[This article so far appears only on my own web site. I do not know of a suitable place for its publication (with sound examples and preferably coloured graphics) and will appreciate constructive suggestions by email via my home page http://nettheim.com.
The author acknowledges helpful comments by Dr. Bruno Repp.
This version is dated 23 August 2001, revised 27 May 2005, re-formatted 26 September 2009, minor revision 3 March 2010.]

Copyright © Nigel Nettheim 2001, 2010.

[After clicking on a footnote or reference, please use your browser's back button to return to the main text.]

A Musical Microscope
Applied to the Piano Playing of
Vladimir de Pachmann

by Nigel Nettheim

Abstract Analysis of Pachmann's playing
Introduction Comparison with Pachmann's rolls; pedalling
Chopin's Nocturne in e minor Comparison with other performers
Pachmann's recordings Evaluation of the present method
Measuring timings Conclusions and future work
Measuring amplitudes Footnotes
Testing the measured timings and amplitudes References
Visual display of the nuances Appendix I: Nuance data
Appendix II: Examples reproduced for comparison


Par 1 Excerpts from a recording by the famous early pianist Vladimir de Pachmann are measured and analysed. Conclusions are drawn on his approach to piano playing. A comparison with the playing of the same music by several other performers highlights Pachmann's originality. The graphical method of displaying the performing nuances is apparently novel, and may be found useful in research and pedagogy.


Par 2 Vladimir de Pachmann (1848-1933) was one of the most famous pianists in the world for nearly fifty years. He was famous on two accounts: his eccentric personality and his playing. His personality produced much humour, usually well received, while his playing produced many supreme moments, especially in the smaller works of Chopin, his admirers including even Liszt. For many years after his death the plentiful anecdotes told about him served as his chief memorial and it is only more recently that, thanks to the wider availability of his best recordings, his playing has been rescued from the inattention it had suffered.  1 

Par 3 In this article I will deal with just one characteristic bar from a recording of Chopin's Nocturne in e minor (the chosen bar is reproduced in the piece, so three bars in all are to be studied). I will first briefly discuss the chosen piece, and then Pachmann's recordings. I have measured the timing and amplitude of the tones in the chosen bars from his 1927 recording, and will present those measurements in a nuanced score display. The display is designed to make clearly visible the departures from literal playing employed in the performance, and to be conveniently read and played at the piano. By analysing the nuances, I will try to infer some of the principles according to which Pachmann approached the interpretation of music on the piano. I will compare Pachmann's recording with those made by some other famous Chopin players: Leopold Godowsky (1928), Artur Rubinstein (1937) and Vladimir Horowitz (1953). I will finally evaluate the present method and offer several conclusions.

Chopin's Nocturne in e minor

Par 4 The Nocturne in e minor, op.72 no. 1, by Chopin (1810-1849) was written in 1827.  2   It will be helpful to be aware of its form, several views of which could be defended (see Table 1): (i) binary (or in earlier terminology three-part song-form), (ii) rondo (or even sonatina, reflecting the key-scheme), and (iii) one-part.  3   Whichever view of the form is adopted, two features are important here. The first feature is the appearance of the main thematic material four times (since "A" is itself a parallel double-period), beginning in bars 2, 10, 31 and 39; since, however, the third of these appearances is ornamented, producing a different number and pattern of notes, I will restrict attention for the present purpose to bars 2, 10 and 39. The second feature is the likelihood that the downbeats are to be felt at each bar line (rather than after each ½ bar or 2 bars); bars 2-5, for instance, constitute a regular four-bar phrase.  4 

Table 1. The Form of Chopin's Nocturne op.72 no.1: Alternative Views

Material   (i): "A" "B" "A" Coda
Material  (ii): "A" "B" "A" "B"
Material (iii): "A" Codetta "A" Codetta
Keys: e-B B e-E E
Bars: 1- 23- 31- 47-57

Pachmann's recordings

Par 5 Pachmann's recordings are of variable quality, some having been made when the recording technology was young, others when he himself was old. Even the best of them could no doubt reflect only a little of what it would have been like to be present at one of his recitals, but some can, I believe, reveal a great deal. Pachmann recorded the present Nocturne three times: as a reproducing roll in 1906 and 1923, and by the electrical process in 1927. We will be concerned here just with the electrical recording, often considered to be one of his finest.

Measuring timings

Par 6 It was desired to measure the timing of each tone in the recorded performance. First, one would like to know the appropriate playback tempo. That is important for two reasons: (i) it is suspected that even numerically small changes in playback tempo can make a well shaped phrase sound not so well shaped, that is, the properties of performance nuances depend to some extent upon the tempo; and (ii) comparisons may be desired between different performances of a given work by the same artist or by different artists. The determination of the correct tempo is, however, not always easy. The story begins with the sound originally created by the artist. Various transfers between media then took place, yielding a physical recording (78, LP, CD etc.). The physical recording is then played back on certain equipment for final listening. There is no guarantee that speed changes were not introduced during that series of processes. The best available clue for checking the speed appears to be the tuning (A=440 etc.) of the piano in the original recording session: if that is matched on final playback, the tempo heard must be correct. However, the original tuning is unlikely to be known with certainty.  5 

Par 7 I chose the Dante ( 1996 ) transfer. On counting the oscillations in the sound wave for the g'' at the beginning of bar 2 it appears that the tuning is approximately A=441.9. That is quite close to the present standard A=440 (the tunings of the nearest neighbours on an A440 keyboard are A#=466.2 and G#=415.3). In view of the difficulties just mentioned, I simply note the tuning in case further evidence on this point comes to light in the future.

Par 8 The timing was carried out by playing the selection at 1/7 of the original speed, without changing the pitch (several computer programs provide this option; I used "Musician's CD Player").  6   Only the tone onsets were timed, not the ends of tones, which were in any case affected by pedalling (in unpedalled music it would usually be desirable to time also the ends of tones). As the music sounded slowly I watched the player's counter and assigned the starting time of each tone, coordinating ear and eye. A re-timing of one selection indicated that in almost all cases the results were reproducible to within 0.1 sec of the slowed-down playing (14 msec of real time). Rarely, they were only within 0.2 sec of 7-fold time (28 msec of real time). Testing by sound synthesis (see later) indicated that this was sufficiently accurate for the purpose.

Par 9 Other methods for measuring timings could be considered. Repp ( 1998 , p.1087) used a waveform to measure nominally simultaneous onsets, but he did not measure the departures from simultaneity which are essential in the present project.  7   Cook ( 1987 , p.271) attempted to automate the timing from waveform analysis but spurious attacks required manual editing; his example was of essentially single-note music. Bowen ( 1996 , p.130) used a method of "tapping along" for timing bars or beats but not the smaller units needed here. Measurement may readily be made from modern recording pianos such as the Disklavier, but the playing of the great artists of the past is not directly available to that technology (it is available indirectly by playing reproducing rolls with a Vorsetzer in front of a Disklavier). Other methods will be mentioned later in connection with piano rolls.

Measuring amplitudes

Par 10 It was desired to measure the amplitude of each tone in the recorded performance. First, one would like to know the general level, set by adjusting the volume control on playback (this question parallels that of the tempo in connection with measuring timings). Unfortunately, very little evidence is available on this, apart from the fact that Pachmann was known as a "pianissimist".

Par 11 In the absence of information on the general level, only relative values can be estimated. This was attempted by ear alone. An obvious question arising in estimating relative amplitudes is that of the consistency of the estimates. Thus when the relative amplitudes of sounds A and B have been estimated, and subsequently of B and C, the question arises whether the estimate comparing A and C matches the implication of those earlier estimates. When more sounds are involved, the number of such comparisons increases quadratically. It is not easy to maintain the level or calibration over a long period of time, even after checking to remind oneself of earlier measurements. Another difficulty is that of comparing aurally the amplitude of sounds of greatly different pitch. It is also hard to judge the amplitude of each of several simultaneous or near-simultaneous sounds. Finally, the sustaining pedal may again add difficulty.

Par 12 Still using "Musician's CD Player" to play the recording at 1/7 of the original speed, I assigned values on a scale from 0 to 100, where 100 corresponds to an approximate maximum attainable.

Par 13 Other methods for measuring amplitude could be considered. The picture provided by the waveform might be adequate for single-note melodies, but it is not sufficiently revealing when several notes sound together. Repp ( 1999 , pp.1973-4) applied mathematical analysis to waveforms, but his purpose allowed him to restrict attention to the notes constituting the main melody; his measurements were derived from the joint contribution of melodic and accompanying notes without attempting to deal specifically with the latter. Cook ( 1987 , p.271) automated the measurement of amplitude from waveforms but did not indicate the details of his procedure; his example was of essentially single-note music. A reference chart of sounds of single notes at various pitches and amplitudes could be prepared beforehand, possibly adopting the "velocity" parameter of the MIDI format, but accompanying sounds are often present in the music in various combinations of pitch and amplitude, so this method seems unlikely to succeed. The earlier comment about the use of the Disklavier for timing applies also to amplitudes. The measurement of amplitudes from piano rolls will be mentioned briefly later.

Testing the measured timings and amplitudes

Par 14 Fortunately, a test for both timing and amplitude measurements is available by converting the measurements to MIDI data; the listener then judges whether the effect of the synthesized playback is sufficiently similar to that of the original. This is only an informal test; factors not fully accounted for include the tone quality of the piano sound and the exact effect of the use of the pedals. Nevertheless it is of some value, and satisfactory confirmation was obtained by this means in the present work.

Visual display of the nuances

Par 15 For the visual display of performance nuances, progression through the piece is most often assigned to the horizontal axis. Timing nuances have usually been displayed vertically and relatively as departures from the nominal values corresponding to a literal reading of the score (e.g. Cook, 1987 , p.259; Desain & Honing, 1993 , p.134).  8   Those displays no doubt suited their authors' analytical and statistical purposes, but here we wish to keep in touch as closely as possible with the artistic task of performance. We therefore display the timing nuances horizontally and absolutely, which more nearly corresponds with the notation to which musicians are accustomed.

Par 16 Given our manner of display, it is a far from trivial question over how long a section of music the timing departures should be cumulated. Different chosen spans will in general lead to different appearances of the data and to different conclusions being drawn from it. If the span were to comprise a whole piece or movement, the displacements would be likely to become large (accumulating over the various accelerandos, ritardandos and other tempo changes that might occur) and so, being divorced from their local surroundings, lose their musical meaning. If the span were to comprise only a note or two, the context would not be taken into account sufficiently.

Par 17 The solution I have adopted is to take the span between successive downbeats. The question naturally arises how to determine the location of the downbeats. They most often occur at each bar line, but by no means always, for they may instead occur after every half bar or every two bars (or rarely another number of bars). There is apparently no term in general use for the span in the score from one downbeat to the next; "conducting span" might perhaps be found satisfactory.  9   No algorithm exists by which to determine the conducting span from the composer's notated score, but rather musical discernment is needed, together with an examination of relevant features of the score in hand; it will be sufficient here to note our assumption, mentioned above when considering the form of the e minor Nocturne, that the conducting span is one bar in that piece, that is, the downbeats occur at each bar line.

Par 18 Not only nuances of timing but also those of loudness (amplitude) are to be shown. It was tempting to use the size of the note-heads for this purpose; this might indeed be the method of choice, given output of suitably high resolution, but the current relatively low resolution of computer screen displays could result not only in irregular shapes appearing but also in a lower value being displayed with more pixels than a higher value, thus giving a misleading impression. The proximity of staff lines can also affect the perceived size of the note-head. I chose to use instead the saturation of colouring in a hue-saturation-intensity colour representation, with fixed hue and intensity.  10  

Par 19 In the following figures the lower score shows the literal rendition of the music; here the horizontal positions of the notes strictly follow their nominal (mathematically exact) timing. The upper score shows the nuanced performance; here the notes are deflected left or right from the nominal positions according to the timing used in the performance. In both scores the main (downbeat) melodic note is printed at the bar line. The light vertical grid lines assist the presentation, the moment when each tone begins corresponding to the left-most point of its note-head. The bars as printed by this method may have different lengths, depending upon the time taken for the performance to traverse each given bar.  11   The horizontal time scale is the same in all the figures, no matter what was the duration of the bar in performance. The two combined scores—original and nuanced—can easily be played from jointly at the piano (or just the nuanced one, if one knows the original score); this opens up the possibility of using the display in music study and instruction, which is potentially an important feature of this display compared with previous ones.

Analysis of Pachmann's playing

Par 20 What can be learned from the measurements thus made and displayed? Let us first study Pachmann's playing of bar 2 (please play Sound Example 1 while watching Figure 1, after checking that your system volume is appropriately adjusted; if clicking on the sound example is not successful, the sound file may be saved to disk for playing by other means).

Figure 1. Chopin, Nocturne in e minor, bar 2, played by Vladimir de Pachmann (1927) (Sound Example 1).

nuanced score 72-1 bar 2 Pachmann

It is obvious both to the ear and to the eye that there is a wide departure from a mechanical (mathematical, or literal) reproduction of the score; that is, the nuanced score differs considerably from the original one. We will document the observed nuances, to that extent taking a scientific approach, but we will also attempt to relate the observations to their possible role in musical performance and expression, to that extent taking an artistic approach. The indulgence of readers who may lean to either side of that potential schism will be appreciated.

Par 21 We define the (main) melody to be the first five notes of the treble staff: b', g'', f''#, e'', d''#, the remaining 13 notes constituting the accompaniment. The melody is played quite loudly, as is indicated by the intense (near-maximal) red colouring in the Figure. Although that is not sufficient to produce the effect of a singing quality, it seems necessary, for, by analogy, if one tries to sing with a withdrawn attitude, the breath will not be poured out sufficiently to engage the vocal apparatus appropriately. The accompaniment is played quietly, which allows the melody to predominate.

Par 22 The nuances in the accompaniment pick out significant relationships within its part, using both timing and dynamics. The notes c', b form an appoggiatura and resolution, and are played closer together than the surrounding accompaniment notes. This applies especially to accompaniment notes number 5 and 6; it also applies to notes 11 and 12, although it will shortly be seen that additional factors come into play there. It is often held that an appoggiatura requires a louder dynamic level than its resolution but it appears that, at least in cases such as the present one, an alternative is available by means of timing alone. On surveying all the accompanying notes, it is observed that the g's are played a little louder than the others, perhaps on account their being sounded, as the third of the chord, only twice compared with four times for the chord root, e, and four for the chord fifth, b; this seems to give the resultant sound a good harmonic balance, but other factors could also be involved here.

Par 23 The second accompaniment note, B, is considerably delayed, allowing a longer-lasting musical gesture to be carried out for the first melodic note of the bar, the downbeat. This is one of the commonest nuances in playing the music of Chopin and some other composers of the romantic period. Further, the opening three notes (b', E, g''), combining melody and accompaniment, are considerably broadened, that is, extra time is allocated to them, so that a specially strong and penetrating gesture can be conveyed, as if the melodic g'' is grasped very strongly. The grace-note, b', written in small size as if the composer did not wish to pin down a fixed length for it, takes part fully in this gesture. It is clear that if any note is relatively lengthened within the bar, others must be correspondingly relatively shortened, once the duration of the bar is known. That is why the choice made earlier of the bar as unit is vitally important for the analysis of the performance: if another unit had been chosen (e.g. the quarter-note, the half-bar, or the phrase—which is most often four bars in length), the graphical display and the analysis would be different, for the point at which the nuanced version is displayed as sounding exactly together with the original version would be different. Such different choices of the unit would not necessarily be wrong, but they would serve different purposes.

Par 24 The last three notes of the melody, f''#, e'' and d''#, are also broadened, again allowing the melodic gesture to be given in a very fully-felt manner. To support this effect, the accompaniment must fit in with the melodic purpose, just as a good accompanist would do for a singer. Thus the accompaniment is gradually accelerated, from its second note, b (which had been delayed for the reason mentioned earlier) to its tenth note, e; indeed, that e arrives a full triplet 8th-note ahead of its literal time. The remaining three accompaniment notes are broadened in sympathy with the last three melody notes. Now a most important nuance is observed: the melodic f''# is not sounded together with its accompanimental e, but follows it. This allows the melody to be heard for itself, as the accompaniment is not allowed to compete for the listener's attention at the moment when the melodic note is sounded. The following c and b in the accompaniment are placed between the three melodic notes, but they also come a little closer together than if exactly so placed, reflecting their pairing as appoggiatura and resolution. The last melodic note, d''#, is most important as the aim of the melodic progression: it is to be given its place in the light and so not subject to the competition of a simultaneously sounded accompanimental f#. But if the f# were to be played before the d''#, then that accompanimental event would to some extent distract attention from the melodic progression; the accompanimental note is therefore played after the melodic d''# (and softly).

Par 25 Reviewing bar 2 as a whole, we see that the melodic progression has been composed using a very common note-pattern, and there is no strong dissonance or unusual harmony present to suggest heightened expression, but Pachmann has nevertheless extracted every ounce of melodic conviction from it. All the means just mentioned assisted the melody to give its best singing effect. The first three notes (b', E, g'') are spread out in the performance so as to convey a strongly grasping gesture, and the same is true of the last three notes of the melody (f''#, e'', d''#). The fact that no comparable event occurs half-way through the bar to some extent supports our specification of one notated bar per downbeat. The broadening at both ends of the bar is made possible by the central hastening. It may be asked why Pachmann has savoured the last three-note gesture even more than the first one: the answer may be that the last gesture has two intervening notes (c' - b) to be delivered as well, whereas the first gesture does not; otherwise, the artistic conception of the performer is the deciding factor.  12  

Par 26 We now consider the reproductions of bar 2 in bars 10 and 39 (Sound Example 2 and Figure 2, Sound Example 3 and Figure 3; all the examples of this article are amalgamated in Appendix II in case the reader wishes to make a more direct comparison between them).

Figure 2. Chopin, Nocturne in e minor, bar 10, played by Vladimir de Pachmann (1927) (Sound Example 2).

nuanced score 72-1 bar 10 Pachmann

Figure 3. Chopin, Nocturne in e minor, bar 39, played by Vladimir de Pachmann (1927). (Sound Example 3).

nuanced score 72-1 bar 39 Pachmann

Several differences are observed in the composition of those bars: the melody is now given in octaves, there is no initial grace-note, and the dynamic level is increased from one reproduction to the next. In this performance bar 39 was given considerably more time than bar 2, and bar 10 most of all. In the first octave in bar 10 Pachmann has played the lower note well in advance of the upper note; together with the playing of the bass note earlier still, a three-tone pattern is produced somewhat analogous to that created in the performance of bar 2, the lower note of the octave playing a role similar to that of the grace-note in bar 2.  13   The remaining octaves are shown in the Figure as played with the upper note sounding slightly before the lower one, though it was not possible to be quite certain about that feature of the measurements (compare footnote  7  ). The broadening of the last three melodic notes in bar 10 is much greater than in bar 2, both relatively to the bar and absolutely, taking into account the greater duration of the playing. Indeed, this musical gesture at the end of bar 10 could well be considered the highlight of the whole performance of the Nocturne and even, when heard in context, one of the supreme moments in the recorded history of the playing of Chopin's smaller works. In bar 39 the gesture is slightly less fully broadened; that is perhaps appropriate to its stronger dynamic level, the effort being exerted, so to speak, vertically down into the music rather than horizontally across it. In both bars 10 and 39 the accompaniment is now hastened even more than in bar 2, the tenth note e now arriving nearly two triplet 8ths ahead of its nominal time, whereas in bar 2 it had arrived just one ahead (compare the notes of the scores with the grid lines in the Figure). The melodic f''# also arrives earlier in bars 10 and 39 than in bar 2—that is, relatively earlier though absolutely later in view of the longer time given to the bars. Apart from the intensification thus achieved, the approach to performing the three bars is on the whole similar.

Comparison with Pachmann's rolls; pedalling

Par 27 It is tempting now to compare the electrical recording just discussed with Pachmann's rolls of the same piece. Rolls can in some respects yield more information than can other recording modes. Such extra information may include: (i) the pedalling used, (ii) the exact timing of notes played nearly simultaneously, and (iii) the duration of the holding down of the keys in certain situations. The tempo is also indicated explicitly (see footnote  5  ). However, questions may arise concerning the exact relation between what was played by the recording artist and what appears on the roll; in particular, a certain amount of editing was often carried out, especially for expression. Piano rolls had very limited capacity to record the differing amplitudes of individual notes played simultaneously or nearly so. The properties of the playback mechanism may give rise to additional questions. I therefore decided to postpone the use of rolls for a future project. It may be reported informally, though, that the general impression made by the rolls of the present piece is generally similar to that made by the electrical recording.  14  

Par 28 In the excerpts studied here from the electrical recording it is obvious that the sustaining (right) pedal was used, but it is difficult or impossible to detect the moments at which the pedal was changed (released and soon afterwards reapplied). In any case, the pedalling may vary with the piano used, the acoustic environment, and the requirements of the recording process, so that detailed observations of its use on any one occasion may have little instructional value. I nevertheless examined the pedalling in Pachmann's rolls of this piece from the encoding on the rolls themselves. In the Welte roll (1906) the changes occur in bar 2 on the notes g''#, d''#; bar 10 g''#, e'', d''#; bar 39 g''#, f''#, d''#. In the Duo-Art roll (1923) they occur in bar 2 on the notes g''#, f''#, d''#; bar 10 the first E, d''#; bar 39 the first E, the second E, f''#, d''#. The una corda or soft (left) pedal is also encoded on the rolls. In the Welte roll it is applied in bars 2 and 10 just for the melodic notes e'' - d''# (naturally taking in also a few accompaniment notes), but not in bar 39. In the Duo-Art roll the soft pedal is not used in the bars referred to.

Comparison with other performers

Par 29 It is generally of questionable value to compare the performances of just one bar by different artists. The artists' conceptions of the work as a whole may differ in many ways, in which case the given bar might play a different role in the various conceptions. It is not artistically viable for a performer to make each bar "the last word" in expressiveness; instead, a limited number of places in the score will be chosen for the greatest expressiveness. Apart from the artist's conception of the piece, his or her mood and the occasion on which it was played can be relevant in any comparison, as well as a certain random element inevitably associated with any one performance. It should also be borne in mind that the present bar and its reproductions were chosen as being considered especially excellent in Pachmann's performance. Certainly there is no intention here to rate any performance as better or worse than another. Nevertheless it is of some interest to apply the present method to other pianists and to see the extent to which Pachmann's nuances may be present.

Par 30 We begin these comparisons with the recording by Leopold Godowsky, also a Chopin specialist; this recording was made in 1928, the year following the recording by Pachmann just considered (Sound Example 4 and Figure 4).

Figure 4. Chopin, Nocturne in e minor, bar 2, played by Leopold Godowsky (1928) (Sound example 4).

nuanced score 72-1 bar 2 Godowsky

One of the first features noticed (by ear and by eye) in Godowsky's performance is that there is very little nuancing in the accompaniment. Loudness nuances were not expected to be featured there but, apart from a slight delay before sounding the 2nd note, B, the timing is also very uniform. Pachmann's time-pairing of the appoggiatura-resolution pair in the 5th and 6th notes, c', b is not present. There is no hastening of the notes towards the 4th quarter of the bar, because there was no intention to broaden the final three melody notes. The initial grace note is tossed off just as an acciaccatura having little melodic function, and the downbeat notes, E and g'', are played nearly together; the combined effect of those two features is that Pachmann's broad three-note initial gesture is replaced by a small two-note gesture. The last half of the bar is an almost exact duplicate of the unnuanced original score, except only for the relative loudness of melody and accompaniment. The performance, though not as mechanical as that of a direct computer realization, may be summarized as quite literal-minded.  15  

Par 31 The next example, recorded in 1937, is by Artur Rubinstein, also well known for his playing of Chopin (Sound Example 5 and Figure 5).

Figure 5. Chopin, Nocturne in e minor, bar 2, played by Artur Rubinstein (1937) (Sound example 5).

nuanced score 72-1 bar 2 Rubinstein

Here the accompaniment is more nuanced than Godowsky's and less than Pachmann's. A delay is present before playing the 2nd note, B; the third degree of the harmony, g, is somewhat heightened dynamically; notes 5 and 6 are brought together by timing; and there is a hastening at the end of the bar, the last b having only a short duration before the following downbeat. The initial grace-note is played quite solidly as if it truly belongs to the melody and the initial accompaniment note, E, is played slightly ahead of the downbeat, so that the initial gesture is broadened a little. The hastening at the end of the bar, just before the following downbeat, applies not only to the accompaniment but also slightly to the melody, the final e'' being played a little later than halfway between the f''# and d''#. Thus the downbeat d''# is approached with a noticeable directness. In summary, this performance lies between Godowsky's and Pachmann's in degree of nuancing, and Rubinstein has replaced Pachmann's broadening at the end of the bar with a slight hastening.

Par 32 A last comparison will be made with the 1953 recording of Vladimir Horowitz, also famous for his playing of Chopin (Sound Example 6 and Figure 6).

Figure 6. Chopin, Nocturne in e minor, bar 2, played by Vladimir Horowitz (1953) (Sound example 6).

nuanced score 72-1 bar 2 Horowitz

The contrast in loudness between Horowitz's melody and accompaniment seems to be nearly maximal. Indeed, his 2nd accompaniment note was not heard at all. The following g and e are relatively loud, perhaps in response to the b not having sounded. The first c' and b are slightly paired off by timing. There is a delay before the second B is played (the 8th note of the composed accompaniment—compare the delay of the other players before the first B is played), but Horowitz has played the E and B so softly that an effect of dividing the bar into two half-bars is not noticeable. Horowitz also played the initial grace-note very softly, apparently indicating that he did not consider it a fully-fledged part of the melody. Since the downbeat notes, E and g'', are played nearly together, there is no broad beginning gesture of Pachmann's type. In the last three melodic notes the f''# and e'' are somewhat paired off by timing, perhaps reflecting their dissonance-consonance relationship. To summarize, the main feature of Horowitz's performance is the extreme dynamic contrast between melody and accompaniment.

Par 33 We now reflect upon the comparisons just made. The most prominent feature is that the broad gestures at each downbeat characterize Pachmann's recording but not the other recordings examined here. That in turn implied a far greater freedom in Pachmann's accompaniment, in order to allow the melody to deliver those gestures. If Chopin had intended such a non-literal reading, he could not have been expected to spell out the details in a score such as our "nuanced" one, so that the insight of the performer would be essential. As was mentioned in Par 25, Pachmann's conception is not obviously required by any harmonic, contrapuntal, rhythmic or formal features of the score, so again a special insight, involving a purely melodic impulse, seems to have been needed for his approach.  16  

Par 34 The freedom in Pachmann's accompaniment is of two kinds. First, his rubato consists of a hastening in the middle of the bar to allow the broadening around the downbeats. Second, simultaneity between accompaniment and melody is avoided, apparently in order to let the melody shine through more clearly, for the accompaniment does not then compete for the listener's attention at the melodic moment. Of course, some other pianists have also shown a willingness to depart considerably from simultaneity, in certain repertoire, for reasons which might be investigated; but the performances of the present excerpt by the other pianists just examined do not show that. Pachmann's accompanimental nuances, however, involve a sacrifice: a uniform smoothly-flowing effect is ruled out, flowing which may well be thought characteristic of pieces of the Nocturne type. Alternative priorities in this music may be formulated as melodic conviction versus background atmosphere. Each of the pianists represented here balanced those alternatives in his own proportions and in his own manner. Similarly, different listeners will have different expectations and different capacities for responding to an alternative approach.

Par 35 It may be noted in passing that the kind of thinking that evidently produced Pachmann's performance could probably not satisfactorily be simulated by any computer program for automated performance. Too many features of the composition play a role (or, to be more accurate, too many types of features). Thus attempts such as that of Clynes ( 1983 ) would not be expected to be capable of producing anything like Pachmann's results; for that, one must look more deeply into the composition than the postulation of a small number of broad principles allows.

Evaluation of the present method

Par 36 We now ask what the present method of study can contribute towards an understanding of recorded performances. We first note that some limitations are present in any method of studying recorded performances: the general level of dynamics is not accurately known, the correct playback tempo is often not known with certainty, and different recording techniques and piano qualities can add difficulty to the task of comparing performances. Those factors cannot be controlled, in general. It must further be acknowledged that a nuanced score as presented here does not contain a complete specification of the performance. The pedalling details have not so far been included, and the duration through which each note is held is not indicated (in an unpedalled piece it would be needed).

Par 37 Apart from those limitations, it might be suggested that the nuances described here could have been gleaned just from a very close listening to the recordings. On the other hand, the amount of detail that emerges in a thorough study is considerable: it is not easy to hold all of it in mind at one time, but that seems much easier with the present visual aids. The eye may direct the ear to details and patterns which might otherwise have escaped its attention. Indeed, in some cases to be documented in future work, the unaided ear cannot even discern which notes are played, as for instance in trills, rapid passages, and faint accompaniment notes. Further, the visual presentation allows the nuances in one or more performances to be compared directly at one time, a feat which the ear cannot successfully duplicate. The verbal formulation of possible rationales for the nuances, an important part of the approach, is also facilitated by the visual documentation; when more examples are available, it will be possible to attempt a confirmation or refutation of such hypotheses.

Conclusions and future work

Par 38 This paper has investigated: (i) a method of measuring and studying recorded piano performances, and (ii) the features of Pachmann's playing in particular, so far in just one bar and its reproductions. The main conclusions are as follows:

(i) The method adopted for measuring recorded piano performances and displaying the measurements for study seems successful in the examples so far seen. The display of timing nuances seems quite clear. The display of loudness nuances seems satisfactory, though other methods could be considered in the future. Different repertoire might require different methods.

(ii) An important feature found in Pachmann's playing is the hastening of the accompaniment in some portions, apparently in order to allow for a broader rendition of the melody in other portions. This involves far greater freedom from a literal-minded reading of the score than was found in several comparisons with recordings by other artists. It is also noted that Pachmann often avoided simultaneity of accompaniment and melody notes, most likely again in order to promote the clear shining forth of the melody.

Par 39 Future work is intended as follows:

(i) The application of the present method in more examples from the recorded repertoire, including longer selections and perhaps the use of piano rolls.

(ii) The further study of Pachmann's playing, including a comparison with his detailed published editions of the scores (which, however, apparently do not include the e minor Nocturne).

Par 40 Finally, it is hoped that the method of study adopted here will throw instructive light upon the great artistry of Pachmann at his best, that the method will be found useful in research and pedagogy, and in particular that it may contribute towards freeing today's and tomorrow's performers from inappropriately literal-minded playing.



1 When Liszt, who was familiar with Chopin's playing, first heard Pachmann he said to the audience "Those who have never heard Chopin before are hearing him this evening" (Blickstein  1969 , 59). For further information about de Pachmann see the author's web site http://nettheim.com/pachmann.
2 The Nocturne was first published posthumously in 1855. In the following I refer to the edition by Paderewski (  1949 ) in which, however, the slurs have been added by the editors.
3 Goetschius ( 1898 par 99b, p.183) indicates three parts but Goetschius ( 1915 par 96, p.127) considers it a rondo, as does Leichtentritt ( 1921 , vol. I, pp.59-60).
4 The question of the location of downbeats will be discussed later in connection with the visual display of nuances.
5 Tunings differed between places and times. It was not until 1939 that an International Conference fixed musical pitch at A440. In 1936 the American Standards Association had settled upon that pitch, confirming what the US music industry had adopted in 1925. But Pachmann's 1927 recording was made in London, and the history of tuning in England seems not well documented. In any case, the tuning of the particular piano used at the time could hardly be known with certainty.
  Some second-order problems may arise too: the pitch of a given tone may change a little during its course (because of the behaviour of the struck string), and the piano may not be perfectly in tune. Fluctuations of speed in processing and/or playback constitute a separate possible problem, not dealt with here.
  In the case of recorded rolls, the speed of the roll was indicated on the roll itself in feet per ten minutes. That reveals the tempo of the performance, on the assumtion that the speed of the recording apparatus was accurately set; that assumption, which seems reasonable, can no longer be tested.
6 The program is currently available at http://www.ronimusic.com/. If the playback is slowed down by a factor of more than about 7, the beginning of the tones may become somewhat fuzzy, making it difficult to assign the starting time by ear. The measurements were made from computer files in WAV format but the examples to be given later have been converted to MP3 format in order to save space. In the conversion a certain amount of detail is lost which is not noticeable in normal playback but which could conceivably be noticeable in slowed-down playback.
7 It is often difficult to determine simultaneity or to separate approximately simultaneous notes from an acoustic recording; for instance, the two notes of an octave might in fact be struck together but one of them might come to be heard only after the sound of the other has decayed a little. This difficulty may increase when the sustaining pedal is used, as overtones come into and out of clear audibility, possibly competing for attention with a note played softly (e.g. Pachmann's 1927 recording, bar 39, the 2nd bass note). I had aimed to time the striking of the keys, or more precisely of the strings, but in such cases it is difficult to deduce that from the evidence of the sounds produced.
8 Desain & Honing apparently there cite Jaffe. Desain & Honing's Figures 2 and 4 seem not to indicate the vertical scale.
9 On the nature of the progression of the music from one downbeat to the next valuable light is thrown by Becking ( 1928 ), to which an introduction may be found in Nettheim ( 1996 ). According to Becking's approach, the span from one downbeat to the next is the span of a single conducting curve (the sense in which "conducting" is used here being carefully defined). Methods for determining the span of the conducting curve are discussed in Nettheim ( 1998 ). Fortunately, even if the span of the curve is judged incorrectly or cannot be determined with certainty, the present method will still yield usable, though then perhaps not optimal, results.
  It should be noted that the approach of displaying separately each conducting curve—in the present case each bar—is not related to the so-called "tyranny of the bar line", for the performance can spill over or around the bar line in any way the performer might wish, and that will be reflected in the present display. The precise moment of the beginning of the downbeat must be decided in cases where nominally simultaneous notes at that point in the score are arpeggiated in the performance; in the present cases that moment is taken as the beginning of the main melodic note at the downbeat (rather than of any accompanying notes).
10 The sethsbcolor operator of the PostScript programming language was used. The red hue was chosen, in which very low saturations are easier to see on a white background than if intensities of grey had been chosen; the intensity was set to its maximum value of 1. It turned out that a linear mapping from estimated loudness to saturation did not produce a visually linear increase in perceived saturation. I therefore experimented with alternative mappings, but was not able to improve the result overall and therefore retained the linear one; further investigation of this matter might be worth while.
11 A study of the performance of the piece from a large-scale point of view (not part of the present project) would deal with the pattern of bar durations in performance.
  The duration assigned to the lower score has been matched to that of the upper score of the given bar, for the present purpose. A Metronome Mark (MM 4th = 69) was provided in the original score. MMs are not always considered entirely reliable or obligatory under all conditions. In any case, they presumably allow for some tempo variability within the piece and so may have limited significance for any one bar. The MMs observed in the performances in Figures 1-6 are, respectively: 64.86, 54.92, 56.74, 69.97, 79.21, 66.85.
12 In this connection, it may be remarked that broadening the playing of the melody in comparison with a somewhat less broad or even hastened playing of the accompaniment can strongly suggest a singing quality. It is as if the singer has a broader conception of the passage than has the instrumental accompanist. That occurs in actual singing and accompanying, particularly clear exaples being heard in the some of the recordings of Elena Gerhardt, for instance of Brahms's Feldeinsamkeit. In such early recordings the accompaniment was sometimes deliberately hastened out of consideration for the limited possible duration of a recording (4½ minutes or less). Even if that merely extraneous motivation were the cause, the artistic/singing effect mentioned is nevertheless present. Such an effect can then be rendered by a solo pianist, the one performer then controlling both parts. Comparisons between the pianist Pachmann and the soprano Adelina Patti were common, and Patti also commented enthusiastically on Pachmann. In the nuances seen here such comparisons can take on a specific practical significance, thus being more than a vague analogy.
13 The downbeat in the performance of bar 10 seems to be shared between g' and g'', that is, between the two notes of the right hand octave. That is suggested by listening to the fade-in at the beginning of Sound Example 2. The effect is achieved by bringing the preceding notes d'#, e', f'# to a strong completion on the lower note, g', before embarking upon the melody starting on the upper note, g''. Thus the downbeat notated at the bar line has been divided into two performed events: a downbeat at the end of the bar just finished and a separate downbeat at the beginning of the bar to follow. Pachmann's separate rendering of these two "Siamese-twin" downbeats has led to the considerable elapse of time between them. This feature is seen in many compositions of Chopin, Schumann and others.
14 Gottschewski ( 1992 ) measured timings from piano rolls, restricting detail to the quarter-bar; he noted "a certain inaccuracy of punching" in the rolls (ibid., p.95). Husarik ( 1986 ) briefly discussed the properties of rolls and measured the degree of non-legato in some of Josef Hofmann's playing.
15 Neither here nor elsewhere do I intend the analytical remarks on the excerpts played by the famous pianists to be taken as pejorative.
16 The primary place of melody in Pachmann's approach is well illustrated in an anecdote from one of his recitals:

Or the charm of some passage strikes him anew. "The melody!" he exclaims enthusiastically, and he marks out the melody for a bar or two, so that the audience may be under no mistake. It is a recital and a lecture in one. (Hadden,  1914 , p.64)


Books and Articles

Becking, Gustav. Der Musikalische Rhythmus als Erkenntnisquelle. (Musical rhythm as a source of insight). Augsberg: Benno Filser, 1928.

Blickstein, Edward. "More than a Clown". High Fidelity Magazine, July 1969: 59-62.

Bowen, Jose Antonio. "Tempo, Duration, and Flexibility: Techniques in the Analysis of Performance". Journal of Musicological Research 16 (1996): 111-156.

Clynes, Manfred. Expressive microstructure in music, linked to living qualities. Royal Swedish Academy of Music Publication No. 39, 1983.

Cook, Nicholas. "Structure and Performance Timing in Bach's C Major Prelude (WTC I): An Empirical Study". Music Analysis 6 (1987): 257-272.

Desain, Peter and Henkjan Honing. "Tempo curves considered harmful". Contemporary Music Review 7 (1993): 123-138.

Goetschius, Percy. The Larger Forms of Musical Composition. New York: Schirmer, 1915.

________. The Homophonic Forms of Musical Composition. New York: Schirmer, 1898. (Reprinted New York: AMS Press, 1970.)

Gottschewski, Hermann. "Graphic Analysis of Recorded Interpretations". Computing in Musicology 8 (1992): 93-96.

Hadden, J. Cuthbert. Modern Musicians: a Book for Players and Singers, London & Edinburgh: T.N.Foulis, 1914.

Husarik, Stephen. "Piano Rolls: Untapped Technical Resources". Clavier April 1986, 14-16.

Leichtentritt, Hugo. Analyse der Chopin'schen Klavierwerke. (Two volumes) Berlin: M. Hesse, 1921-22.

Nettheim, Nigel. "A Schubert fingerprint, related to the theory of metre, tempo and the Becking curve". Systematische Musikwissenschaft (Systematic Musicology) 6 (1998): 363-413.

________. "How Musical Rhythm Reveals Human Attitudes: Gustav Becking's Theory". International Review of the Aesthetics and Sociology of Music, 27/2 (1996): 101-122.

Repp, Bruno H. "A microcosm of musical expression. I. Quantitative analysis of pianists' timing in the initial measures of Chopin's Etude in E major." J.Acoust.Soc.Am. 104(2) Pt 1 (August 1998): 1085-1100.

________. "A microcosm of musical expression: II. Quantitative analysis of pianists' dynamics in the initial measures of Chopin's Etude in E major". J. Acoust. Soc. Am. 105(3) (March 1999): 1972-1988.

Musical Scores

Chopin. Complete Works. Edited by Ignaz Paderewski. Polish Music Publications, Cracow, 1970 (1949). Vol. 7, Nocturnes.

Sound Recordings

Chopin, Nocturne in e minor op.72/1, played by:

Leopold Godowsky: LP MJA (1967), recorded in London, 23 June 1928.

Vladimir Horowitz: LP RCA VH 018 (1975), recorded 25 February 1953.

Vladimir de Pachmann: CD Dante HPC 056 (1996), recorded in London, November 1927.
Reproducing roll Welte Licensee C-7202 (1925), recorded in Leipzig, 1906. Recut by Leedy Bros., Michigan, 2001.
Reproducing roll Duo-Art 016 and 6675-4 (1923), recorded in London, 1923. Recut by Meliora, Atlanta, Georgia, c1999.

Artur Rubinstein: LP EMI Electrola C 187-00 162/63 (date?), recorded April 1937.

Appendix I: Nuance data

Table 2. Nuance Data

N = Note name (melody shown in blue)
T = Score position in triplet 8th-notes x 10
P = Performance identification
S = Performance tone onset in seconds
E = Performance tone onset range extended between 0 and 120
D = Difference E - T (these are the timing nuances)
A = Amplitude (loudness) 0-100 (shown in pink)
Fixed reference values are shown in green.
Note that the order of the columns is in some cases not the order of the notes played.

N b' E g'' B g e c' b E B g e f''# c' e'' b d''# F#
T -5 0 0 10 20 30 40 50 60 70 80 90 90 100 105 110 120 120
P Pachmann (1927) bar 2:
S -0.59 -0.37 0.00 0.37 0.57 0.87 1.14 1.33 1.64 1.89 2.14 2.41 2.61 3.01 3.19 3.40 3.70 3.86
E -19.0 -12.0 0.0 12.0 18.5 28.3 37.1 43.1 53.3 61.2 69.5 78.3 84.8 97.8 103.3 110.3 120.0 125.1
D -14.0 -12.0 0.0 2.0 -1.5 -1.7 -2.9 -6.9 -6.7 -8.8 -10.5 -11.7 -5.2 -2.2 -1.7 0.3 0.0 5.1
A 45 23 87 14 28 21 18 16 15 14 32 28 60 24 51 19 81 17
P Godowsky (1928) bar 2:
S -0.11 0.04 0.00 0.39 0.67 0.94 1.20 1.47 1.73 2.04 2.30 2.59 2.56 2.87 3.00 3.17 3.43 3.44
E -4.0 1.5 0.0 13.5 23.5 33.0 42.0 51.5 60.5 71.5 80.5 90.5 89.5 100.5 105.0 111.0 120.0 120.5
D 1.0 1.5 0.0 3.5 3.5 3.0 2.0 1.5 0.5 1.5 0.5 0.5 -0.5 0.5 0.0 1.0 0.0 0.5
A 19 27 86 22 32 40 23 28 24 15 23 25 84 24 77 26 83 30
P Rubinstein (1937) bar 2:
S -0.40 -0.16 0.00 0.39 0.60 0.86 1.13 1.31 1.57 1.87 2.04 2.31 2.29 2.57 2.70 2.83 3.03 3.04
E -15.8 -6.2 0.0 15.3 23.8 34.0 44.7 52.1 62.3 74.2 80.9 91.7 90.6 101.9 107.0 112.1 120.0 120.6
D -10.8 -6.2 0.0 5.3 3.8 4.0 4.7 2.1 2.3 4.2 0.9 1.7 0.6 1.9 2.0 2.1 0.0 0.6
A 63 14 94 24 32 39 10 16 21 35 39 30 88 17 84 19 78 21
P Horowitz (1953) bar 2:
S -0.33 -0.09 0.00 0.73 0.99 1.27 1.51 1.71 2.11 2.39 2.59 2.67 2.97 3.07 3.23 3.59 3.56
E -11.0 -2.9 0.0 24.4 33.0 42.5 50.7 57.4 70.8 79.8 86.5 89.4 99.4 102.8 108.0 120.0 119.0
D -6.0 -2.9 0.0 4.4 3.0 2.5 0.7 -2.6 0.8 -0.2 -3.5 -0.6 -0.6 -2.2 -2.0 0.0 -1.0
A 9 15 97 0 32 33 16 6 11 17 20 16 90 8 90 10 85 12

N E g' g'' B g e c' b E B g e f''# f'# c' e'' e' b d''# d'# F#
T 0 0 0 10 20 30 40 50 60 70 80 90 90 90 100 105 105 110 120 120 120
P Pachmann (1927) bar 10:
S -0.63 -0.47 0.00 0.41 0.64 0.94 1.16 1.43 1.81 2.07 2.31 2.59 2.83 2.86 3.39 3.67 3.70 4.00 4.37 4.39 4.41
E -17.3 -12.9 0.0 11.4 17.6 25.9 31.8 39.2 49.8 56.9 63.5 71.0 77.6 78.4 92.9 100.8 101.6 109.8 120.0 120.4 121.2
D -17.3 -12.9 0.0 1.4 -2.4 -4.1 -8.2 -10.8 -10.2 -13.1 -16.5 -19.0 -12.4 -11.6 -7.1 -4.2 -3.4 -0.2 0.0 0.4 1.2
A 69 42 87 3 20 24 23 18 25 18 31 17 41 24 22 46 19 28 40 18 22
P Pachmann (1927) bar 39:
S -0.29 0.04 0.00 0.49 0.70 0.94 1.21 1.49 1.81 2.01 2.26 2.50 2.71 2.76 3.31 3.46 3.47 3.74 4.23 4.24 4.13
E -8.1 1.2 0.0 13.8 19.9 26.8 34.5 42.2 51.5 57.2 64.1 70.9 77.0 78.2 94.1 98.1 98.5 106.2 120.0 120.4 117.2
D -8.1 1.2 0.0 3.8 -0.1 -3.2 -5.5 -7.8 -8.5 -12.8 -15.9 -19.1 -13.0 -11.8 -5.9 -6.9 -6.5 -3.8 0.0 0.4 -2.8
A 63 89 98 10 27 31 16 19 29 26 37 43 65 61 20 60 50 22 52 46 24

Appendix II: Examples reproduced for comparison

Figures 1-6 and their sound links are reproduced here to assist comparison among them.

Figure 1. Chopin, Nocturne in e minor, bar 2, played by Vladimir de Pachmann (1927) (Sound Example 1):
nuanced score 72-1 bar 2 Pachmann
Figure 2. Chopin, Nocturne in e minor, bar 10, played by Vladimir de Pachmann (1927) (Sound Example 2):
nuanced score 72-1 bar 10 Pachmann

Figure 3. Chopin, Nocturne in e minor, bar 39, played by Vladimir de Pachmann (1927). (Sound Example 3):
nuanced score 72-1 bar 39 Pachmann
Figure 4. Chopin, Nocturne in e minor, bar 2, played by Leopold Godowsky (1928) (Sound example 4):
nuanced score 72-1 bar 2 Godowsky
Figure 5. Chopin, Nocturne in e minor, bar 2, played by Artur Rubinstein (1937) (Sound example 5):
nuanced score 72-1 bar 2 Rubinstein

Figure 6. Chopin, Nocturne in e minor, bar 2, played by Vladimir Horowitz (1953) (Sound example 6):
nuanced score 72-1 bar 2 Horowitz

Nettheim home page                  Index of Pachmann items