLX scores were collected after each noise condition, not after each level of task. Thus, interpretation of results must include the consideration that reported NASA TLX scores here include an amalgamation of all three levels of the cognitive task. As such, they likely more so reflect a subjective attitude toward working in the noise rather than the work itself. Future research should examine the differential effect of noise exposure on subjective ratings of workload for distinct levels of cognitive work, which would require workload ratings to be collected after each level of task (McGarrigle et al., 2021).
Significant effects of noise type on task load index scores were observed for the subscales effort, frustration, and temporal demand. For all three of these self-reported task load factors (effort, frustration, temporal demand), noise was always rated as more burdensome than silence. In the case of frustration, both C1 and C2 were rated as more burdensome than silence. But, for effort the more burdensome noise was C1 and for Temporal Demand it was C2. Recall that C2 was both sharper (3.24 vs 1.54), more tonal (0.39 vs 0.17), and was rated as overall more annoying than C1 (Sung et al., 2018 participants rating 6.80 vs 4.81). On the other hand, C1 was rougher (5.44 vs 2.51) and had a greater fluctuation strength (0.012 vs 0.008). Because these noises differed from one another in terms of multiple acoustic properties, it is not possible to draw strong conclusions about which properties might lead listeners to feel differently about a task when performing it in a noise, though sharpness and tonality have both been linked to annoyance (Lee et al., 2017). Future research should experimentally vary distinct acoustic properties of noise in a parametric way in order to identify characteristics of noise most associated with aversive responses.
Also of note is the range of scores in each of the rating subscales, which was large (from 10 to the maximum 19), especially for the subscales of mental demand, frustration, temporal demand, and performance. This pattern illustrates the individual variability in the current sample in terms of subjective attitudes toward the demands of the task, underscoring the need for work in the field of occupational noise exposure to consider the role individual differences play in the effect of noise on employees. It is worth mentioning, however, that the large range in reported performance ratings could be due to an inexplicable reversal of axes in the questionnaire—for all subscales except performance, the scales proceed left to right from “very low” to “very high,” but for performance the scale moves from “perfect” to “failure.” Should the participant fail to read the scales carefully, this reversal could cause them to mark a rating reflecting “perfect” performance when in fact a low performance score was intended. We note this as a drawback of the current NASA TLX form, and consideration should be given to adjusting the form accordingly.
D. Physiological responses
1. Effects of N-back level on physiological responses
We presented the working memory task in level order (0-, 1-, 2-, then 3-back). While this does introduce a confound—participants may get better at the task even as demand increases from block to block—this was an intentional study design decision. As the intention was to expose participants to annoying noise, which may be stressful, a simple study design that minimized the burden of instruction on participants was ideal. Further, the goal of this research was to investigate the influence of noise (not working memory task difficulty) on physiological and psychological markers of stress. Should future research wish to focus on the effects of cognitive demand while working in noise, we would recommend counterbalancing working memory task level as well as noise condition. Though we do see a main effect of N-back level on working memory task performance as expected, we cannot rule out an influence of order on physiological differences across N-back blocks due to either fatigue or experience.
The manipulation of cognitive demand in the N-back task generally had the intended physiological effect, such that listeners generally showed evidence of lower physiological arousal in the 1-back condition as compared to the 2-back, and the 2-back compared to the 3-back. This pattern was most obvious in the heart period and RSA data, such that heart period was shorter and RSA was lower (indicating greater autonomic arousal and cognitive engagement, respectively) in the 2- and 3-back conditions than in the 1-back. Corrugator supercilia activity recorded via EMG also showed a similar pattern, in which the 1-back condition showed significantly less activity than the 3-back condition with the 2-back between the two but not significantly different from either. The two exceptions to this overall trend are found in the electrodermal response data and could be related to the structure of the experiment. The first is SCL, where no effect of N-back level was shown. SCL is a tonic measure of the electrical characteristics of the skin over time and does therefore not change quickly in an event-related manner. Therefore, it is possible that SCL here is simply reflecting the fact that the participant was in noise for a time (∼15 min), rather than reflecting task-specific demands. The second exception is SCR rate, for which both the 1- and 2-back conditions showed a higher rate of SCR than in the 3-back condition. In this case, it is possible that participants were “giving up” more in the difficult 3-back condition, therefore making fewer responses to target stimuli (resulting in fewer event-related SCRs) in the 3-back compared to the 1- and 2-back conditions.
2. Effects of noise type on physiological responses
One of the primary goals of this study was to investigate the possibility that the two types of noise might differentially affect physiological or self-report measures associated with discomfort. This expectation was borne out, at least to some extent. Across measures, participants tended to show greater arousal in the noise as compared to the silence conditions. This was most obvious in the heart period measure, which was significantly lower in both noise conditions than in the silence condition, indicating that listeners' heart period decreased (heart rate increased) more compared baseline in the presence of noise than in silence. The other physiological measure showing a main effect of noise type, RSA, suggests a more mixed pattern of responsiveness. Working in noise C1 induced a greater decrease in RSA, typically associated with increased cognitive demand or possibly increased attentional focus (Muth et al., 2012; Overbeek et al., 2014), than did either noise C2 or silence. However, the widest range of effects differentiating the two noise types was found in the interaction between noise type and noise sensitivity, with the greatest differences between noise and silence conditions being observed in less noise sensitive individuals. These significant effects must be interpreted with caution, however, as it is not yet obvious what degree of difference in these measures might be meaningful in a physiological sense (cf. discussion of “smallest effect size of interest” by Anvari and Lakens, 2019).
There are two important considerations in the interpretation of the preceding results. First, because the N-back task was always presented in the same order (1-back, then 2-back, then 3-back), it is possible that some effects of time in noise might be confounded with effects of working memory demand. Second, in the current study, participants were given the opportunity to read task directions in between blocks at their own pace, meaning that the amount of time each participant spent in noise varied slightly depending on reading speed. We recommend that future studies look specifically at effect of time on task in a more controlled manner.
3. Effects of noise sensitivity on physiological responses
Although there was no main effect of noise sensitivity observed in the present dataset, noise sensitivity did interact with noise type in multiple measures. This interaction is reflected in terms of differences in slope of the relationship between noise sensitivity and a given physiological measure (SCR rate, RSA, BVPA, and EMG) in different noise conditions, as shown in Figs. 3, 5(b), 6, and 7.
For SCRR, both more noise-sensitive and less noise-sensitive people show an increase in SCRR vs baseline, but that is stronger for less noise-sensitive people. Less noise-sensitive people showed a greater increase in SCRR doing work in all 3 noise conditions compared to baseline. The slopes of the relationship between noise sensitivity score and SCRR are significantly steeper for noise C2 than for noise C1.
Similarly, both more noise-sensitive and less noise-sensitive people showed a decrease in RSA vs baseline, but that decrease is stronger for less noise-sensitive people. The slope of the effect of noise sensitivity on RSA is steeper for noise C1 than for silence, such that individuals with lower self-rated sensitivity to noise show more reduction in RSA during exposure to noise C1 than in silence.
With respect to BVPA, significant differences are observed in slope between the silence and both noise conditions. For less noise-sensitive people, silence shows an increase in BVPA and both noises C1 and C2 show a decrease. This relationship is more complex for more noise-sensitive people, who show little change from baseline in noise but a decrease in BVPA in silence. One issue with interpreting BVPA in the present study is that it may turn out to be more effectively used in an event-related manner (cf. Francis et al., 2016), rather than measured over a longer period as we have done here. Although some early studies suggested that measures related to pulse volume might reflect a tonic psychophysiological response to noise during physical work (Kryter and Poza, 1980), and typical cognitive effects on pulse volume may be obtained over spans of a minute or more (e.g., Iani et al., 2004), other studies have found no effect of longer periods of noise on BVPA (Francis et al., 2021). As discussed by Francis and Oliver (2018), such variability across studies suggests the need for more detailed studies of the time-course of change in peripheral vasoconstriction in response to noise.
For EMG, both less noise-sensitive people and more noise-sensitive people showed an increase in EMG activity during the working memory task compared to baseline. The effects of noise C1 on EMG activity varied more as a function of noise sensitivity than did the effects of either silence or noise C2—more noise-sensitive people exhibited a larger increase from baseline in EMG activity compared to less noise-sensitive people during noise C1.
Though it is not possible to draw strong conclusions based on these observed interactions, they seem to suggest that individuals who were more noise-sensitive showed seemingly weaker changes from baseline. One reason for this could be that the study was advertised as involving doing work in noise and the NoiSeQ was administered prior to starting the behavioral task. As a result, participants may have approached the study with a greater degree of arousal to begin with and/or might have needed more than 3 min of relaxation time to stabilize physiologically, in a manner analogous to the effect of stereotype threat discussed by Ryan and Campbell (2021). This could also be related to study design and properties of the assessment chosen to score noise sensitivity. The NoiSeQ is designed to score individuals on noise sensitivity in a variety of contexts: work, home, leisure, communication, and sleep. It seems likely that the NoiSeQ might be measuring something more akin to sensitivity to environmental noise (i.e., living close to an airport or a wind turbine), and therefore might not be predictive of how a person responds to a specific noise event like the ∼15-min exposure to HVAC noise used here.
V. CONCLUSION
The goal of this work was to explore the interaction between acoustic properties of realistic annoying noise and individual differences in response to working in that noise. We examined affective physiological responses in individuals working in silence and in two acoustically distinct HVAC equipment background noises. Our results showed that these distinct noises affected physiological arousal in different ways, possibly due to differences in the acoustic properties of sharpness, tonality, roughness, and fluctuation strength. However, further research is necessary to definitively link specific noise types/qualities and physiological response in humans.
ACKNOWLEDGMENTS
We thank Patricia Davies for helpful discussion related to the initiation and design of this study. We also thank current and past SPACE Lab members Anna Brown, Breeah Carey, Shelby Claflin, Jessica Lorenz, and Theresa Nelson for their help in subject recruitment and data collection and analysis.
See supplementary material at https://www.scitation.org/doi/suppl/10.1121/10.0006383 for detailed descriptions of physiological data collection and preparation methods, modeling technique, and summary data of physiological measure log ratio scores.