Xylophones are percussion instruments made of a series of tuned wooden bars of varying length, width and thickness (the ‘keyboard’). They are played by performers who use mallets to strike the bars (xylophones, marimbas, vibraphones and other pitched instruments are often referred to as ‘keyboard percussion’ or ‘mallet instruments’ in Western classical music). A common way to organize the keyboard is from low to high range (modern Western xylophones are a good example) but other configurations can be found around the world (for a review on xylophones, see (1)). Some xylophones have resonators mounted below the bars to enhance sound quality, resonance and sometimes tuning (2), but a common feature of xylophones is to produce sounds that are relatively short, especially in the high range (3). As a result, in most cultures, xylophone repertoire tends to privilege music at fast tempos where many notes are played in quick succession, and being able to play at high speeds with great accuracy is the hallmark of excellent performers: a good example in the Western tradition can be found in the numerous recordings of xylophone virtuoso George Hamilton Green, dating from the first half of the 20th Century (4).

During xylophone performance, the only contact between the performer and the keyboard is indirect and happens during the very brief moment of impact between the head of the mallet and the bar. Contrary to the practice of instruments such as the violin for example, where haptic feedback is rich and integral to the interaction between instrument and performer (5), xylophone performance provides a reduced amount of haptic feedback. The sense of sight is thus particularly solicited during training phases and performance, and the capacity to visualize the notes to be hit is a topic often raised during discussions among peers, as well as in the teaching studios of Western practitioners (6). Moreover, percussionists have to work with instruments whose shapes are asymmetric: bars differ in width and length between low and high ranges (in the low register of a marimba, which could be described as a bass xylophone, the bars may be twice as wide in the low range as in the highest range), and between models (even in the world of Western classical music where instruments are mass-produced, xylophone dimensions vary from one brand to another). Therefore, not only must percussionists accurately hit many different targets at high-speeds in a pre-set order (the different pitches that constitute the melody that they play), they must also adapt to the built-in variability that is inherent to the nature of their instrument.

Stemming from over 25 years of observation and practice as a professional percussionist and teacher of Western classical music, as well as an ethnomusicologist researching music from Central African Republic and Cameroon, the main question behind this research is fairly simple: where do performers look while they are playing the xylophone?

Internationally acclaimed marimba soloist Leigh Howard Stevens stresses the importance of vision and the role of anticipation in performance: “For me, the most important thing is visualizing what notes I’m going to hit” ((6) p. 37). This observation illustrates the well-known role of the anticipation of the gaze in many motor-control tasks (7) and leads us to focus our research question: how do xylophone players anticipate the action of their hands with their gaze while they are performing? In other terms, how are the eyes and hands synchronized in xylophone performance? Two related questions stem from this main interrogation: do all percussionists anticipate with their gaze in the same way? And what relationship exists between performance speed and gaze anticipation?

Research on music and eye-movement is largely based on experiments where musicians read the music they are playing, the experimental conditions varying between sight-reading and different levels of rehearsed performances. When the music is performed from a score, musicians tend to look ahead to the next notes to be played: the distance between the note being played and the note that is fixated upon by the eye is called Eye-Hand Span (8), or EHS. EHS can be denoted in duration (typically in milliseconds) or by the number of notes between hand and eye position, and depends on multiple factors such as the performance speed, the level of expertise of the performer or the complexity of the music to be performed (9, 10, 11, 12). The notion of awareness span (13) was introduced to measure the adaptation of EHS before a glance at the keyboard of the piano, taking into account the moments when the player needs to look at the keyboard a few milliseconds after reading the score, in order to perform accurately. This measurement helps to understand how pianists navigate between two systems of reference: the code that represents a musical score, and its application on the physical plane via the keyboard of the piano. However, when music is learnt by memory, EHS becomes less relevant since there is a ‘more-or-less complete dissociation of the eye movements from the written music […], presumably because the buffer is now obtaining its input form some other source, and the eye movements are no longer relevant’ (9). In other terms, the eye movements of musicians playing by memory are no longer driven by the necessity to refer to a physical musical score (paper or electronic device), and are thus focused differently (on the instrument or the environment).

Most of the research on anticipation, memorization and eye-movement in music is indeed based on music played from a score, and music performance in score-free conditions has seldom been studied. Moreover, xylophone performance has never been explored using eye-tracking devices: we do not know where xylophonists look when they perform, and what strategies they use during their performances to ensure the best accuracy. This article presents preliminary results from a larger interdisciplinary project dedicated to the study of instrumental gesture (14). It investigates how a xylophonist’s gaze moves during a performance, focusing on score-free musical tasks. Two contrasting contexts are examined in separate case-studies: classically-trained Western performers from the conservatory and university systems in Canada forming one group, and musicians belonging to three different ethnic groups from Cameroon (Bedzan Pygmies, Tikar and Eton; Figure 1), trained according to oral tradition by immersion, observation and imitation (15) in the other group.

It may seem paradoxical to lead a score-free study with classically trained musicians, who spend a large part of their musical practice reading musical notation and playing from a score; however, xylophone training is largely based (as is the case with many other instruments) on exercises and routines that are entirely performed by memory. In most cases performers also play their pieces by memory, and this is particularly true for orchestral excerpts, which are short passages from the orchestral repertoire that are used to test and select performers during professional auditions: playing by memory allows musicians to look at their keyboard in order to ensure the greatest accuracy throughout their performance.

For the first case-study, the measurements took place in Canada in two phases: the first phase involved 9 percussionists and involved a series of technical exercises based on scales, typical of Western musical training practices, as well as a musical excerpt from standard orchestral xylophone repertoire, recorded at the Centre for Interdisciplinary Research in Music, Media and Technology (CIRMMT, Montreal). For each task, musicians were asked to perform at different speeds to evaluate the effect of speed on eye-hand synchronization. In sight-reading tasks, different methods were used regarding the choice of tempo: self-paced, fixed by a metronome or a recorded accompaniment, or a combination of both. The latter method was used by Lehmann and Ericsson (16) who used trials with fixed tempo to measure differences in performance, but noted that the reasons behind the selection of tempo in self-paced trials were varying and not always clearly stated, and that variations in tempo could occur "without violating the range of acceptable tempi for performance" (p. 193). In the first phase of my research, xylophonists chose their own performance speed and played an excerpt of their choice: the freedom in the choice of performance speed aimed to collect data relative to potential thresholds linked to speed that would be dependent on individual skills. The second phase was run in the faculty’s percussion studio, with 3 musicians, one of them having already taken part in the first set of experiments but whose measurements were not of sufficient quality (high data loss rate). In order to facilitate inter-subject comparisons in this second phase, these 3 musicians were asked to perform an identical musical excerpt at 3 different speeds, which were based on the performance of another player from the first phase, for a total of 12 versions of the excerpt (4 players, each performing the excerpt at 3 different tempos). In both phases, the instrument used was a marimba, which is a type of xylophone (Figure 2a), so that the size of the bars would be comparable to the instruments used by some of the performers from Cameroon.

For the second case-study, I worked with 21 xylophonists from Central Cameroon (a region where I have conducted field research in ethnomusicology since 1999 (17, 18, 19)), who use two types of instruments: Tikar and Bedzan Pygmies play on banana-tree trunk xylophones (c and 2d), standing side by side in pairs, striking the ends of large bars loosely placed on banana-tree trunks, while Eton musicians play individually on multiple-resonator xylophones (Figure 2b), where the bars are attached to the frame and struck in the centre, organized in small ensembles ranging from 3 to 7 instruments. In order to preserve the performing conditions as close as possible to the normal conditions of performance, the measurements were realized in context (i.e. during live performances), in the villages, in outdoor conditions. The repertoire was based on the main pieces known by all of the instrumentalists within each village, and among different villages of the same ethnic group, to facilitate comparison. The identification of the pieces to be performed was based on interviews with the members of the different communities (musicians, elders), cross-referenced between different groups and villages, as well as on recordings and transcriptions, following methods described in (20) and (21). However, although the original goal of this study was to compare Western and African xylophone players, the measurement conditions in Cameroon (bright sun light, position of the instrument requiring downward gazes with the eyelids being half-closed) led to a dataset that was not consistent enough after post-processing, reducing the data to 5 musicians only, coming from 3 different ethnic groups. As a result, this second case-study reports observations based on individual players rather than on inter-subject comparisons.

The results of the first case-study, dedicated to Western players, are based on the analysis of three of the musical tasks: 1) a C-Major scale, played in strict alternation of the hands over two-octaves, starting and finishing on the central C; 2) an exercise based on the same scale, where the hands move in opposite directions, going in progressively larger sweeping motions from the central C to the next octave (Figure 3c and Figure 9); 3) an excerpt of ‘Porgy and Bess’ (Gershwin).

The first general finding is related to the position of the gaze on the keyboard: when playing on the natural keys, percussionists all fixed their gaze on the portion of the keyboard close to the area where both natural and accidentals (‘white’ and ‘black’ keys) are slightly superimposed, whereas they usually focused on the middle of the accidental keys when they played on those bars. Additionally, when playing several consecutive strokes combining natural and accidental keys within one gaze fixation, they often fixed a position on the bar that was close to the very corner of a bar, and located at the centre of an area covered by two or more consecutive strokes (Figure 6).

Main findings reveal a correlation between performance speed and a) the number of fixations, b) changes in lateral gaze movements for certain players. The comparison of 4 players for the musical excerpt of ‘Porgy and Bess’ gives insights on: a) the high reproducibility of gaze patterns for each individual across several performances, these patterns being nonetheless extremely individualized, and b) the relationship between Fixation-Duration, Eye-Stroke-Span and Note-Pattern.

The first task consisted of the execution of a C-Major scale, played in strict alternation, going up over two-octaves before returning to the original note, with 4 notes per beat for a total of 29 notes. All performers were asked to play at slow, medium and fast tempos; some also performed at an “extra-fast” tempo. Table 1 displays the individual speeds chosen by each performer for each speed category, along with the total number of eye fixations; there was no significant difference in the number of eye fixations depending on the trajectory of the musical scale (left to right for the first part of the exercise when pitches go up, right to left for the second part), t(31) = 1.57, p=0.12, (two-tailed paired Student’s t-test).

While the interpretation of ‘slow’, ‘medium’ and ‘fast’ varies significantly from one performer to the next, it is interesting to note that the number of eye fixations decreases with the augmentation of speed for all performers (Figure 7a). At a slow tempo, the xylophonists take the time to fix their gaze on most of the bars (up to 29 fixations for a total of 29 strokes). At a medium tempo, the decrease in the number of fixations varies quite significantly from one performer to the next, and at a fast tempo, half of the performers have decreased their number of fixations by half compared to the slow tempo. The extra-fast category sees further decreases in the number of fixations, and it should be noted that players start to lose accuracy.

A one-way ANOVA revealed a significant effect of tempo condition (Slow, Medium, Fast) on the number of fixations (F(30,2)=26.73, p<0.0001) as shown in Figure 7a. Post-hoc tests revealed significant differences across all three conditions (p<0.005). The ‘Extra-fast’ condition was not taken into consideration in this comparison since only 3 participants added this fourth category.

These results indicate that fixations are correlated to speed, typically going from a fixation for every note or almost every note (every xylophone bar) at the slowest tempo, to almost 3 notes at the fastest tempo. At the slowest tempo (42bpm), player #4 realizes 29 fixations, the duration of each note being 375ms, while at the opposite end of the spectrum in the ‘fast’ category (187bpm), each note is 80ms long and player #7 uses 10 fixations to perform the entire exercise, for a Fixation-Duration mean of 232ms; while the duration of each note is more than 4 times shorter in the second case, the number of fixations is not reduced proportionally. Figure 7b reveals a strong correlation between the augmentation of performance speed and the reduction in the number of fixations (r(34) = 0.907, p <0.01) which does not linearly scale but rather follows a logarithmic function.

Comparing the ratios between ‘slow’, ‘medium’ and ‘fast’ categories for performance speed and their corresponding inversed ratios for gaze fixations (Table 2) helps to underline that ratios vary considerably from one player to the next (see for example players #4 and #10 in the comparison between ‘slow’ and ‘fast’ speeds). Moreover, there are few exceptions where the ratio is similar (player #8 for the ‘slow’/‘medium’ comparison), or inverted (players #7, ‘slow’/‘medium’; players #5, #6, #10 and #11 for the ‘medium’/‘fast’ comparison).

The next results refer to an exercise based on the same scale (C-Major), where the hands move in opposite directions, making progressively larger sweeping motions from the central C to the next octave (Figure 3c and Figure 8). Performance speed was chosen by the players according to the same categories (Slow, Medium and Fast, as well as Extra-fast for a few players who performed another run).

For this exercise, players tend to cast their gaze ahead by 3 notes (ESS note index of 3): when the right hand reaches a high note, the gaze shifts immediately to the next lowest note to be played by the left hand, skipping the two central notes (C) that are played in between; however, there is an influence of performance speed on the eye-movements of the players. At slow speeds, xylophonists have enough time to look at every bar and anticipate not only the notes that are at the extreme end of the range that they need to cover, but they also focus on the central note of the exercise (C). When speed increases, they skip this intermediate step to jump from one extreme to the next, and when they reach the limits of their skills, their gaze tends to lose focus and synchronicity with the bars to hit, leading to pitch errors (Figure 9) or they stop moving their eye laterally to focus only on the central note of the range (C).

As was the case for the first set of results, there are numerous discrepancies between performers who use different strategies: while some of them try to anticipate all of the striking positions at all speeds, i.e. extremes and centre of the range (Table 3, Player #6), some stop switching their gaze from left to right when they reach a certain speed, fixing their gaze only on the central note C (Players #1, #4, #10 and #11). Player #8 begins applying this technique at the slowest speed. In general, the gaze-stop on the central note is associated with slower speeds, while the unique focus on the central note is associated with higher speeds, but the gaze-stop technique reappears at higher speeds when performers start to make some errors on this note.

This second set of results indicates that performance speed has an impact on the way performers fix the keyboard with their gaze, in this instance on lateral shifts. Xylophonists use different techniques with their gaze to combine accuracy and performance speed, in turn or in partial combination throughout the exercise: fixing on all of the striking spots, shifting between extremes, or focusing only on the central note of the range.

The third set of results is based on the analysis of the musical excerpt of the xylophone part from ‘Porgy and Bess’. Based on the performance of the first player (Player #2), three other xylophonists recorded the excerpt at the same respective speeds: 127bpm, 132bpm and 144bpm, all being within the range of an actual performance tempo with an orchestra. Data loss occurred once for each player, at different tempi: 6 notes for Player #1 at 127bpm (including one blink); 4 notes at 144 for Player#2; 6 notes at 144bpm for Player #3; 2 notes at 132bpm for Player #4 (blink). The data was either taken out of the results, or compensated by checking the eye-camera video, to detect any displacement or change in the gaze: if the eye and the head were both immobile, as could be verified respectively with the eye-camera and the scene-camera, it was assumed that the fixation was continuing through the data loss; else the data was discarded during the corresponding frames.

The number of fixations over the entire musical excerpt varies greatly from one player to the next (Figure 10): Player #1 uses less than half the fixations than Player #4. All players use more fixations at the slowest speed, however the number of fixations is the lowest at the intermediate speed for three of them. Player #3 displays the largest amount of variability in the number of fixations across the three versions (21.9%).

The variability in the total number of fixations depends on the duration of these fixations: less fixations in total implies longer single fixations, although the Fixation-Duration observed for Player #1 are not systematically twice as long as the Fixation-Duration of Player #4. The comparison of the Fixation-Duration for these players shows that Player #4 relies solely on short Fixation-Duration of 1 to 5 notes, while Player #1 uses fixations that can last up to 20 notes, i.e. the equivalent of 5 beats or 2.36 seconds (Figure 11). The longest Fixation-Duration was found in the performance at 132bpm of Player #1, with fixations lasting 23 and 24 notes (2.72 seconds) in musical sections A (a1 + a2) and A’-t2 (a1-t2 + a2-t2) (Figure 4).

When comparing individual players across versions, it appears that longer Fixation-Durations take place in the same musical segments, which is not true when realizing an inter-subject comparison. To facilitate comparison between versions and subjects, Fixation-Durations were grouped into 3 categories chosen to reflect recent findings by Cara about the limits of musical information processing of 7 +/- 2 notes (13), that confirm Sloboda’s findings (8): 1 to 4 notes, 5 to 9 notes, 10 and more. These categories create the opportunity to obtain a profile for each player, depending on their tendency to use shorter or longer Fixation-Durations (Figure 12). The influence of speed on the variation of Fixation-Duration for each category remains very individualized, each player displaying different variations from one speed to another.

As mentioned earlier, Fixation-Duration, ESS and Note-Pattern are interrelated, but not identical.

Such is the case for the relationship between Fixation-Duration and Note-Pattern: if a fixation is 5 notes long, it does not mean that the corresponding Note-Pattern will have the same duration (see Figure 5, for example); however, there is a strong link between both, and the longest Note-Patterns (24 and 25 notes long) are found in performances by players who have the longest Fixation-Durations. An interesting characteristic of Note-Patterns is that their segmentation remains relatively stable across versions for each of the players, who tend to amalgamate small chunks in longer patterns. The comparison of the versions at 132bm and 144bpm at the beginning of section A’ for Player #2 is a good example (Figure 13): from one version to the other, Player #2 amalgamates two chunks that are 3-note and 5-note long into a single chunk that is 8-note long (highlighted in yellow in Figure 13), ), proceeding in the same manner for the following chunks (highlighted in green). This process is not systematic, as can be seen with the last note-pattern in Figure 13, where a 6-note long Note-Pattern is divided in 2 sub-groups (in orange).

The categorization in 3 classes (1 to 4 notes, 5 to 9, 10 and more) has also been applied to Note-Patterns, to compare versions and players. For the intra-subject comparison, the influence of speed on the variation of Note-Pattern durations remains very individualized, each player favorizing different variations from one speed to another. The inter-subject comparison presents a profile quite similar to the Fixation-Duration, but the category 5 to 9 notes is larger for Note-Patterns than for Fixation-Duration for all players except for player #1 (Figure 14).

The possibility for a musician to perform identical musical passages with the same Note-Pattern segmentation, but with different Fixation-Duration, is based on the flexibility of the Eye-Stroke Span: such is the case, for example, for Player #2 in the musical segment a1 in sections A and A’, at 132bpm (Figure 15).

Duration of ESS vary from 0 to 6 notes across all versions, 0 indicating a moment when the player looks at the bar at the moment of the strike, which happens rarely. The majority of ESS have a duration of 2 or 3 notes. ESS durations can be grouped in 3 categories: 0 to 1 note, which indicates minimal gaze anticipation; 2 to 3 notes, corresponding to the majority of cases and to the strategies observed in the scale exercises; 4 and more notes, indicating a gaze anticipation of at least one beat. By color-coding these 3 categories, it is possible to produce a sort of ‘map of anticipation’, where red corresponds to minimal anticipation and blue maximal anticipation (Figure 16), that helps to proceed to intra-subject comparison. Such comparisons show that players tend to exhibit shorter or longer anticipations in the same musical passages, beyond variations in ESS.

A comparison between all players across all versions reveals that all performers favour an ESS of 2 or 3 notes, but that the ESS seems to be more independent of the other variables, as demonstrated by the repartition of categories across players that differs from Fixation-Duration and Note-Patterns graphics (Figure 17).

The results of the second case-study are based on measurements and observations realized among 3 different ethnic groups in Cameroon. As aforementioned, data quality was uneven after post-processing and the results are based on a few individual players, rather than on inter-subject comparisons. However, even when the eye-tracking data were not exploitable, the image of the eye-camera provided precise information on the opening or closing of the eye, and the scene camera, combined with the other video camera placed in front of the player, reported additional information on the orientation of the head and of the gaze of the performer.

The main findings are linked to the time musicians spend looking at or away from their instrument, and to ESS. The main difference between xylophonists observed in Cameroon and in Canada is that Cameroonian players do not always look at their keyboard while playing. Two types of performance practices can be distinguished, based on the variety and the complexity of the music being performed. Among the Bedzan and Tikar, players use banana-trunk xylophones, where each person strikes a limited range of 4 to 6 notes. The xylophone bars are also quite large at the striking position (6-9 cm), and as the music is extremely repetitive with limited melodic variations, the necessity to look at the instrument to improve accuracy is less prevalent for those musicians. The same conditions apply to xylophonists who have an accompanying role in Eton music, though the musical material presents more variations (pieces include several sections that create more melodic and rhythmic variations; there is more interaction between players who react to one another’s variations; an accompanist might take short solos). The situation is somewhat different for the soloists whose role is to lead the Eton ensembles: their musical task is divided between sections in which they play repetitive patterns with few variations – often singing at the same time – and sections in which they improvise on their instruments, without singing. During the singing or steady pattern sections, they look at their accompanists (looking at their faces or at their hand motions) or at the audience, whereas during the solo sections, they look at their own xylophone.

Most players spend a significant amount of time looking away from their keyboard: in Bedzan and Tikar villages, musicians generally look away from their instrument 60 to 75% of the performance time. Their gaze is focused on singers or dancers surrounding them, or on their fellow xylophonists’ sticking patterns. Recordings in the Tikar village of Ngambé showed that when musicians were looking at their own sticks or bars, the time spent on their zone of playing ranged from 0.2 to 3 seconds, with the vast majority of time spent on this zone under the duration of a full musical cycle (around 2 seconds). It is also not rare for musicians to close their eyes during a performance, and one of the most experienced xylophonists among the Tikar was observed fully closing his eyes over the span of 20 consecutive seconds, which represents 10 iterations of the musical cycle.

When they look at their keyboards, Bedzan and Tikar focus their gaze close to the point of impact (at the extremity of the bar) or just beyond the extremity of their stick or mallet (Figure 18). Eton players also focus close to the point of impact, which is in the middle of the bar for their xylophones.

Among Bedzan Pygmies, the performer playing in the high range executed a short cycle of 6 beats subdivided in 3 notes at 178bpm, using only 4 bars (minimal operational values are 112ms long), the cycle being repeated with small variations. Each hand struck one of the two central bars to execute the basic pattern, while the two bars on the outside were used to add variations, generally on the sixth beat of the cycle. Eye movements were limited, focusing mainly on the bar played by the right hand during the basic pattern, leading to Fixation-Duration lasting most of the cycle, with slight repositioning on the bar every beat or second beat. The main gaze shifts occurred one or two beats before the variation (ESS of 3 or 6), the Fixation-Duration on the new note lasting for an entire beat (Figure 19), as well as the Note-Pattern. Occasional gaze shifts also took place towards the bar played by the left hand in the basic pattern, with an ESS of 2 or 3 notes, and for a Fixation-Duration of 2 to 3 notes also.

The solo xylophones in Eton ensembles count up to 20 bars, tuned in a heptatonic scale. In solo sections, the xylophonist executes melodic and rhythmic variations that include techniques such as passages in parallel thirds or octaves (simultaneously playing bars that are respectively set apart by one bar, or by six bars), or rolls on ascending or descending scales, which consist of a fast series of repeated strikes on each bar before moving to the next. These techniques could be extracted from their original context and executed separately to record specific eye-movements. For parallel thirds, the performer repeated each double-stroke once before moving to the next double-stroke, playing two double-strokes per beat at 164bpm, in a general ascending motion from the lowest bar to the highest bar (for Eton players, this means playing from the right to the left, the lowest bars being positioned to the right of the performer). The player consistently looked at the usual point of impact of the mallet with the bar, i.e. at the centre of the bar. For the first part of the exercise (6 beats), the performer had to hit the bars positioned on his right side and focused his gaze on the bar that separated his mallets. A shift occurred around the 7th beat, while he was playing on the bars located in front of his body, and his gaze shifted towards the bar that was struck by the left hand (leading the motion towards the upper register). Gaze shifts took place after the repetition of the second double-stroke, the gaze shifting to the next fixation just before the beginning of the next set of double-strokes, for an ESS of 0.5. Gaze shifts lasted 120 to 150ms.

For roll sequences on continuous scales, the performer repeated each note 3 or 4 times before changing note, playing 14 to 15 strokes per second. The gaze was always focused on the first or second bar to the left or to the right of the bar that was played, depending on the global motion (towards the upper or the lower register). ESS was of 3 or 4 notes. Saccades were 33 to 50ms long, a duration comparable to those reported in Lehman and Kopiez (22), and they usually took place right after the last stroke of a group of repeated notes on the same bar.

It should be noted that the data corresponding to the portion of the keyboard located in front of the body of the player was lost.