**1. Introduction**

Understanding and recognizing human emotions has been identified as a main interest area in smart systems [1–5]. Such systems are being applied in many fields like well-being and healthcare [6–11], safe driving [12], smart cities [13] and smart environments [14,15], among others. Pleasure, arousal and dominance are three independent emotional dimensions to describe people's state of feeling [16,17]. Arousal was conceived as a mental activity describing the state of feeling along a single dimension ranging from sleep to frantic excitement and linked to adjectives such as stimulated–relaxed, excited–calm and wide awake–sleepy to define arousal [18].

The arousal level changes constantly, and it has a profound influence on performance during everyday activities [19]. Fluctuations in arousal are regulated by the autonomic nervous system, which is mainly controlled by the balanced activity of the parasympathetic and sympathetic systems [20]. Electrodermal activity (EDA; or skin conductance) has also frequently been used as a measure of arousal. The advantage of EDA is that it is unambiguous, given that it is innervated entirely by the sympathetic nervous system (SNS) [21]. Within the domain of music emotion research, physiological measures such as EDA, heart rate, respiration, and body temperature have been frequently used as correlates of emotional arousal. Among these, EDA is generally a preferred measure as it is highly sensitive and under strict control of the sympathetic nervous system and is therefore largely involuntary. Furthermore, a relationship between EDA as indicator of emotional arousal and experienced pleasure in response to music has previously been demonstrated [22]. At the same time, in their studies with volunteers the participants' feelings have been obtained by questionnaires in the form of Likert scales, self-assessment manikins (SAM) and free text [23–26].

This paper introduces arousal detection from EDA signals using musical stimuli. Several studies have reported that using music to elicit emotions is one of the most effective methods of emotion induction [27–30]. Music plays a key role in most people's lives, frequently being used to explore and regulate emotions. The proposal is linked to our current research elicitation of emotions in elderly people to trigger processes of emotional self-regulation [31–33]. Those processes should help elderly people to improve their mood and mental state. The importance of emotional self-regulation is related to the fact that older people, especially when living alone, are at high risk of suffering from diseases such as depression and anxiety [34,35].

Specifically, people over 60 years old from the region of Murcia, Spain, were recruited as participants to listen to a series of musical pieces similar to those played in their younger years in order to study the level of arousal produced by each musical genre. Although many protocols have investigated physiological responses to music, the present work explores the physiological responses to pieces of music composed specifically for this experiment. The use of original pieces of music, which had not been heard by the listener before, is a novel research technique that has yielded interesting results so far [27–30]. The use of this type of music fragments provides a high level of experimental control and allows knowledge of the influence of the independent variables on the dependent ones. Experimental control is especially important when analyzing physiological responses like EDA [36]. The signals collected during the experiment were used in conjunction with a SAM questionnaire to undergo a couple of studies oriented towards discriminating the arousal. One study analyzed some EDA features only, and the second, based on classifiers, examined possible correlations between the objective detection of the arousal level from processed physiological EDA signals and the level of arousal subjectively perceived by participants when answering the SAM questionnaire.

The remainder of the paper is as follows. Section 2 shows the materials and methods needed to perform the experiment successfully, as well as the investigation methods and metrics used. In Section 3, the results obtained are shown and a discussion about the results obtained in the context of the experiment is provided. Finally, in Section 4 the results obtained in this study are presented.

#### **2. Materials and Methods**

This section describes the methodology and materials required to carry out the proposed experiment. First, an introduction is made about the electrodermal activity as a biomarker of activation detection. Then, a description of the material used, and the processes required to detect the level of activation is made. Next, the methods of data collection (SAM questionnaires) and how they are used within the experiment are explained. Afterwards, a detailed explanation of the experiment is given. Finally, the process of data segmentation and feature extraction for further analysis is explained.

#### *2.1. Electrodermal Activity*

Electrodermal activity (EDA) reflects the output of the attentional and affective and motivational processes integrated within the central nervous system that act on the body [37]. When emotional arousal increases, the accompanying activation of the SNS results in increased sweat gland activity and skin conductance. The validity of EDA as a measure of emotional arousal has been established in studies showing that EDA varies linearly with self-reported arousal when viewing emotional pictures [38]. Therefore, EDA is outstanding in behavioral medicine as a biomarker of individual characteristics of emotional response. EDA monitoring has been used for multiple applications, including assessment of anxiety and stress, detection of orientation response, providing neurofeedback for epilepsy, recognition of emotional state, and many others. In addition, EDA can be very effective in discriminating patients with depression from healthy controls [39]. Specific patterns of electrodermal hypoactivity may be a reliable marker of a depressive state at population level, but they should be carefully combined with other physiological and non-physiological indicators when used for preventive and diagnostic purposes.

EDA covers the electrical variations that occur on the surface of the skin due to changes in sweat secretion. EDA signals are obtained by measuring the potential when a small constant current is applied between two metal electrodes (for example, chrome-silver electrodes). The skin usually responds to stress by producing an increase in sweat. Consequently, the skin's conductivity increases. On the other hand, sweat production stops and skin conductivity is reduced when a person is subjected to a calm or neutral induction. In this study, EDA is measured at the wrist, bearing in mind that wrist biosensors are being widely adopted in conventional and commercial devices. The bracelets provide excellent surfaces for attaching the electrodes to the skin. Ideally, the proposed system should be further miniaturized to record EDA in the areas of the palm where the activity of the skin conduction response (SCR) is most pronounced, without being intrusive or interfering with daily activities.

#### *2.2. Data Acquisition and Empatica E4 Device*

The commercial Empatica E4 wristband has been used to carry out our experiment. The Empatica E4 bracelet is a device that allows the collection and measurement of physiological signals such as EDA, blood volume pressure, temperature and acceleration. This device has been used with good results in some previous works [14,40,41]. In this work, we have used only the EDA signals to study the possibility of determining whether significant differences occur when a participant is subjected to different musical stimuli.

An essential component of our proposal is to acquire, process and obtain a set of data that will be used for identification of the listener's arousal. The Empatica E4 device must be firmly attached to the wrist so that the electrodes touch the skin correctly. Otherwise, if the device is not properly connected, the captured data are not valid due to manifold artefacts.
