Abstract
usic has been shown to be capable of improving runners' performance in treadmill and laboratory-based experiments. This paper evaluates a generative music system, namely HEARTBEATS, designed to create biosignal synchronous music in real-time according to an individual athlete's heartrate or cadence (steps per minute). The tempo, melody, and timbral features of the generated music are modulated according to biosensor input from each runner using a combination of PPG (Photoplethysmography) and GPS (Global Positioning System) from a wearable sensor, synchronized via Bluetooth. We compare the relative performance of athletes listening to music with heartrate and cadence synchronous tempos, across a randomized trial (N=54) on a trail course with 76 ft of elevation. Participants were instructed to continue until their self-reported perceived e ort went beyond an 18 using the Borg rating of perceived exertion. We found that cadence-synchronous music improved performance and decreased perceived e ort in male runners. For female runners, cadence synchronous music improved performance but it was heartrate synchronous music which signi cantly reduced perceived e ort and allowed them to run the longest of all groups tested. This work has implications for the future design and implementation of novel portable music systems and in music-assisted coaching. Keywords: algorithmic composition; biosynchronous music generation; running; physical activity; music mediated perceived e ort; music perception 1. Introduction Trail running is an outdoor sport with a long history which has recently become the focus of academic investigation [13], particularly as the emerging eld of biophilia [4] suggests that there are potentially physical and mental
music systems and in music-assisted coaching. Keywords: algorithmic composition; biosynchronous music generation; running; physical activity; music mediated perceived e ort; music perception 1. Introduction Trail running is an outdoor sport with a long history which has recently become the focus of academic investigation [13], particularly as the emerging eld of biophilia [4] suggests that there are potentially physical and mental health bene ts to engagement with the natural world. Music has been suggested as both a performance enhancer, and a mediator of psychological and physiological discomfort whilst engaging in sport [5,6]. Music can be used to increase motivation, or as a tool to help an athlete achieve a desirable mental state before partaking in sport, due to the ability of music to in uence emotional states. For a full treatment of the use of music in enhancing athletic performance, the interested reader is referred to a relatively recent survey in [7]. The use of music for improving athletic performance as measured by physically quanti ed improved performance and mental performance improvement markers such as reduced perceived di culty, have been the subject of a signi cant body of research, including ndings that: Sensors2020,20, 4528; doi:10.3390 /s20164528 /journal/sensors
Sensors2020,20, 4528 2 of 14 Music can mediate physical responses to pain [8]. Musical cues, particularly the soni cation (Soni cation is a practice of auditory display whereby data are auralized. A simple example would be an alarm) of biomedical data, can encourage good form in strength and conditioning activities [9]. Music can improve athletic performance, and reduce the perceived e ort involved [10,11]. Appropriate music selection is not trivial [12,13]. Previous work has evaluated the role that music selection might play in mediating heartrate and perceived exertion in running for 20 min periods [11], in a treadmill condition [14], and in more recent studies investigating a ectively-driven music selection on the basis of biosensing [15], whereby music selection is made according to a target heartrate. Music selection has been shown to enhance running performance, by encouraging optimal running cadence by means of music [16]. The selection of music is critical to informing the research documented in this paper. Systems for automatic music selection and curation now exist [17], but our particular focus is on harnessing the power of algorithmic music generation [18], which provides techniques that can power systems to create new music with no predetermined time-frame, that might operate synchronously with an athlete's own biosignals. When listening to our favourite music, our bodies respond physically, inducing reactions such as pupil dilation, increased heart-rate, blood pressure, and skin conductivity [19,20]. These biomarkers are particularly interesting as existing work has shown that there is a neurological and physiological connection between emotional state, human performance, and music listening [21], which recent advances in wearable sensor technology and portable programmable music systems might now potentially address. A number of suggestions for optimal running cadence have been made (where cadence, or gait, is the number of full cycles per minute, i.e., by both feet) but these values are highly variable, dependent on individual stride length, technique, and other bio-mechanical factors [22]. Nevertheless, cadence selection has been shown to reduce running-related injuries and reduction of time taken to fatigue [23,24]. Treadmill based experiments do not necessarily translate well to real-world running conditions outdoors, and
the number of full cycles per minute, i.e., by both feet) but these values are highly variable, dependent on individual stride length, technique, and other bio-mechanical factors [22]. Nevertheless, cadence selection has been shown to reduce running-related injuries and reduction of time taken to fatigue [23,24]. Treadmill based experiments do not necessarily translate well to real-world running conditions outdoors, and as such we will focus on the design of an experiment considering the use of music in an ecological context-outdoor running. The prototype system we have designed, described in Section, is the subject of an experimental analysis in this paper. We believe that, in future, these types of system also have the potential to be adapted to real-time biosynchronous feedback (for example, audio-based coaching), a use case which music and trail running are particularly well suited to from a user safety perspective because of the lack of vehicle tra c or the need for visual interfacing. 1.1. Synchronous and Asynchronous Music as A ective Correlates and Athletic Performance Enhancers Speci c choices of asynchronous classical music have been shown to reduce the heart rate and perceived exertion (amongst other physiological measures) of athlete's during running [25], wherein participants exhibited decreased heart rate and blood pressure in the music condition. Some more recent work suggests that asynchronous music was less motivating than synchronous music for treadmill running [26]. Beyond running speci c evaluations, asynchronous music was suggested to have a greater degree of in uence on valence (valence is often used as a means of quantifying the positivity of a particular emotional state) [27,28]. Some research suggests that synchronous music has a greater in uence on mood [29], promotes greater endurance [30], and may further reduce limbed discomfort and increase arousal level (arousal as an a ective correlate is a means of quantifying the activation strength of a particular emotional state) [31]. There is also increasing evidence of a correlation between athletic endurance and levels of valence [32]. In the existing literature, three speci c functions of music have been widely reported as having a signi cant in uence on athletic
level (arousal as an a ective correlate is a means of quantifying the activation strength of a particular emotional state) [31]. There is also increasing evidence of a correlation between athletic endurance and levels of valence [32]. In the existing literature, three speci c functions of music have been widely reported as having a signi cant in uence on athletic performance: Auditorymotor entrainment, i.e., the tendency of the listener to synchronise their own movements with musical features.
Sensors2020,20, 4528 3 of 14 Dissociation: the ability to distract an athlete from discomfort by means of listening to music. The ability of music to increase emotional arousal (where arousal is synonymous with emotional intensity or activation strength) Music has been shown to be able to narrow an athlete's focus and attention whilst exercising, diverting them from unpleasant sensations such as fatigue [31,33], and from other distractions like environmental cues (for example in the case of trail running, bad weather might be of particular concern as opposed to treadmill or laboratory based experiments). There is a psychological parallel with the sensation of total immersion or ow [34], wherein the athlete is so involved in the activity, i.e., listening or exercise, that other such distractions are pushed out of their cognitive processing. These distractions might include, in the case of trail running, thoughts about competitors, or a focus on the runner's own form and technique, which would ideally become autonomic processes. Music can thus be used to reinforce skill development in the athlete [5]. However, there could be cases where this development might be hindered by music, for example when music becomes a distraction in its own right. Again, such ndings reinforce the case for careful curation of music selections to maximize athletic performance. A range of musical features have been proposed as correlates for emotional and perceptual states in listeners, including rhythmic responses whilst exercising [31]. Perhaps unsurprisingly, musical rhythm has been found to be in uential in many cases, both by encouraging motor responses which are synchronous with the temporal characteristics of music (such as when people dance to music), and also with asynchronous e ects [27]. Tempo, for example, can be used synchronously by matching the speed of a piece of music to an athlete's heartrate [35,36], or asynchronously by switching between slow and fast tempo in order to try and heighten the athlete's motivation and increase work output, with marked e ects shown across a range of sports [2527], including rowing [37] volleyball [38] and cycling [39]. Musical features which have a strong correlation with induced or
a piece of music to an athlete's heartrate [35,36], or asynchronously by switching between slow and fast tempo in order to try and heighten the athlete's motivation and increase work output, with marked e ects shown across a range of sports [2527], including rowing [37] volleyball [38] and cycling [39]. Musical features which have a strong correlation with induced or perceived arousal include tempo (higher tempi are often correlated with higher arousal), mode (where minor keys are often associated with lower valence, and major keys with higher valence, or positive states), and rhythmic density (with less dense patterns being associated with decreased arousal) [40]. Sporting activities, particularly those with an element of competition, are very likely to cause an increase in arousal for participants and spectators alike. Music can be used to stimulate or attenuate arousal (e.g., to calm an athlete down before undertaking competitive activity), or to increase arousal before or during the activity itself [41]. This can be imagined as a Yerkes-Dodson curve wherein stimulating or attenuating e ects of music operate on an inverted `U' shape in relation to performance, rather than as a linear function. Stimulation/music induced arousal can be a positive e ect in relation to performance, but only up to a point. Note, this is not the same as saying music can become a distraction, this is a separable attention-related e ect. In brief terms, when performance is suboptimal it is possible that music may have led to over or under stimulation. Musical amplitude and tempo have both been previously evaluated in the context of improvement in running performance [10], whereby louder, faster music resulted in a quicker running pace. Previous research investigating the e ect of synchronous music on movement found a marked ergogenic in uence on running shorter distances [42], with a signi cant increase in work output (faster sprint times) and endurance capacity [31]. Whilst we do not draw any speci c hypothesis from the literature regarding the in uence of music on running performance across gender, we designed our experiment to also capture this information and to examine whether there
ergogenic in uence on running shorter distances [42], with a signi cant increase in work output (faster sprint times) and endurance capacity [31]. Whilst we do not draw any speci c hypothesis from the literature regarding the in uence of music on running performance across gender, we designed our experiment to also capture this information and to examine whether there were any signi cant gender di erences in our participants. Speci cally, we examine gender di erences in a range of synchronous and asynchronous music conditions as part of our experimental analysis in Section. 1.2. Overview of Algorithmic Composition A huge variety of music generation systems exist, with perhaps the earliest example being Mozart's Musikalisches Würfelspiel, the `dice game', which uses the roll of a dice as a control signal to
Sensors2020,20, 4528 4 of 14 inform the selection of pre-composed musical segments. Music generation by algorithmic means is not novel. Algorithmic composition systems include transformative systems, wherein source musical passages are adjusted according to various functions (transposition, retrograde inversions, etc.), and purely generative systems, whereby rulesets and constraints are used to modulate otherwise randomly generated source material. The latter approach is particularly well suited to data-driven approaches. For the purposes of the work described here, the use of algorithmic composition over selection of existing musical stimuli has the advantage over `traditional' music selection in that we can create continuous music playback (hence no pauses between song selections whilst our participants are running), and most crucially the ability to be precisely synchronous with an individual's given heartrate or running cadence as measured using a lightweight wearable biosensor. Our work is closest in this regard to recent work using biosensors to control music based on emotional estimation [43], and the interested reader might want also see related evaluation of sensor type and mappings with speci c musical examples (singing) [44], or in a broader range of work in the recent review of musical interfaces, which includes biosensor interfaces, given by Frid [45]. 2. Materials and Methods The system described and evaluated here, namely HEARTBEATS, was developed to track a runner's cadence and/or heartbeat to generate synchronous music which might promote optimal running performance in the listener. It makes use of generative music production techniques, as described in [46], to create new music according to a second order Markov model specifying rhythmic and melodic musical feature generation in real-time with varying degrees of repetition. This is of note because previous work has noted how repetition can create a range of emotional responses in listeners [47]. We are particularly interested in the use of music with athletes in an ecological context (i.e., real-world outdoor environment), and the potential of biophysiologically synchronous, computer-generated music to aid performance. Cadence had been documented in the literature as a prominent feature, but matching optimal cadence to running (between 150180 steps per minute) which the reader should note would
listeners [47]. We are particularly interested in the use of music with athletes in an ecological context (i.e., real-world outdoor environment), and the potential of biophysiologically synchronous, computer-generated music to aid performance. Cadence had been documented in the literature as a prominent feature, but matching optimal cadence to running (between 150180 steps per minute) which the reader should note would be further challenging in a system using traditional music selection. There are simply not vast quantities of existing music at 180 bpm. In HEARTBEATS, music is generated at a synchronous pace for the athlete based on either their heartrate or cadence whilst running. Latency is minimal but transformations occur at bar breaks, which gives the impression of being instantaneous at all but very slow tempo values. We therefore designed a system to investigate the relative performance between heartrate and cadence synchronous music by experiment with the following research question: Does cadence or heartrate synchronous algorithmically generated music improve athletic performance or reduce perceived exertion in trail runners? 2.1. Music Generator An overview of the music generation system is shown in Figure. The rhythmic, melodic, and timbral features of the generated music are modulated according to a combination of biosensor input for tempo and metrical data, and random number selection of a series of prescribed musical feature mappings based on the features described above. A preloaded selection of source sound les is combined and mixed in various combinations, and can also be resampled, time-stretched, or pitch shifted, according to the desired cadence or heartrate. These e ects have a result in the perception of the music speeding up or slowing down, depending on the desired tempo.
Sensors2020,20, 4528 5 of 14Sensors 2020, 20, x FOR PEER REVIEW 5 of 14 Figure 1. Overview of music generation system. Tempo, melody, and timbral features in the generated music output are modulated according to biosensor input from individual users. Timbre mapping is by 2nd order hidden markov model. The rhythmic, melodic, and timbral features of the generated music are modulated according to a combination of biosensor input for tempo and metrical data, and random number selection of a series of prescribed musical feature mappings based on the features described above. A preloaded selection of source sound files is combined and mixed in various combinations, and can also be resampled, time-stretched, or pitch shifted, according to the desired cadence or heartrate. These effects have a result in the perception of the music speeding up or slowing down, depending on the desired tempo. 2.2. Experiment Design A group of 57 volunteers were recruited from a local (county level) trail running club to participate in an experimental trial of the system on a real-world trail course. These volunteers were recruited via the club social secretary as part of a series of social runs-volunteers were not screened according to ability or experience. The music condition was assigned randomly. Of the 57 trials returned, 29 were carried out with cadence synchronous music, and 28 with heartrate synchronous music, however not all of this data were used in the final analysis (54 out of 57 participants data were used; for further detail, see Section 3). Runners were self-timed on a course segment with 76 ft of elevation including a steep section with a 26% grade, as shown in Figure 2. Participants were instructed to continue until self-reported perceived effort went beyond an 18 on the Borg scale from 6–20, with 6 being no exertion at all, and 20 corresponding to maximal exertion. Participants were given a short training session giving instruction on the use of the scale prior to beginning the experiment. The course segment was chosen to be representative of a variety of trail race conditions, including fell and mountain racing, and trials were
Borg scale from 6–20, with 6 being no exertion at all, and 20 corresponding to maximal exertion. Participants were given a short training session giving instruction on the use of the scale prior to beginning the experiment. The course segment was chosen to be representative of a variety of trail race conditions, including fell and mountain racing, and trials were conducted over a weeklong period in December 2017, with runners submitting their results via a commonly used social networking tool. As such, there is an element of trust involved in any individual reporting their own time and distance honestly, with the expectation being that this was an experiment, not a race, and runners should take part within the framework of an ‘honour’ system. Figure 1. Overview of music generation system. Tempo, melody, and timbral features in the generated music output are modulated according to biosensor input from individual users. Timbre mapping is by 2nd order hidden markov model. 2.2. Experiment Design A group of 57 volunteers were recruited from a local (county level) trail running club to participate in an experimental trial of the system on a real-world trail course. These volunteers were recruited via the club social secretary as part of a series of social runs-volunteers were not screened according to ability or experience. The music condition was assigned randomly. Of the 57 trials returned, 29 were carried out with cadence synchronous music, and 28 with heartrate synchronous music, however not all of this data were used in the nal analysis (54 out of 57 participants data were used; for further detail, see Section). Runners were self-timed on a course segment with 76 ft of elevation including a steep section with a 26% grade, as shown in Figure. Participants were instructed to continue until self-reported perceived e ort went beyond an 18 on the Borg scale from 620, with 6 being no exertion at all, and 20 corresponding to maximal exertion. Participants were given a short training session giving instruction on the use of the scale prior to beginning the experiment. The course segment was chosen to be representative
Description
The study investigates the impact of cadence and heartrate synchronous music on trail running performance.