Designing a Computer Model of Drumming: The Biomechanics of Percussive Performance

ISSN: 1795-6889 www.humantechnology.jyu.fi Volume 13(1), May 2017,109–141 109 DESIGNING A COMPUTER MODEL OF DRUMMING: THE BIOMECHANICS OF PERCUSSIVE PERFORMANCE Abstract: Becoming a competent musician requires significant practice, including rehearsal of various musical pieces. Complex sequences of musical notes and the associated bodily movements must be choreographed and memorized so that the human body can reproduce these sequences consistently. Such bodily movement occurs within the instrumental performance space, with some instruments, notably the drum set, requiring more bodily movement than most. Choreographed bodily movement in drumming is fundamental for producing the timbral and timing variations crucial in delineating human vs. computer percussive performance. Current computer models designed to simulate percussive performance focus on the cognitive aspects of performance or the musical structure to determine the simulation, while other systems focus on reproducing the physics of musical instruments. The focus of this paper is on the complexities of human movement in drumming, with a view toward proposing, as part of a larger research project, a background understanding and methodology for extracting empirical data from human performance for interactive computer-based percussive performance modeling applications. Keywords: percussion, performance, modeling, drums, biomechanics, computer music. © 2017 John R. Taylor and the Open Science Centre, University of Jyväskylä DOI: http://dx.doi.org/10.17011/ht/urn.201705272520 This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. John R. Taylor MARCS Institute Western Sydney University Australia and Sydney Conservatorium of Music University of Sydney Australia Taylor 110 INTRODUCTION Considerable literature exists concerning the field of computational modeling of expressive music performance (Gabrielsson, 1999, 2003). This research includes a diverse array of approaches, owing to the complexity of human performance (Widmer & Goebl, 2004). Arguably, one of the more complex instruments to computationally model is the drum set because of the complexity involved in playing the instrument. In percussive performance, the interaction between the player and instrument is perhaps the most significant variable in timbre production. This interaction is manifested in different techniques, skill levels, musical knowledge and experiences, and the physical attributes of the performers themselves (e.g., height, body mass, fitness, etc.). The act of musical performance encompasses a variety of contributory aspects (Gabrielsson, 1999, 2003; Palmer, 1997), specific examples of which include the physiological (Fujii, Kudo, Ohtsuki, & Oda, 2009; Lee, 2010), cognitive (Dahl & Friberg, 2004; Laukka & Gabrielsson, 2000; Repp, 1999), technical (Dahl, Großbach, & Altenmüller, 2011), and musical (Repp, 1997), as well as both theoretical and empirical perspectives (Shove & Repp, 1995). Several aims of music performance modeling have been identified in the literature, encompassing the design of interactive music performance systems, virtual music environments, and compositional software tools. One aim of performance modeling seeks to generate humanlike computer performance; therefore, it is useful to consider how human performance is distinct from computer performance. In this research, the analysis regarding human and computer performance addresses particularly the context of modeling percussive performance on a ninepiece drum set comprising bass drum, snare drum, hi-hat, floor tom, low tom, medium tom, high tom, and ride and crash cymbals. Such a drum set configuration is typically used in rock, jazz, and pop music genres. Firstly, the empirical research into the physics of percussion instruments shows that a number of physical factors are involved in timbral variation, such as strike location, construction, material, and so forth (Fletcher & Rossing, 1998; Rossing, 2000; Taylor, 2015). Secondly, timbral variations in drum sounds are important to listeners’ overall perception of music (Rath & Wältermann, 2008). Because playing the drums is a time-sensitive endeavor, the human movement involved in percussive performance can be considered to be “chronemic movement” (Sutil, 2015), in which the qualitative determinations of speed, sustain, attack, or delay (Sutil, 2015, pp. 35–37) in musical and timbral qualities are directly related to instrumental interaction and trajectory control. This article presents part of a wider research investigation into the computational simulation of human percussive performance and presents discussion of relevant literature as a prequel for further empirical work. Why Is Modeling Human Performance on a Drum Set So Difficult? Drumming comprises a variety of set drum patterns and techniques that are learned and performed in different rhythmical and musical contexts, often in an improvisatory manner. To perform these drum patterns and techniques, the drummer must choreograph the human movement of the patterns within the biomechanical constraints of his/her abilities in order to execute them within the rhythmic and time constraints of the music. Each drum must be played optimally at all times, with the performer able to add nuances, such as gestural embellishments or timbral variations that could affect either the timbre or the timing, in each strike. Consequently, a drum performance can Designing Computer Models of Drumming Performance 111 be regarded as multiple patterns containing choreographed sequences of human movement. A computer model of drumming therefore encapsulates the choreographed movements contained within a performance and the transitional movements between choreographies. The main aim of this article is to deconstruct and discuss the key aspects of human movement that lead to timing and timbral imperfections in percussive performance on a nine-piece drum set. This analysis enables the identification of a methodology that can be used to analyze human percussive performance with the goal of creating a computer model that represents, musically, the continuum of percussive performance movement in the physical world. More specifically, this analysis identifies ways in which real-world drumming interactions can be captured and how a human might interact with a system that models that interaction. Such a computer model could be used in interactive systems, virtual music environments, and in compositional software tools. This article evaluates methodologies for measuring human movement in order to create a framework that utilizes interactive computer algorithms to simulate percussive performance. This paper begins with a description of drum rudiments and drummers’ development goals that are fundamental to learning optimal movement and form in drumming. The paper then presents an analytical framework, based upon information processing systems and human motor control, with which to understand the underlying causes of performance variation, particularly regarding instrumental interaction and physical control. This will involve a summary review of the literature in the discipline of biomechanics and the subsequent application of these principles in relation to percussive performance of a ninepiece drum set. This article does not address the different options and timbral and acoustical effects of striking implements (see Halmrast, Guettler, Bader, & Godøy, 2010, pp. 204–207), nor is it intended to be an exhaustive discussion. Many specific aspects have been omitted, including the effect of batter head models on timbre (Henzie, 1960; Lewis & Beckford, 2000); the effect of disuniform tension; potential tonal evolution due to the age (and usage) of the head; tempo (Desain & Honing, 1993); feedback conditions (Brandmeyer, Timmers, Sadakata, & Desain, 2011; Dahl & Bresin, 2001; Pfordresher & Palmer, 2002), and temporal independence (Goebl, 2011). In addition, aspects such as style and genre which, with their obvious contextual performance differences, will not be discussed in detail. Drum Rudiments and Development Goals Drummers develop their technique by learning drum rudiments established by the international drum rudiment committee, part of the Percussive Arts Society. The rudiments currently consist of 40 techniques (Percussive Arts Society, 20141) that often are choreographed independently and have been derived from various musical styles to form a pedagogical method for learning percussion. This method is designed to provide an “orderly progression for the development of physical control, coordination, and endurance” (Carson & Wanamaker, 1984, p. 3). Although not explicitly defined, these development goals can be interpreted and summarized as follows:  Physical control, referring to the performers’ management of stick and instrument interaction, which comprises wrist and hand movement and arm control;  Coordination, referring to the strike accuracy and the performer’s ability to exert physical control over sequences of strikes in different locations; and, Taylor 112  Endurance, referring to performer attributes, instrumental configuration, and the complexity of piece being performed. Although these development goals can be considered independent of each other, there is considerable interdependence among the three. One example of this is where a performer has good stick and instrument management but poor coordination. The result is a drummer who could play rhythmic sequences and timbres correctly but not necessarily hit the drum in time. Choreographically, this could be attributed to a disconnect between the sequence design and poor motion control or form. Another example of independent development goals can be observed in a performer’s ability to maintain arm control and coordination during prolonged movements in complex percussive sequences. Control and coordination will deteriorate at varying rates depending on the endurance levels of the performer. Essentially these development goals are individually important to the successful execution of a choreographed movement and contribute towards the overall form of the movement and the sound of the performance. The relationship between the development goals is described in Figure 1. There is no “magic spot” among these development goals because each drum rudiment requires a unique mix of the three components, depending on the percussionist’s current developmental stage and the demands of the choreographical context. Although these goals are fundamental to the development of a percussionist’s skill, obtaining an understanding of percussive performance by way of deconstructing principles of human movement from these goals is difficult due to the effect of environmental factors on skilled movements (Dahl, 2005). Such factors could include, among others, the effect of temperature and altitude on endurance, auditory feedback on coordination, and stick thickness on physical control. As a result, it is both difficult and impractical to account for all these independent variables. Figure 1. A diagram outlining the interdependency among the three development goals in drumming. Individually and collectively, the developmental goals impact performance. Adapted from Carson & Wanamaker (1984). Designing Computer Models of Drumming Performance 113 The reduction of independent variables in the analysis of human movement, extending to environmental variables, is not new. In fact the dimensionality of variables in understanding human movement has been the subject of investigation since Nikolai Bernstein first proposed the theory of the degrees of freedom (DOF) in 1967. He theorized that because there are an almost infinite number of ways a movement could be executed through the large network of muscles, joints, and cells in the human body, there are an infinite number of ways that muscles can achieve the different movements. The control of the nervous system on the musculoskeletal system is highly complex: For any given movement, there are a high number of DOF. This complexity is illustrated during the activation of a single muscular element either in isolation or in any particular sequence (Bernstein, 1967). Thus, if the nervous system controls movement by controlling synergistic groups rather than individual muscles and joints, the number of DOF (and therefore the dimensionality of variables) is reduced (Turvey, 1990). Bernstein (1967) also suggested that sensory feedback from the environment interacts with the nervous system to reduce the number of DOF. Turvey (1990) substantiated the omission of environmental factors within the context of Carson and Wanamaker’s (1984) development goals for this framework by arguing that, “If the environment to which the movement system relates is interpreted as just another large set of variables, then the juxtaposition of an animal and its environment would amplify the problem of degrees of freedom” (Turvey, 1990, p. 940). Juxtaposing environmental factors onto percussive performance would not only concern human movement and the number of DOF but would necessitate extending environmental variables to the vibrational behavior of each of the nine drums under investigation as well. Because the speed of sound increases with air temperature (Fletcher & Rossing, 1998, p. 70), a bigger picture emerges regarding the inherent difficulty in adequately applying several environmental factors as variables across the different themes noted in this article. In light of this, Turvey’s (1990) position will be considered to be the most appropriate view and, consequently, environmental factors will be considered outside the scope of this discussion. THE ANALYSIS OF HUMAN MOVEMENT: A THEORETICAL FRAMEWORK In 1982, neuroscientist David Marr presented a tri-level hypothesis by which information processing systems could be analyzed. These levels of analysis can be summarized as follows (Marr, 1982, p. 25):  Computational level: What does the system do?  Algorithmic/Representational level: How does the system do what it does?  Physical level: How is the system physically realized? Marr (1982) described how these three levels of analysis are not intrinsically dependent upon one another and that, in some circumstances, analysis can be achieved by using only one or two levels. The choice of analytical level is critical in correctly understanding certain systems. More importantly, Marr described how the computational level of analysis is essential in understanding certain phenomena, particularly where there are significant levels of abstraction between the understanding of a system and the computational representation. Examples of this include a priori understanding of the nature of biological or perceptual Taylor 114 processes prior to computational representation, rather than by analyzing the computational representation of such process in a given computational environment (Marr, 1982, p. 27). David Rosenbaum (2010), in his book on motor control, described how Marr’s three analytical levels of information processing systems also represent “the study of human motor control” (p. 4). At the computational level of analysis, Rosenbaum described how, during physical activity, animals and humans plan their movements using what he described as “implicit equations” (p. 5). These implicit equations are derived from Marr’s (1982) computational level, where a system must achieve a function whose representation is often described mathematically. However, for humans and animals, this refers to the mental representation of task to be performed. One example of this is the mental representation a rock climber has of a “dyno” (a jump or leap) to the next position. In the context of percussive performance, this can be a mental representation of an impending drum fill and the drum striking sequence following from the “current position.” Critically, the current position is spatiotemporally unique, thus requiring transitional or linkage movements between sequences. Rosenbaum (2010) noted that the computational level of analysis does not include the execution of the action, which is unsurprising considering the number of DOF. In applying Marr’s (1982) second level, the algorithmic/representational level, Rosenbaum (2010) noted that a computer’s algorithms are designed to enable a system to undertake their functions with guaranteed success. In the natural world, movements operate in real time (analogous to runtime algorithms) without guaranteed success. As examples, a rock climber might not jump high enough to grab the next hold (and thus fall to the safety net below) and the drummer can hit the wrong instrument or strike the shell of the drum by accident. As Rosenbaum pointed out, each of these real time movements relies upon a procedure, and the person executing the action will draw upon behavior and cognition in order to execute and verify the movement, hence Rosenbaum’s extension of this term as the “procedural level” (Rosenbaum, 2010, p. 5). Rosenbaum (2010) described the final level of Marr’s (1982) analysis, the implementation level, as the physical aspects of the movement. These biological elements are described by Rosenbaum (2010) as muscle operation and brain activity (e.g., a rock climber will use leg muscles to jump, stretched arms to grab the hold, and fingers and forearms to grip and maintain the hold). For the drummer playing a snare drum followed by a ride cymbal, muscle operation can include the fingers and hand for gripping the stick, adduction of the lower arm for the strike, followed by a lateral rotation and abduction of the arm to reach cymbal height. Such movement can be considered either a choreographed pattern or a transitional or linkage movement. These examples are highly simplified, as it is in this analytical level that the DOF problem is encountered. Rosenbaum’s (2010) biological adaptation of Marr’s (1982) tri-level analysis provides a solid approach to understanding the movement process. If this three-stage analysis is undertaken in the context of Carson and Wanamaker’s (1984) development goals, it is possible to objectively evaluate existing research and literature on human movement, specifically for percussion. Furthermore, the bottom-up nature of the three analytical levels in relation to performing a drumming action allows for a more comprehensive and structured discussion. This recontextualization is described in Table 1. Understanding the nature of percussive performance variation requires only the computational level of analysis to gain an understanding of the relevance of human performance on timbre and timing and to uncover critical aspects of human movement in physical performance. Although other additional aspects in the other levels contribute to performance variation, this article presents Designing Computer Models of Drumming Performance 115 Table 1. Three Analytical Levels Applied to a Drummer’s Development Goals. Adapted from Carson and Wanamaker (1984) and Rosenbaum (2010). Level/Goal Physical Coordination Endurance Computational Planning the control of  The physical movement  Instrumental interaction Planning coordinated movements  Coordinating simultaneous multiple physical events  Multiple instrumental interactions Planning movement for improving endurance  Economy of movement Procedural The behavioral and cognitive aspects of carrying out a physical movement, relating to  Timbre  Timing The behavioral and cognitive mechanisms for  Measuring current position  Verifying next movement  Anticipating next timbre/timing The behavioral and cognitive aspects of improving endurance  Performance psychology Implementation The physical aspects of carrying out a movement  Muscle activity  Brain function The physical aspects of coordinating multiple instruments  Interlimb coordination  Muscle activity  Brain function Physical ways of improving endurance  Training  Warm up protocols  Performer impairments rationales regarding why the majority of these are outside of the scope of this investigation due to their highly individual and highly subjective natures, as well as the challenges in adequately proving these. The first aspect of the framework outside of the scope of this investigation is the behavioral and cognitive aspects of carrying out a physical movement (physical/procedural). This is because behavior and cognition are highly individual, as well as highly dependent on the context of the performance (e.g., genre). An important cognitive element of this analytical level and context includes sensorimotor synchronization (SMS), which is the rhythmic coordination of an action with a regular external event (Repp, 2005). As a result, the computational representation of SMS would be difficult to realize, and the empirical testing required for such a model is outside the scope of this investigation. For further reading on this subject, consider Fujii et al. (2010), Hove, Keller, and Krumhansl (2007), Repp (2005, 2006), Wing, Church, and Gentner (1989), and Wing and Kristofferson (1973a, 1973b). Another area of the framework outside of investigative scope is the physical aspect of carrying out a movement (physical/implementation), particularly regarding muscle and brain activity. This particular area presents two separate problems. In terms of muscle activity, the most significant modeling challenge lies with the DOF problem and determining which classifiers and representative organizational systems of muscle activation to model. One such solution would be to use a single DOF as a representative for all similar movements in the model. In the case of a drummer, more than one DOF would need to be modeled to cover all Taylor 116 limbs. In addition, determining the most appropriate DOF for the movement, and even the process of making such assumptions, will produce theoretical shortcomings (particularly for neurophysiologists). Modeling muscle activations also presents problems regarding the relationship between abstracted models of muscle movement and timbre production—a problem that also is found in modeling brain activity. Further reading on muscle activation and brain activity during performance is available in Fujii et al. (2009), Fujii and Moritani (2012a, 2012b), Gabrielsson (2003), and Todorov and Jordan (2002). The behavioral and cognitive mechanisms associated with performance feedback (coordination/procedural) encompass a range of methods of feedback acquisition. These include auditory, visual, tactile, haptic, and kinesthetic, and combinations of one or more. Each of these individual types of feedback has different effects on cognitive and behavioral mechanisms and varies depending on the performance conditions. With so many combinations of feedback conditions and environmental variables, finding an appropriate representative model is difficult. Additionally, modeling specific effects of certain feedback conditions would have limited practical application. Therefore, aspects of performance are outside the scope of this research and the reader is directed to Brandmeyer et al. (2011), Dahl and Bresin (2001), Fujii et al. (2010), Gabrielsson (2003), Petrini et al. (2009), Pfordresher and Palmer (2002), and Pfordresher and Benitez (2007). It was noted above that modeling muscle activity was challenging given the DOF problem, the high level of abstraction from timbre production, the timing of both muscle activity and brain function, and the selection of suitable organizational systems for modeling control and muscle activation. This problem is compounded when considering interlimb coordination as a physical aspect of coordinating the strikes of multiple drums (coordination/implementation), particularly in complex tasks such as rhythm production. In creating complex rhythms bimanually, task complexity between the hands (which include cooperative and disjointed tasks) together with the dexterity levels and handedness of the individual will affect the brain’s organizational control of the two hands. In the case of drumming, it is more likely to include leg control for operating the bass drum and hi-hat. This would result in a highly complex study with too many variables to allow for meaningful conclusions relevant to performance modeling. Further reading on this subject, however, is available from Bernstein (1967), Calvin, Huys, and Jirsa (2010), Iannarilli, Vannozzi, Iosa, Pesce, and Capranica (2013), and Kelso, Southard, and Goodman (1979). Endurance is unique to individuals and can be increased with correct training. However, during performance, endurance can be affected by an individual’s level of physical exertion, which can be mitigated by designing sequences of movement that require less movement or that increase their economy of movement. Other behavioral and cognitive aspects of improving levels of endurance fit firmly within the realms of performance psychology, which are difficult to represent in a computational performance model. Similarly, the modeling of training and warm up protocols also is outside of the scope of this investigation in that they do not bring any direct benefit to the modeled system. No benefit would be gained by modeling a performer with an impairment, such as modeling a drummer with low levels of endurance, because the system would be designed with a level of performer obsolescence, resulting in poor playing after a period of time. Therefore computational, procedural, and implementation levels of analysis relating to endurance are outside the scope of this investigation. However, further reading is Designing Computer Models of Drumming Performance 117 available from Abernethy, Hanrahan, Kippers, Mackinnon, and Pandy (2005), Gabrielsson (1999, 2003), and Shaffer (1989). Thus in the following sections, discussion will focus on physical movement, instrumental interaction, and bodily movement in the context of human movement in the physical world. The aim of this research is to identify a method for analyzing the critical elements of music performance movement for electronic representation in either an interactive music or a virtual system. It is worth noting that, although some aspects of the framework are specifically identified as being outside the scope of investigation, there are overlaps between some of the variables mentioned and aspects of performance that will be discussed in the following sections. Their inclusion within the discussion serves to highlight the complexity of percussive performance and demonstrates the wide reaching implications and importance of the discussion. CONTROLLING INSTRUMENTAL INTERACTION Why is physical control so important? Striking an object with another object has two repercussions. Firstly, when the struck object produces sound, vibration in the stick travels through the fingers to the hand. In some instances, and depending on the force of the strike and the materials involved, this can extend into the arm. In severe cases, this can cause discomfort (e.g., using a metal bar to strike a large mass of solid metal with extreme force). Secondly, striking an object can cause the striking tool to be deflected away from the surface and, depending on the elasticity of struck materials, the level of deflection will be either minimal (e.g., a hard metal surface) or more significant (e.g., a membrane under tension). Because playing the drums requires striking many objects consisting of different materials, and striking them at different strengths, the amount of vibration experienced in the player’s body varies among the instruments and which, during drum set performance, is exacerbated by deflections of the striking implement caused by different elasticities in the struck surfaces, the angles of the initial strikes, and the strike forces across the individual components of the drum set. Strike location plays a significant role in modal frequency excitation, subsequently affecting the timbre of the drum. Moreover, because playing the drums often requires multiple strikes, it is important for timbral consistency that the drummer maintains physical control of the striking implement across a diversity of potential strike interactions. Understanding how a performer maintains physical control of a striking implement is important in contextualizing how timing and timbral variations occur in a drumming performance. This information also is useful for developing a performance ontology in which the system either simulates the elements or the results of physical control or transitions into new states as a result of identifying embodiments of physical control as input parameters. This section provides a bottom-up approach to discussing and reviewing the literature concerning instrumental interaction, starting with stick contact and grip, stick rebounds, and preparatory strike movements, to coordinating bodily movement and drum strike trajectories across multiple drums. This approach facilitates a detailed discussion of the complex nature of percussive performance and helps in identifying emergent themes in human percussive performance movement and biomechanics. Taylor 124 The overriding goals of these components draw parallels to Shaffer’s (1989) description of the motor geometry in piano performance: Getting the fingers to the right locations on an instrument is important but only part of the motor task in playing. The performer can learn to shape the trajectories of movement so as to achieve timing of rhythm and variation of dynamic and tone quality with an economy of motor effort. (Shaffer, 1989, p. 383) It is evident from both Dahl (1997b) and Shaffer’s (1989) description of musical performance that drumstick management comprises technical elements of playing the drums, particularly the control of rebounds and the control of stick at the height of the strike motions. Technical elements in drumming contribute toward accuracy in timbre production and timing control. Although Dahl (1997b) described variations in the overall motion among the participants (especially at varying skill levels), the curvilinear trajectory followed the findings by Kelso et al. (1991). Bodily Coordination One important concept of choreography that contributes toward motion and form is that of balance arrangement, particularly whether the body is symmetrical or asymmetrical, which is indicative of stability and equilibrium or irregularity and imbalance. Playing the drums requires both bilateral movement (both limbs moving in unison) and unilateral movement (one limb moving at a time). Although drumming can be considered symmetrical (mirrored) or asymmetrical, depending on the combination of individual drums being played (i.e., the context), the process is inherently asymmetrical owing to the configuration of the components of the drum set. This article discusses the effects of the inherently asymmetric environment and how a drummer responds to and uses asymmetry in designing drumming sequences. Aruin and Latash (1995) investigated the effect of opposing bilateral fast movements on the shoulders (with and without load) of subjects standing on a force platform. They found that anticipatory postural muscle adjustments (APAs) in the trunk and leg muscles were made by the subjects to maintain balance, with adjustments increasing to a maximum when arms were moved in a forward or backward motion and decreasing to no APAs when moving the arm along the sides (i.e., the coronal plane). Furthermore, the authors found no significant difference in muscle adjustment as a result of additional load on the arms. These APAs were evident by changes in the subjects’ anterior, posterior, and vertical centers of pressure and gravity on the force plate prior to the movement. In the case of drumming, it is quite common for the drummer to be in a seated position with much of the player’s weight supported by the seat. Consequently, the leg muscles play a lesser role in redistributing centers of force and gravity for an APA. The redistribution of weight using the legs is further complicated by their use in applying independent pressure to the hi-hat and bass drum pedals. Consequently, upper body stabilization is carried out by the trunk, specifically the erector spinae (ES) and rectus abdominis (RA), irrespective of the types levels of support in the legs (Aruin & Shiratori, 2003). These findings were supported by Santos and Aruin (2008), who also found that the lateral muscles contributed to upright posture control in feed-forward movements (i.e., movements relying on anticipatory correction), akin to feedforward movements in drumming and where the level of muscle activation being is directionally specific. With both legs in a fixed position for operating the hi-hat and bass drum, a drummer’s Designing Computer Models of Drumming Performance 125 directional posture control is of great importance, particularly in controlling movements requiring axial rotation of the upper body. Thus APAs in compound multijoint movements—especially those involving changes in direction (Holmes, 1939, pp. 17–19) such as bilateral fast movements of shoulders coupled with point-to-point axial rotation—are critical in maintaining postural stability. However, in addition to bilateral movements, a drummer’s arm movements often are unilateral, are not directly opposing, and are executed at different strengths and speeds relative to the location and distance between subsequent drums to be struck. Where a drummer has different maximum arm heights relative to the horizontal plane, as well as different maximum distances in arm reach required from the center of the torso between strikes, then postural control and stability also affects movement on the vertical (i.e., sagittal) plane. Thus, consequently, a hunched-over position is not conducive to playing strikes at greater heights. With this in mind, it is easy to imagine the variations in the centers of pressure and gravity on a player during the course of a percussive performance. In fact, Alén (1995) suggested similar links between movement and performance variations. In his analysis of the Cuban music genre tumba francesa, particularly a type of performance called a toque macota, Alén described how the large size of a Cuban bulá drum may have affected the performer’s stabilization, requiring torso movements that could contribute towards timing deviations. Although there are vast differences between the drum set and the bulá, it is conceivable that Alén’s (1995) links also apply to playing the drum set. One theoretical view is that a performer mitigates these effects by maintaining a postural equilibrium, with extreme changes in postural stability countered by APAs stemming from performance planning and musical read-ahead, both of which can be linked to performance skill and having repercussions on musical gesture as a learned deviation.3 In summary, one general rule of drumming performance variation is that the greater the distance and angle of movement (relative to the torso) prior to the strike, the greater the inequality between the opposing reach angle and distance of the other hand, the greater the synchrony/asynchrony of the arm movements, the more complex the biomechanical and neurophysiological process and the increased likelihood of performance variation. Drum Strike Trajectory The trajectory of a drum strike is important in drumming to such an extent that drum strike trajectory was used as an important component in the compositional specification of Karlheinz Stockhausen’s composition Zyklus (1959). As described previously, rebound control can be used to affect the trajectory of the subsequent strike in a sequence of percussive hits. Between rebounds, the player must move the stick from one strike location to another at a speed sufficient for maintaining correct timing. The success of this aim is largely dependent upon trajectory, defined by Abend, Bizzi, and Morasso (1982, p. 331) as “the path taken by the hand as it moves to a new position and the speed of the hand as it moves along the path.” In their study of hand trajectory to target, Abend et al. (1982) found that the majority of subjects who were asked, with no instruction, to move their hand deliberately to a target, opted for a straight line. With the shortest distance between two points being a straight line, one would expect movements with straight trajectories to have a shorter duration than curved trajectories to the same target. Although this was found to be true, movement duration also is dependent on speed, which Abend et al. found to be more irregular during curved trajectories. Taylor 126 However, in cases where the average speed was low, even straight trajectories showed irregular speed patterns, suggesting greater difficulty in controlling the movement. In a performance context, a lower movement speed and, therefore, a lower strike velocity, will produce weaker rebounds. Thus, the interaction with the instrument in terms of rebound control and the movement between the strikes is harder for the player to control. Regarding the irregular speed profiles of the curved trajectories in Abend et al. (1982), it was noted previously that the movement of a drumstick during a strike has curvilinear resemblances due to the phasing of muscle movements (Kelso et al., 1991). However, a connection between the two cannot be drawn because there were differences in planar movement in these studies. The participants in Abend et al. (1982) operated on a horizontal plane, compared to sagittal movements in Kelso et al., (1991) and compared to both sagittal and horizontal movements in Dahl (2000). Despite this, there was a correlation in the increased irregularity in hand speed relative to the antiphase nature of the angular velocity of the shoulder and elbow—in other words, a joint-focused dichotomy with parallels to Kelso et al.’s (1991) muscle synergies. Drumming invariably uses multiple joints, each with different torques applied from the muscles that, in a multijoint movement, extend to the interaction of other joints and torques in the movement. In the case of multijoint movement, each joint will be subject to different velocity interactions at various points in the movement. Where a trajectory is changed midair and not using a rebound (e.g., at a higher preparatory stick height, as in Dahl et al., 2011), the joint torques will change depending on the new trajectory. Such a movement is subject to interactional forces during the planning and control of the movement—such as the Coriolis, centripetal, and reaction torques (Abend et al., 1982, p. 331)—although the effects of these forces change dynamically over the movement. Hollerbach and Flash (1982) observed such behavior in relation to a curved trajectory where “the velocity interaction torques in fact completely dominate the dynamics at the movement midpoint because the inertial torques go through zero as the movement switches from acceleration to deceleration and the arm is moving the fastest at this point” (Hollerbach & Flash, 1982, p. 76). In the case of a single stroke, as measured in Dahl et al. (2011), the midpoint would be the arc at the peak of the preparatory movement. In some instances, a change in trajectory at this point would have three benefits. Firstly, this enables a greater preparatory stroke height for the next strike. Secondly, the greater height enables higher maximum acceleration and downward velocity. Thirdly, as a point with the least amount of inertial torque, the player can prepare for the joint torque of the next movement. Such torque control can mitigate timing variation. In terms of accuracy, it has been found that the trajectory of aimed movement can be learned. These learned trajectory movements were demonstrated by Georgopoulos, Kalaska, and Massey (1981) during a study of aimed movements in Rhesus monkeys. They found that practice over a period of time reduced the mean variability of the trajectory towards a target, together with improved accuracy, irrespective of target location. The implication here is that a human drummer is likely to do the same using the drums as targets. However, as previously noted, drumming requires bilateral and unilateral arm movement, and humans can be either left handed or right handed. Each of these have been demonstrated to be contributing factors towards target accuracy (Garry & Franks, 2000), with increases in reaction time for bilateral strikes with targeting aimed by the weaker hand compared to unilaterally mirrored targeting. Designing Computer Models of Drumming Performance 127 The effects of this can be minimized through drum set configuration, with little impact on multijoint bilateral movement. Although several factors can affect trajectory and control during percussive performance, the most significant factor occurs during multijoint movement, where joint torques impact not only the choice of trajectory but also the control and speed of the movement. In the case of drumming, sequences involving multijoint movements can often include multiple simultaneous planes of motion and axes of rotation. Such an action is illustrated in Figure 4, where a drummer’s movement is described between changes of strike location, from a strike on a snare drum to a strike on a crash cymbal. In the example in Figure 4, during the movement of the right hand from the starting position (snare drum) to the crash cymbal, there is abduction and extension of the right shoulder on the frontal plane with a posterior axis of external rotation. There is also an elbow and wrist extension on the sagittal plane with a lateral axis of rotation. Assuming no movement to the left arm, then there is also a vertical axis of rotation of the trunk on the horizontal plane to allow the drummer to position the body for reaching the new target. Kinetically, each of these axes of rotation and movement in this multijoint sequence contain torque forces that affect the movement. If the drummer in the figure had not included a strike at the crash cymbal but a repeat strike to the snare drum, there would have been minimal changes to the existing patterns of joint torque and muscle activation. Additionally, another drum located at the same height as the snare drum, but closer to the crash cymbal, would cause the drummer to make a trunk and shoulder rotation. However, because the drums are at a similar height, there would be less movement over the three planes. Therefore, movements spanning multiple planes of motion and axes of rotation are most likely to affect the movement of a drummer and, subsequently, the timbre and timing variations. Multijoint movements, such as those in Figure 4, are considerably Figure 4. An illustration showing the typical movements associated with a change of strike location by a drummer moving from a snare drum strike to a crash cymbal strike. Taylor 128 problematic to model because there are 17 DOFs in movements of the shoulder, elbow, and wrist: 9 kinematic net moments and 8 dynamic with optimized muscle forces (Chadwick & van der Helm, 2003, p. 15). This discussion presents some clear difficulties in modeling percussive performance from body, movement, and spatial perspectives. Firstly, stick management plays an important role in the interaction between the stick and the drum in the way that stick contact times can be influenced to alter the vibration of the drum (and the subsequent timbre). Similarly, stick grip influences the rebound of the stick from the drum, which has two effects on drumming: force contact dampening of a drum after a strike and positive and negative rebound use for subsequent strikes, particularly in sequences of drums operating at different angles and locations relative to the torso. Although in most cases stick control can be executed during the strike, due to time constraints much of the rebound and strike control is managed during preparatory movements. During the downward motion of a strike, a curvilinear trajectory was observed in Dahl’s study (1997b); this can be explained by the phasing of muscle activity in the homologous muscle groups of the arm (Kelso et al., 1991). In-phase muscle activity produced greater arm stability and economy of movement, which is a contributory factor in stick control. At the apex of a strike, a fishtail motion was described (Dahl, 1997b), which further exploits the existing synergy between muscle activities by taking advantage of the upstroke to minimize additional muscle activity in the upper arm. In bimanual and unilateral arm movements, which are common occurrences during drumming, APAs were observed as a means to maintain postural stability. These involved small muscle movements that compensated for changes in force (e.g., changes in the center of gravity) resulting from arm extension. The effect of this, when in a seated position, is that the trunk is responsible for postural stability in the upper body. With more complex arm movements in drumming sequences, compared to the simple arm movements as studied in previous research, the potential need for constant postural anticipation and control was highlighted, particularly in arrhythmic unilateral strikes at nonopposing angles and at various distances from the torso. TOWARDS A TEMPORAL MOVEMENT CONTEXT This study aimed to assess the current literature relating to human percussive performance on a nine-piece drum set. This was done in order to understand human movement and rehearsal as choreographed motion. The notion of choreography in the real-world context of drumming, as rehearsed sequences of movements, stems from the internationally recognized drum rudiments intended to develop percussionists’ physical control, coordination, and endurance. In the virtual world context of interactive computer systems, a clear understanding of how these development goals relate to instrumental interaction, biomechanics, and human movement provides an opportunity to explore the interaction possibilities between human and machine. The tri-level Marr/Rosenbaum framework (Table 1) for analyzing information systems and motor control, applied through the lens of the development goals, facilitated a bottom-up analysis of real-world human interaction with drums. The significance of this approach from a human-to-machine interaction perspective is that real-world, bottom-level interactions (i.e., end effector interactions) can have a significant impact on any generated simulation (e.g., force Designing Computer Models of Drumming Performance 129 contact affecting the timbre of a drum by altering the drum’s vibrational characteristics). Such conditions may be either too difficult to mathematically represent or computationally too expensive, particularly across the nine individual components of a drum set. Understanding the nuances of real-world interaction provides for more informed decision making when designing interactive virtual systems and human interaction. Design options include substituting mathematical representations (algorithms or subsystems) of some interactions with similar level representations (algorithms or subsystems) of other interactions. The measurement of such interactions may allow design choices that defer interaction from a dynamic mathematical representation to a sensor-based human interface (e.g., measuring force contact duration on a computer-enabled surface), thus reducing computational overhead while simultaneously maintaining real-world interactive authenticity. The hierarchical nature of a bottom-up analysis allows each real-world interaction to be contextualized within a larger set of movements. In the virtual world, this is equivalent to merging two subsystems to form a larger complex system representative of a more abstract function. From an interaction perspective, this may mean deriving a force contact duration from a series of assumptions about the current state or the context of the system. Such an abstraction could include the representation that stick control is more difficult with weaker rebounds, thus weaker striking leads to longer durations of force contact. However, as each interaction is abstracted to a larger set of movements, human interaction with the system becomes more abstract. Consequently, in designing an interactive system that simulates human percussive performance, there are trade-offs in deferring simulation functions to either horizontally integrated subsystems or abstract layers with regards to the level of similar realworld human interaction with the system. This analytical framework provides a unique way of investigating human percussive performance while concurrently analyzing computational aspects relevant to the system design. The convergence of these two paradigms manifest themselves in system interactivity and how the system represents real-world movements that are inherently both compromised and unique, depending on the vision of the system. Representing Human Movement It is clear that significant issues exist in using the biomechanical considerations of the human body during percussive performance as a method of generating both performance context and in algorithmic control of representative computational musical output. Fundamentally, the main problem in modeling drum set performance is that it is predominantly asymmetrical: The performer’s arm movements (e.g., reach distance, height, and angle) are often unequal, and the rhythmic striking of these can be irregular. The inequality of arm location and irregularity in drumming constantly changes the joint torques and the force interactions that affect trajectory control, movement stability, and postural stability, which subsequently affect strike control, strike accuracy, rebound control, and stick management, ultimately causing variations in timbre and timing. This problem is compounded by an almost infinite number of combinations of movements between Cartesian strike coordinates during drumming and, if one takes into account the DOF problem, there is an extreme abundance of potential system representations. Such an abundance of potential representations would be hard to implement computationally; yet, the selection of a smaller number of representatives is difficult to justify theoretically. As Abend et al. (1982) noted, there would need to be an inverse kinematic transformation of the Cartesian-to- Taylor 130 joint coordinates and then, using inverse dynamics, the joint torques would need to be calculated. This has significant implications for both the selection of particular variables that would form the basis of any computational model and on the way in which these variables are represented. One way of representing human body movement in percussive performance would be to examine a specific movement and identify the most likely used DOF in the joints activated during that movement, such as the methods used by Bouënard, Gibet et al., (2011). This would require investigating the effect that each individual joint can have on the overall movement, including the selection of multiple DOFs and subsequent joint angles on the outcome of the movement. Such an approach would allow for an assessment of whether a variation in DOF at a joint closer to the instrument has a greater impact in producing biomechanical errors than variations in DOF at joints closer to the torso. In addition, preferences and/or trends in planar movement for each joint for a given movement could be identified, together with the impact of these planar preferences on biomechanical error. One way of computationally representing this approach is to design an algorithm that uses weighted probability to calculate the likelihood of a selected DOF or angular movement in a given joint. An example of a method of representing this computationally would be a Gaussian distribution of values to represent a joint angle (e.g., shoulder) and a Markov chain to determine the next selected joint angle (e.g., elbow), and so on until a joint angle value is determined for the wrist. At this point, a movement assigned a unique identification number could be used to trigger a predetermined timbral or temporal variation to represent the level of biomechanical error in the movement (as compared to a theoretical ideal). Determining the probabilities of joint variation within movements would require significant analysis of multijoint, multiplane movement and would require also measuring a quantifiable error from the various movements. Furthermore, deciding which movements to investigate can be problematic in that their relative importance is highly subjective. In addition, identifying preferences or trends in movement at joint level may require significant sample sizes and may generate large quantities of data, particularly should the three axis planar movement be measured at high frequencies. Finally, a link between joint variation and performance variation would need to be quantified and would require multiple methods of analysis, for example, correlating data from joint movement with audio to identify the trends in performance variations associated with combinations joint values. Exactly how the many combinations of joint values represent performance variation also is critical in reproducing human percussive performance in a computer environment, as it relates to a method of controlling one or more aspect of musical parameters, such as timbre or timing. Another potential method lies in the representation of the drumming techniques by creating an algorithm that represents the DOFs associated with a particular drum rudiment. A skilled drummer will have a standard repertoire of drumming techniques at his/her disposal; so it may be possible to assign various combinations of movement to a given technique that then generates a musical output that closely resembles that technique. However, the execution and application of these techniques will differ across performers and performances, notwithstanding the stylistic differences of the performed music. As a result, the process may produce disjointed sounding performances because the selected techniques are inappropriate, unusual, or humanly impossible for the given musical or performance context. Of course, it may be possible to concatenate algorithmic representations of techniques to form a coherent performance, but that depends upon whether the techniques have unique muscle and joint activations that are reproducible and relevant. The most significant challenge in this approach is identifying and empirically measuring Designing Computer Models of Drumming Performance 131 these multiple techniques within a performance and determining a unique movement value. The difficulties associated with collecting joint information as described above are suddenly increased when more variables are introduced into the performance context. It is clear that, in designing a system that simulates human percussive performance, the sheer number of biomechanical and performance considerations pose significant challenges in computationally representing any meaningful or situationally specific interaction. Therefore, it is useful to consider the biomechanical considerations and performance context at a lower level of detail and in the context of playing the drums. So what do we know at this point? One broader view that can be taken for the purposes of modeling percussive performance is that large multijoint movements operating on multiple planes of motion are more likely to generate performance variations for two reasons. Firstly, a percussive performer playing the drum set is inherently constrained by his/her number of limbs regarding how many instruments can be struck simultaneously. For example, a nine-piece drum set has 64 potential combinations of simultaneous instrument strikes using only the two hands. With the feet fixed in position, the main areas of movement lie in the upper body and torso, which relates to the complexity and equality of bimanual drumming. Secondly, because the number of drums limits the combinations of arm movements, the complexity of the movement is largely affected by prior arm location. Collecting Data from Human Movement The literature discussed in this article present various methods for obtaining observational and empirical data from human percussive performance. Despite varying research aims, these studies reveal important insights into real-world human interaction with drums. In order to virtualize this human–drum interaction, various methods can be exploited for system design and control. The task then is to determine whether the system should be event-driven (i.e., software that changes behavior in line with an event), data-driven (i.e., software whose embedded data controls the flow of the program), or a combination of the two. An event-driven system would respond to human interaction, such as playing a typical commercially available electronic drum machine in which electronic drum pads measure the strike force and play a sampled drum sound in response. One key consideration of this approach is to ensure that the interaction between human and machine accurately simulates real-world, stick-to-drum interaction. However, most modern electronic drum pads account for this need. In fact, most modern electronic drum sets have begun to incorporate different zones into the drum pads in order to trigger different timbres, thus mirroring the action of physical drums. Obtaining a measurement of contact duration ubiquitously from the drum pad may be useful for filtering a triggered sound to simulate membrane dampening. Pressure sensors mounted in or below the drum stool (similar to the force platform used by Aruin & Latash, 1995) could be used to control additional timbral parameters, although how the center of mass relates to meaningful system output would depend on system representation. Consequently, such functions would produce limited meaningful additional interaction. A completely eventdriven system such as an electronic drum kit has two drawbacks. Firstly, it relies on the human user to interact with the system and will only produce sounds relative to the skill level of the user. Secondly, with music being a time-sensitive task, system responses from human interaction would need to be extremely low latency. This may not be possible, depending on the speed of the performance and the number of events that need to be handled. Taylor 132 A data-driven system could use data captured from actual human performance in order to create an embedded database or for use in real-time interaction. One method of capturing human performance uses infrared cameras and sensors, as demonstrated by Dahl (2004), Dahl et al. (2011), and Kelso et al. (1991). This approach also was used by Bouënard, Gibet et al. (2008, 2011) to capture timpani performances to create a motion database. Although this approach addressed some limitations in physics-based modeling of performers, Bouënard, Gibet et al. noted that the instrument could obstruct the infrared markers. In relation to a ninepiece drum set, this is a significant limitation in that a typical drum set has components at various points around the performer. In addition, Bouënard, Wanderley, Gibet, and Marandola (2011) described limitations in capturing nuances in performance, such as stick grip. With this in mind, real-time control of a percussive system using infrared cameras and sensors would need to take place in an environment devoid of obstructions because, from a human–machine interaction perspective, a significant portion of the stick-to-drum interaction is lost, such as the rebound and force contact. The descriptions above provide the polar exemplars of event-driven versus data-driven systems, with each having completely distinct aims and outcomes. The event-driven system with the electronic drum kit is typical of a performance system, while the data-driven system with the motion capture is reminiscent of virtual character animation (Bouënard, Gibet et al., 2011). Interactive systems that employ a combination of event-driven and data-driven methods include interactive computer-generated performance tools and electronic composition tools that render a performance based on human interaction. Such systems require a larger amount of abstraction on the data side complemented by human interaction to trigger events and computer state changes. In the case of abstracting data, several methods are available for creating a representation of an aspect of performance. One method of empirical data collection that could help to identify the levels of movement in a drumming performance is the video capture of multiple performances by different performers with the comparative analysis of the movement level in the video across the performers. It also is possible to attach accelerometers to the performer’s hands to measure the amount and direction of their acceleration. In addition, audio data could help to identify the extent of performance variation by enabling a temporal analysis of performance events and comparing these with elements of the video and sensor data. This could yield information regarding the temporal stability of performance. Although this methodology would facilitate a more generic representation of percussive performance, much information can be obtained from this multimethod approach. Firstly, any empirical performance data obtained could be used in a data-driven model. Secondly, it is possible to infer more generic rules surrounding the use of multiple combinations of instruments and avoid the need for generating multiple variables to cater to the DOF problem. Finally, this approach is more practical because broader observations can be made from a relatively fewer number of participants than would be required to calculate the median joint angle averages for multiple percussive techniques. Therefore, by identifying complex bodily movement in an instrumental performance space, algorithmic logic can be created that can simulate the performance context that forms the fundamental logic of a system that controls the levels of variations simulated in a computer model of drumming. This approach supports rethinking the choreography of performance and its role in designing computersimulated human interaction, as well as rethinking empirical movement data to contribute towards new concepts of computer generated rhythm systems. Designing Computer Models of Drumming Performance 133 Sound generation and instrumental representation are important components of any performance modeling system. With a variety of techniques available for system-modeling consideration, the main considerations are computational overhead, expressivity of the synthesis, and accuracy of the representation (Kahrs & Brandenburg, 1998). The complexity of accurately synthesizing a nine-piece drum kit comprising membranophones and idiophones places some physical modeling synthesis techniques firmly out of scope, particularly when considering the computational overhead and time sensitivity of the system. Consequently, sound generation is more efficient when the computational overhead is transferred to decision making, database matching, and sound playback, as opposed to calculating complex equations and resythesizing the sound at run time. Therefore, using a comprehensive sample database to sonically represent the instruments would augment the realism of the simulation by allowing timbral variations to be linked to inferred representations of performance. Towards a Theoretical Model Adopting a physical-based approach presents two levels of conceptual representation of the system. The first relates to David Marr’s (1982) representational level, whereby the relationships between the samples in a database conceptually represent an instrument. The second level further abstracts performance and presents a more contextual understanding of the variables affecting the relationship between the samples by inferring a relationship between the instrumental representations themselves. This is described in Figure 5, a simplified diagram showing the context of two instrumental representations. Holistically, the instrumental representations should be part of a larger conceptual construct related to performance context. With this in mind, a theoretical model is presented in Figure 6 that shows a performance model derived from information in the sample data, augmented by representations of performance context. Figure 5. Intrainstrumental performance context. Pulse-code modulation samples of strikes on a single instrument represent only one occurrence of the performance. Therefore, it is necessary to use the sonic content of each sample to provide a contextual representation of the instrument, by inferring a relationship between the spectral features of each sample. Taylor 140 Halmrast, T., Guettler, K., Bader, R., & Godøy, R. (2010). Gesture and timbre. In R. Godøy & M. Leman (Eds.), Musical gestures: Sound, movement, and meaning (pp. 183–211). New York, NY, USA: Routledge. Henzie, C. (1960). Amplitude and duration characteristics of snare drum tones. Unpublished doctoral dissertation, Indiana University, USA. Hollerbach, J. M., & Flash, T. (1982). Dynamic interactions between limb segments during planar arm movement. Biological Cybernetics, 44, 67–77. Holmes, G. (1939). The cerebellum of man. Brain, 62(1), 1–30. Hove, M., Keller, P., & Krumhansl, C. (2007). Sensorimotor synchronization with chords containing tone-onset asynchronies. Attention, Perception, & Psychophysics, 69(5), 699–708. Iannarilli, F., Vannozzi, G., Iosa, M., Pesce, C., & Capranica, L. (2013). Effects of task complexity on rhythmic reproduction performance in adults. Human Movement Science, 32, 203–213. Kahrs, M., & Brandenburg, K. (1998). Applications of digital signal processing to audio and acoustics. New York, NY, USA: Kluwer Academic Pub. Kelso, J. A., Buchanan, J. J., & Wallace, S. A. (1991). Order parameters for the neural organization of single, multijoint limb movement patterns. Experimental Brain Research, 85, 432–444. Kelso, J. A., Southard, D., & Goodman, D. (1979). On the nature of human interlimb coordination. Science, 203(4384), 1029–1031. Laukka, P., & Gabrielsson, A. (2000). Emotional expression in drumming performance. Psychology of Music, 28(2), 181–189. Lee, S. (2010). Hand biomechanics in skilled pianists playing a scale in thirds. Medical Problems of Performing Artists, 25(4), 167–174. Lewis, R., & Beckford, J. (2000). Measuring tonal characteristics of snare drum batter heads. Percussive Notes, 38(3), 69–71. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. Pacific Grove, CA, USA: Brooks/Cole Publishing Company. Palmer, C. (1997). Music performance. Annual Review of Psychology, 48(1), 115–138. Percussive Arts Society (PAS). (2014). PAS international drum rudiments. Retrieved from http://www.pas.org/resources/education/Rudiments1/RudimentsOnline.aspx Petrini, K., Dahl, S., Rocchesso, D., Waadeland, C. H., Avanzini, F., Puce, A., & Pollick, F. E. (2009). Multisensory integration of drumming actions: Musical expertise affects perceived audiovisual asynchrony. Experimental Brain Research, 198(2-3), 339–352. Pfordresher, P., & Benitez, B. (2007). Temporal coordination between actions and sound during sequence production. Human Movement Science, 26(5), 742–756. Pfordresher, P., & Palmer, C. (2002). Effects of delayed auditory feedback on timing of music performance. Psychological Research, 66(1), 71–79. Rath, M., & Wältermann, M. (2008). Exploring the perceptual relevance of inherent variability of drum sounds. In R. Kronland-Martinet, S. Ystad, & K. Jensen (Eds.), Computer music modeling and retrieval: Sense of sounds (pp. 303–312). New York, NY, USA: Springer. Repp, B. H. (1997). The effect of tempo on pedal timing in piano performance. Psychological Research, 60(3), 164–172. Repp, B. H. (1999). Detecting deviations from metronomic timing in music: Effect of perceptual structure on the mental timekeeper. Perception & Psychophysics, 61(3), 529–548. Repp, B. H. (2005). Sensorimotor synchronization: A review of the tapping literature. Psychonomic Bulletin and Review, 12(6), 969–992. Repp, B. H. (2006). Musical synchronization. In E. Altenmüller, M. Wiesendanger, & J. Kesselring (Eds.), Music, motor control and the brain (pp. 55–76). Oxford, England: Oxford University Press. Designing Computer Models of Drumming Performance 141 Rosenbaum, D. A. (2010). Human motor control (2nd ed.). Burlington, MA, USA: Elsevier. Rossing, T. D. (2000). Science of percussion instruments. Republic of Singapore: World Scientific Publishing Company. Santos, M. J., & Aruin, A. S. (2008). Role of lateral muscles and body orientation in feedforward postural control. Experimental Brain Research, 184(4), 547–559. Shaffer, L. H. (1989). Cognition and affect in musical performance. Contemporary Music Review, 4(1), 381–389. Shove, P., & Repp, B. (1995). Musical motion and performance: Theoretical and empirical perspectives. In J. Rink (Ed.), The practice of performance (pp. 55–83). Cambridge, England: Cambridge University Press. Stockhausen, K. (1959). Nr. 9 Zyklus für einen Schlagzeuger [Cycle for a Percussionist]. London, England: Universal Edition. Sutil, N. S. (2015). Motion and representation: The language of human movement. Cambridge, MA, USA: MIT Press. Taylor, J. R. (2015). Ontology of music performance variation (Doctoral dissertation; Record No. 2123/13279). Sydney, Australia: University of Sydney. Todorov, E., & Jordan, M. I. (2002). Optimal feedback control as a theory of motor coordination. Nature Neuroscience, 5(11), 1226–1235. Turvey, M. T. (1990). Coordination. American Psychologist, 45(8), 938–953. Wagner, A. (2006). Analysis of drumbeats:Interaction between drummer, drumstick and instrument (Master’s thesis). Stockholm, Sweden: KTH Royal Institute of Technology. Retrieved from https://www.nada.kth.se/utbildning/grukth/exjobb/rapportlistor/2006/rapporter06/wagner_andreas_06047.pdf Widmer, G., & Goebl, W. (2004). Computational models of expressive music performance: The state of the art. Journal of New Music Research, 33(3), 203–216. Wing, A. M., Church, R., & Gentner, D. (1989). Variability in the timing of responses during repetitive tapping with alternate hands. Psychological Research, 51(1), 28–37. Wing, A. M., & Kristofferson, A. (1973a). The timing of interresponse intervals. Perception and Psychophysics, 13(3), 455–460. Wing, A. M., & Kristofferson, A. (1973b). Response delays and the timing of discrete motor responses. Attention, Perception & Psychophysics, 14(1), 5–12. Author’s Note All correspondence should be addressed to John R. Taylor Western Sydney University The MARCS Institute for Brain, Behaviour and Development Building 1 Bullecourt Avenue Milperra NSW 2214 Australia [email protected] Human Technology: An Interdisciplinary Journal on Humans in ICT Environments ISSN 1795-6889 www.humantechnology.jyu.fi