Although many keyboard players may never have encountered a situation in which they cannot hear themselves, few would be surprised to learn that it is possible to play quite accurately in the absence of auditory feedback. Generations ago, concert artists often practiced on portable "dumb" keyboards during their long travels, an experience that now can easily be recreated on an electronic instrument. However, even though it is possible to translate a printed score or a memorized representation of music into appropriate actions on a silent keyboard, there is nevertheless something strange and unnatural about this task, and the accompanying auditory imagery is only a poor substitute for the real sound of an instrument. Also, good pianists always carefully monitor the sounds they produce and may adjust their playing on the basis of what they hear. This raises the question of whether silent playing can be really as precise and as expressive as performance with normal auditory feedback.

Several earlier studies have investigated the effect of auditory feedback deprivation on keyboard performance. Ebhardt (1898) compared repeated performances of various musical excerpts on a piano with and without sound. He found that absence of auditory feedback, which he achieved by pulling out the keyboard and action from underneath the strings, resulted in a somewhat slower tempo of performance. However, only a nonsignificant tendency in the same direction was found by Gates and Bradshaw (1974), who asked six musicians to rehearse and play a study on an electronic organ with and without sound. Neither of these studies assessed accuracy or expression. Recently, Banton (1995) counted errors in sightreading and found no difference with and without feedback.

A more detailed study of the effects of auditory feedback on keyboard performance was conducted by Finney (1997), who took advantage of MIDI technology. Eleven pianists of various skill levels played excerpts from two Bach Inventions on an electronic keyboard with and without sound. There were no significant differences between the normal and silent conditions with regard to number of errors, overall tempo, overall key-press velocity, or variability in note onset timing. In a subsequent perceptual judgment task, performances that had been produced without feedback were judged to be worse than normal performances 62% of the time, but apparently this result was not significantly different from chance.

Finney’s study did not focus specifically on expression, and his musical excerpts and unweighted keyboard were not ideally suited for that purpose. In fact, his main concern was the effect of delayed or altered auditory feedback on performance, and the silent condition was included merely as a control. The present study focuses specifically on expressive aspects of piano performance with and without auditory feedback, in order to find relatively subtle and perhaps previously overlooked effects of this factor.

The musical excerpt used was the beginning of Chopin’s Etude in E major, op. 10, No. 3, a famous passage that requires fine control over timing, dynamics, and pedaling. A computer-generated score, without slurs and expression marks, is shown below. The excerpt is terminated here with a chord. The melody in the soprano voice is divided into several rhythmic groups, each ending with a long note, as indicated by the brackets above the score. The alto voice exhibits continuous motion in sixteenth notes. The tenor voice establishes a syncopated rhythm, and the bass underlines the harmonic alternation of tonic and dominant. Following the initial eighth-note upbeat, there are 36 nominally equal note inter-onset intervals (or IOIs) corresponding to sixteenth notes, as indicated below the score.

Six pianists participated. They included one middle-aged amateur (B.R.) and five young pianists with professional-level skills: one with a master’s degree in piano performance from the Yale School of Music (D.G.), one graduate student (T.C.) and one undergraduate student (H.S.) in the same program, and two excellent undergraduates from Yale College, one senior (K.S.) and one sophomore (M.S.). After a brief warm-up, each pianist performed the Chopin excerpt 20 times from the score on a Roland RD-250s digital piano equipped with a simple sustain pedal switch. For the first ten performances, auditory feedback was provided over earphones; for the second ten, the earphones were disconnected. None of the pianists had recently practiced the Chopin Etude; in fact, four of them had not played it previously but knew it well from listening. The performances thus may be characterized as fluent sightreading. The pianists were asked to play with expression but not to change their interpretation across the 20 repetitions. The performances were recorded in MIDI format. The data were converted to text format and analyzed using a spreadsheet program.

The results will be described under five headings: horizontal timing (the timing of successive notes), horizontal dynamics (the relative intensities of successive notes), vertical timing (the asynchronies among nominally simultaneous note onsets), vertical dynamics (the relative intensities of simultaneous notes), and pedaling. This analysis is almost exhaustive; only articulation (gaps and overlaps among successive notes) was not examined, as it plays little role in this excerpt, due to continuous use of the sustain pedal.

Horizontal timing. Horizontal timing refers to the sequence of IOI durations in the music, defined by the onsets of successive primary notes. Primary notes were defined as the highest-pitched notes in all simultaneities. Thus, they included all sixteenth notes of the soprano melody and those of the alto accompaniment that occurred during longer melody notes. The initial eighth-note upbeat was ignored, so that all IOIs in the excerpt were nominally equal but of course unequal in any actual performance, due to expressive timing. The sequence of IOIs within a performance has three aspects: (a) the average IOI duration (inversely related to the basic tempo), (b) the range of variation of the IOIs (the "temporal modulation depth"), and (c) the pattern of the IOIs (the "timing profile"). In addition, there is the variability of IOI durations across different performances.

Each pianist had his or her individual timing profile, although there were also considerable similarities among pianists. All profiles resembled a typical timing profile for this music, established in an earlier study (Repp, 1997a). Peaks in the profiles represented slowing at salient points in the musical structure, particularly at the ends of melodic-rhythmic groups. The timing profiles for the two playing conditions were highly similar for all six pianists, as indicated by correlation coefficients of 0.95 to 0.98. Nevertheless, they were significantly different for all six pianists, as assessed by the condition-by-position interactions in two-way analyses of variance. Two pianists (T.C., K.S.) greatly increased their final ritard when feedback was absent, and the others showed smaller tendencies in that direction. However, even when the final position was omitted from the statistical analyses, the profiles were still significantly different for all pianists. The differences were relatively small, however, and seemed idiosyncratic. Only two pianists (T.C., M.S.) showed a significant main effect of condition, due to a somewhat slower basic tempo without feedback, as was observed by Ebhardt (1898). T.C., who played with very little temporal modulation to begin with, seemed to play somewhat more expressively without feedback.

The standard deviations of the 36 individual IOIs across the 10 performances in each condition were plotted as a function of the mean IOI durations in that condition for each pianist. The regression lines all had positive slopes, indicating that IOI variability increased with IOI duration. This increase is generally observed in motor behavior for IOIs longer than 300 ms and has also been shown to hold in expressive music performance (Repp, 1997b). Differences in IOI variability between the two conditions were assessed by sign tests. Two pianists (B.R., T.C.) showed a significant increase in variability when auditory feedback was withheld, and a third pianist (H.S.) showed a marginally significant tendency in that direction.

Horizontal dynamics. Horizontal dynamics refers to the relative intensities of successive primary notes, as defined earlier. The measure of relative intensity used here is MIDI velocity, reflecting the velocity of key depressions on the digital piano. The acoustic intensity of the piano tones is a monotonic but nonlinear function of MIDI velocity. Like horizontal timing, horizontal dynamics has three aspects: (a) the average MIDI velocity (the dynamic level), (b) the range of variation (the dynamic range), and (c) the pattern of velocities (the dynamic profile).

Peaks in the average dynamic profiles corresponded to soprano melody notes, valleys to alto accompaniment notes. The profiles were highly similar among the six pianists, with M.S. showing a greater dynamic range than the others. Also, the dynamic profiles for the two playing conditions were quite similar for all pianists, with correlations ranging from 0.88 to 0.99. Nevertheless, the condition-by-position interactions in the two-way ANOVAs were significant in all cases, even for M.S., due in part to the large number of degrees of freedom. Four pianists (B.R., D.G., T.C., K.S.) also showed a significant main effect of condition, reflecting an increase in average dynamic level in two cases but a decrease in the other two cases. However, there seemed to be a consistent effect of withholding feedback for these four pianists, namely a slight reduction of dynamic rangeùa raising of the profile valleys in two pianists (D.G., T.C.) and a lowering of the profile peaks in the other two (B.R., K.S.).

The standard deviations of the MIDI velocities of the primary notes were plotted as a function of their mean velocities in the two playing conditions. The slopes of all regression lines were negative, indicating a tendency for dynamic variability to decrease as dynamic level increased; in other words, melody notes tended to be less variable than accompaniment notes. However, there was no evidence for greater dynamic variability in the absence of auditory feedback. On the contrary, one pianist (D.G.) showed significantly less variability in the no-feedback condition, and another one (K.S.) showed a marginally significant difference in the same direction.

Vertical timing. Asynchronies among nominally simultaneous note onsets were computed relative to the alto voice because this is the only voice that has a note onset in every sixteenth-note position. The harmonic filler notes that occasionally appear between the soprano and alto voices were called "mezzo". The data were examined in the form of scatter plots pitting the two playing conditions against each other. There were 62 data points in each plot, representing individual asynchronies averaged across the 10 performances in each condition. Some pianists (B.R., D.G.) played much more synchronously than others. Negative asynchronies for the soprano voice indicated that all pianists tended to lead with the melody. (Note that this differentiation occurred within the right hand.) Mezzo notes, also played by the right hand, tended to coincide with or lag behind the alto voice. The same was true for the tenor notes, which were played by the left hand. The pianists differed in their timing of the bass notes: While the bass tended to lag behind the alto voice in four pianists, K.S. and especially M.S. had a tendency to advance bass notes where they coincided with a melody note, so that the bass note sometimes even preceded the melody note. This is reminiscent of a playing style often found in pianists of earlier generations, traces of which can also be found in some of today’s young pianists (see Repp, 1996a).

The patterns of asynchronies looked reasonably similar for the two conditions in all cases. The correlations, computed across all voices, were substantial (0.73 to 0.89) but lower than those for horizontal timing and dynamics. Apart from a few outliers, the data points clustered around the main diagonal in all cases, which suggested little systematic change. Nevertheless, in two-way ANOVAs on all 62 asynchronies, all six pianists showed significant two-way interactions with condition. Thus, once again, there were effects of feedback deprivation that were both subtle and seemingly idiosyncratic. The standard deviations of the asychronies across the 10 performances in each condition were compared by means of sign tests. Two pianists (B.R., K.S.) showed significantly greater variability in the absence of feedback, but one (D.G.) showed significantly lower variability.

Vertical dynamics. Like the asynchronies, differences in MIDI velocities among nominally simultaneous notes were computed relative to the alto voice. This aspect of the data relates to tonal balance and texture. Scatter plots of the MIDI velocity differences showed that the soprano melody was consistently more intense than the alto voice. The bass notes were usually of about the same intensity as the alto notes, while tenor notes were weaker. The relative intensity of mezzo notes differed greatly according to their position in the music. The correlations between the two playing conditions were high (0.87 to 0.96), and the data points generally clustered around the main diagonal. Nevertheless, ANOVAs on the 62 velocity differences revealed significant two-way interactions with condition for all six pianists (though they were only marginally significant in the case of D.G. and K.S.). Again, it must be concluded that subtle differences existed.

Differences between the two conditions with regard to the variability of velocity differences were assessed by sign tests. Only one pianist (D.G.) showed a marginally significant difference, but it indicated lower variability in the no-feedback condition. He is the one who had also shown reduced variability of horizontal dynamics and of asynchronies in the absence of feedback.

Pedaling. Pedal actions have two aspects (Repp, 1996b): (a) Pedal use, that is when the pedal is depressed and when it is not, and (b) pedal timing, the exact times at which the pedal is depressed and released. In the present musical excerpt, the sustain pedal is depressed almost the whole time, so that its use is characterized sufficiently by the frequency and position of pedal changes (i.e., the quick succession of a pedal release and a pedal depression). Pianists must change pedal at least with every harmonic change in the music (i.e., 6 times in the Chopin excerpt) or, perhaps more realistically, with each beat (9 times), but they can also change pedal with every melody note within beats and with the accompaniment notes in bar 4 and possibly elsewhere (i.e., as often as 28 to 36 times). A comparison of average pedal change frequencies in the two playing conditions for the six pianists revealed, for the first time, some large effects of withholding auditory feedback. The largest effect was shown by B.R., the only real amateur in the group. He tended to pedal very frequently in the normal condition, but much less often without feedback. Two of the young professional-level pianists (H.S., M.S.), who pedaled less frequently than B.R. to begin with, also showed a substantial reduction in the no-feedback condition, essentially to the bare minimum. The two most experienced young pianists in the group (D.G., T.C.) showed smaller, but nevertheless significant reductions. However, K.S. showed a significant increase in pedaling frequency.

The times of pedal releases and depressions were expressed as percentages of the IOIs in which they occurred (i.e., relative to the onsets of the preceding and following primary notes). These relative times were averaged across the 10 performances in each condition, and the averages were subsequently edited to eliminate those based on only a single occurrence or showing unusually large variability. Scatter plots of the remaining average pedal times showed that pedal releases usually occurred shortly after the onset of a primary note, which created the desired legato effect. T.C. had a tendency to release the pedal early, whereas H.S., K.S., and especially M.S. often released the pedal late, thereby producing considerable overlap between successive tones. The times of pedal depressions were highly dependent on position in the music. The data points generally clustered around the main diagonal, and four pianists showed high correlations between the two conditions (0.93 to 0.98). Two pianists (D.G., M.S.) showed lower correlations (0.75, 0.76), in each case due to a single outlier. In D.G.’s case, this data point represents the first pedal depression right after the initial upbeat, which he for some reason made much later in the absence of auditory feedback. In M.S.’s case, however, the outlier represents a pedal depression in bar 4.

Because pedal use was not entirely consistent across performances, the matrix of pedal timing data contained many empty cells, and therefore the data were not subjected to statistical analysis. However, it seems likely that there were again small but significant differences between conditions. Several pianists (D.G., H.S., K.S.) showed a tendency to release the pedal a little earlier when feedback was absent, thereby reducing the legato overlap between successive notes.

So, is expressive performance affected by the absence of auditory feedback? The answer is: yes and no. The detailed analyses presented here reveal significant differences between the two playing conditions in all parameters of expression that were examined: horizontal timing, horizontal dynamics, vertical timing, vertical dynamics, pedal use, and probably also pedal timing. In addition, some pianists exhibited greater variability in some of these parameters when feedback was removed. However, nearly all of these differences were small in size, showed different patterns for different pianists, and were often difficult to describe and interpret. It seems likely that most of these differences are perceptually and aesthetically insignificant. One exception to this conclusion may be the substantial reduction in pedal use observed in some pianists. It seems to confirm the point, often made in the pedagogical literature (e.g., Neuhaus, 1973, p. 162), that pedaling is "governed by the ear". Heinlein (1930) observed long ago that accurate pedaling is difficult in a variety of conditions different from normal performance and concluded that "[t]he damper-pedal response is so highly integrated into a unity of effect with the other phases of a pianoforte performance that alteration of any single factor or group of factors ... may seriously modify a pianist’s customary pedal interpretation" (p. 527). It should be noted, however, that some pianists’ pedal use and all pianists’ pedal timing was not seriously affected by auditory feedback deprivation. A perceptual study in which listeners are asked to discriminate performances drawn from the two playing conditions remains to be conducted; it will probably be a difficult task.

A final comment is in order on the representativeness of the present data. One certainly could imagine situations in which larger effects of auditory feedback deprivation might be obtainedùfor example, if pianists had no opportunity to practice the music before they play on a silent keyboard or if they are beginners or if the music poses great technical challenges. However, these situations do not seem of great theoretical or practical interest. The more interesting issue is whether highly practiced, professional-level performance suffers a necessary degradation in the absence of auditory feedback. This refers to situations in which the effects of auditory feedback deprivation would most likely be even smaller than they were in the present study. Therefore, the data tend to support the conclusion of earlier authors that absence of auditory feedback has only a negligible influence on competent keyboard performance. This suggests that the complex motor activities of performance are guided almost entirely by mental representations and plans, including auditory imagery, and that they do not require immediate auditory feedback, although kinaesthetic feedback is probably important. This is not to deny that delayed auditory feedback can wreak havoc with a performance (Gates & Bradshaw, 1974; Finney, 1997) and that normal auditory feedback is vitally important in learning a piece, refining its interpretation, and fine-tuning a performance to an unfamiliar instrument. After all, music is meant to be heard, not just to be felt and imagined, and it is highly unlikely that any pianist would prefer silent playing to real music-making.