Proceedings of the International Symposium on Musical Acoustics, March 31st to April 3rd 2004 (ISMA2004), Nara, Japan

Once again: The perception of piano touch and tone.

Can touch audibly change piano sound independently of intensity?
Werner Goebl1 , Roberto Bresin2, Alexander Galembo3
1
Austrian Research Institute for Artificial Intelligence (ÖFAI), Vienna, Austria
2
Department of Speech, Music, & Hearing, Royal Institute of Technology (KTH), Stockholm, Sweden
3
Sechenov Institute of Evolutionary Physiology and Biochemistry, St. Petersburg, Russia
werner.goebl@oefai.at,roberto@speech.kth.se,galembo@galembo.mail.iephb.ru

Abstract reduce the problem to its very essence: Is it possible to

This study addresses the old question of whether the tim- produce two isolated piano tones without using the pedal
bre of isolated piano tones can be audibly varied inde- with identical final hammer velocities, but with audibly
pendently of their hammer velocities — only through the different sounds?
type of touch. A large amount of single piano tones were Apart from ‘historic’ studies starting from the 1920s
played with two prototypical types of touch: depressing [1, 2, 3, 4, 5, 6, 7], the more recent literature found dif-
the keys with the finger initially resting on the key surface ferent kinds of noise that emerge when the key is struck
(pressed), and hitting the keys from a certain distance in different ways [8, 9, 10]. The most prominent noise
above (struck). Musicians were asked to identify the type emerges when the hammer hits the strings (hammer–
of touch of the recorded samples, in a first block with all string noise, “attack thumb” [8]). This noise characterises
attack noises before the tone onsets included, in a second the specific sound of the piano, is most prominently au-
block without them. Half of the listeners could correctly dible in the treble strings, but cannot be varied with type
identify significantly more tones than chance in the first of touch independently of hammer velocity.
block (up to 86% accuracy), but no one in block 2. Those When a key was hit from a certain distance above, a
who heard no difference tended to give struck ratings for characteristic finger–key noise was found to occur 20–
louder tones in both blocks. 30 ms before the actual tone (“touch precursor” [8],
“early noise” [9]). This noise was clearly visible in au-
1. Introduction dio wave form plots. Although the authors reported that
listeners could easily distinguish between tones that were
For almost a century, physicists and musicians have been played from above and those played from the keys, no
arguing whether it is only the final hammer velocity that systematic listening test was reported [9].
determines the sound of a piano tone or whether a pi-
Although measurement tools improved since the first
anist can additionally influence the piano timbre by vary-
systematic investigations in the 1920s, no more conclu-
ing the way of touching the keys. Pianists study for many
sive results could be obtained as to whether the touch-
years extremely intensively to advance their technique of
variant noise components (especially finger–key noise)
touching the keys so that the outcoming sound satisfies
can be aurally perceived by listeners not simultaneously
their (and their teachers) high artistic demands. They es-
involved in tone production. This study investigates
tablish and refine various kinds of accelerating the keys
whether musically trained participants in a controlled ex-
in order to obtain finest timbral shades so that it might be
perimental situation are able to distinguish between dif-
hard for them to believe that piano timbre might be ex-
ferent types of touch, even if the finger–key noises are
pressed by a single physical parameter. The physicist, on
removed.
the other hand, argues that the pianist loses control over
the hammer after the jack is escaped by the let-off button.
Therefore, it is only the endmost velocity of the hammer 2. Method
that determines the intensity and thus the timbre of the 2.1. Participants
piano tone.
The two positions are apparently contradictory, but The 22 participants were aged between 23 and 46 years
they are not necessarily. A pianist manipulates naturally with a mean of 31.2 years. All were practising musi-
all expressive parameters simultaneously when playing cians or musically well trained; 13 of them had piano as
a piano, so there might be audible differences in sound main instrument, the others play violin, guitar, violon-
when two tones are played with different types of touch. cello, and clarinet. They had been playing their instru-
However, it is very likely that in such an uncontrolled ments between 8 and 36 years (mean = 21.7); 18 of them
case, more parameters than only final hammer velocity had studied their instrument at a post-secondary level for
have been varied. For scientific investigation, we have to an average period of 8.8 years (2–18 years).
2.2. Stimuli a) WG pressed, mHV = 1.837 m/s

The piano tones were samples recorded on a 173-cm Key velocity track (m/s)
0.4

Amplitude / Key velocity (m/s)

Yamaha grand piano (the same recordings as in [11]). Audio wave form
0.3
The middle C (C4, 261.6 Hz) was played by two pianists
(WG, RB) with two different types of touch: one with 0.2
the finger initially resting on the key surface and pressing
0.1
it down (pressed), and one hitting the key from a certain
distance above (touching it already with a certain speed, 0
struck).1 The hammer movement was monitored with one −0.1
accelerometer mounted at the front end of the hammer
shank and another one at the front side of the key. The −0.2
microphone was placed close to the strings (about 10-cm −0.3
distance), and the digital recordings were sampled mono
at 16 kHz with 16-bit word length (for details, see [11]). −0.4
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Each sample started 250 ms before hammer–string con- b) RB struck, mHV = 1.884 m/s
tact (as measured from the hammer accelerometer) and
lasted for 750 ms, so all possible noises emerging from
0.4

Amplitude / Key velocity (m/s)

hitting the keys were included in the stimuli.
From the recorded samples, we selected 100 isolated 0.3

tones from the two pianists and for the two touch condi- 0.2
tions so that each of the four groups comprised 25 tones
0.1
and case-wise similar maximum hammer velocities. The
hammer velocities ranged from 0.37 to 4.07 m/s with a 0
mean of 1.65 m/s.2 In Figure 1, we show two typical pi-
−0.1
ano tones. Their hammer velocities are almost equal, but
they were played with two types of touch. The pressed −0.2
tone (Fig. 1a) exhibits a gradual increase of key velocity, −0.3
whereas the struck tone (Fig. 1b) shows a very sudden
jerk. Parallel to this first key impulse, there is a clearly −0.4
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
visible (and audible) knock in the audio wave form. Time (s)

2.3. Procedure Figure 1: Audio wave form and key velocity pattern (dot-
The stimuli were presented to the participants via head- ted line) of two piano tones with similar hammer velocity.
phones. A graphical user interface provided play buttons The upper tone was played with the finger initially rest-
and rating radio buttons for all stimuli of a block at once ing on the key surface (a), the other struck from a certain
arranged in random order. The participants had to judge distance above (b).
whether a piano key was originally pressed or struck by
the pianists in a two-alternative forced choice paradigm. They found the test quite difficult throughout. A minor
They could listen to each stimulus as often as wanted and portion of them admitted that they did not hear any dif-
in any order they liked until they were sure about all their ferences and that they had to guess. In contrast to the sec-
judgements. In a first block, they listened to all 100 tones. ond block, where the overall accuracies were throughout
After a short break, they listened to a selection of 48 tones at chance level, the overall recognition rates were much
in which the first 250 ms before hammer–string contact better in the first block: Four of the 22 participants got
were replaced by silence so that all attack noises prior to 80–86% correct, five 70–80%, two 60–70%, and the other
the sound were removed. The task was then the same as 11 rated at chance level.4 Overall, half of the participants
before.3 identified significantly more tones correctly than chance,
the other half did not.
3. Results & Discussion In Figures 2 and 3, the relative frequencies of cor-
rect ratings are plotted separately for touch condition and
The participants needed usually 20 to 25 minutes to ac-
pianist for the two blocks, respectively. The error bars
complish the first block and 5 to 7 minutes for the second.
indicate the range of an insignificantly different accuracy
1 In previous papers, we used instead of “pressed – struck” “legato – according to the χ2 statistic with n = 22 × 25 = 550
staccato,” a terminology introduced by [12]. We changed to the present and p < 0.05. In the first block, participants identified
terminology to avoid confusion with terms referring to articulation. around two thirds of the pressed tones correctly on aver-
2 The standard deviation for each of the 25 hammer-velocity quadru-
ples ranged between 0.005 and 0.237 m/s with a mean at 0.065 m/s. 4 Accuracies between 59.8% and 40.2% are not significantly dif-
3 The entire stimulus material can be accessed at http://www. ferent from chance according to the χ2 statistics with n = 100 and
oefai.at/˜wernerg/pianosound/. p < 0.05.
 Block 1 (with touch noise), n = 22 Block 2 (without touch noise), n = 22
100 100

69.7%
65.2% 73.5%
75 75
Percent Accuracy

Percent Accuracy
70.4%
66.0% 67.3%
54.2%
58.0% 56.4%
50 50
50.7%
41.3% 40.2%

25 25 Block 2
Block 1 Corresp. Block 1
Pianists (n=13) Block 2, Pianists
Non−Pianists (n=9) Block 2, Non−Pianists
0 0
Pr (RB) Pr (WG) St (RB) St (WG) Pr (RB) Pr (WG) St (RB) St (WG)

Figure 2: Relative frequency of correct ratings (accu- Figure 3: Accuracy of all participants for block 2 (with-
racy) of all participants for block 1 (with touch noise), out touch noise, circles with error bars), as in Figure 2.
separately for two touch conditions (struck, “St” and Empty diamonds denote the ratings of the corresponding
pressed, “Pr”) and two pianists (“RB” and “WG”). Er- stimuli from block 1. Triangle plots show ratings for pi-
ror bars denote the range of an insignificant difference anists and other musicians, respectively.
according to the χ2 distribution with p < 0.05. Addition-
ally, triangle plots show the accuracy for pianists (up-
ward pointing triangle) and other musicians (downward 6 pianists and 5 non-pianists. It is evident, that pianists
pointing triangle). and other musicians were able to judge equally well.
From those 11 participants who did not hear any dif-
ference (rated at chance level), 7 rated the stimuli clearly
age. They had the same accuracy for the struck tones, but according to their intensity: The louder a tone the more
only when RB played them; those played by WG were they tended to judge it a struck one. Evidence came from
judged randomly. This result is due to differences in the highly significant (positive) correlation coefficients be-
stimuli played by the two pianists. The finger–key noise tween rating and hammer velocity (0.27 < rpb < 0.78,
was clearly more present in RB’s struck tones than in all p < 0.01).5 This tendency was also found with
those played by WG. It might be that although both pi- some of the good identifiers. From them, four had sig-
anists hit the key surface from a certain distance above, nificant correlation coefficients between 0.24 and 0.32
RB played with a slightly different position of the finger (p < 0.05). This finding suggests that when participants
tip that produced a more prominent impact noise. were unable to identify the actual cue in this listening test
(touch noise), they assigned touch by tone intensity. This
In block 2, they heard a selection of the first-block
is both obvious and interesting. On the one hand, louder
stimuli with the touch noises before the actual tone re-
sounds always involve larger and faster body movements
placed by silence. As Figure 3 shows, the listeners could
(hands, arms, etc.), while softer tones require smaller,
not correctly identify the type of touch anymore. The
more controlled movements. This applies not only for
ratings for pressed tones were not significantly different
the piano, but also for other instruments, as e.g. string
from chance. Those for the struck tones had a small
or percussion instruments. Therefore it is not surprising
but significant tendency to be erroneously perceived as
that some participants connect loud with struck and vice
pressed tones. These results confirm that the cue for dif-
versa. Moreover, a struck touch generates typically loud
ferentiating the two types of touch were the touch noises
and loudest tones, while a pressed touch provides more
before the actual tone.
tone control and is typically applied for soft and softest
It might be assumed that pianists perform in this tones (as reported in [13]). On the other hand, this find-
touch identification task better than other musicians, be- ing provides empirical evidence for a phenomenon that
cause they are especially familiar with piano sounds and might be essential for synthesis of sounds communicat-
different touch conditions. In Figures 2 and 3, the accu- ing body movements in general.
racies are plotted separately for pianists and non-pianists The present results are also essential for the concep-
(upward and downward pointing triangles, respectively). tual design of computer-controlled reproducing pianos.
The differences are small and insignificant. These results They rely entirely on the hammer-velocity-only assump-
are consistent with the overall accuracies we considered tion, measuring and reproducing only the final hammer
before: From the 11 participants who heard significantly
better than chance were 7 pianists and 4 other musicians, 5 Using the point-biserial correlation coefficient between rating (di-
whereas among those eleven who identified nothing were chotomous variable) and hammer velocity (continuous variable).
velocities — up to now with extremely convincing re- References
sults. However, noises different from the human fingers [1] O. Ortmann, The Physical Basis of Piano Touch and Tone.
emerge at machine playback. London, New York: Kegan Paul, Trench, Trubner; J. Cur-
We tested an extreme condition: single tones at a 10- wen; E. P. Dutton, 1925.
cm distance from the strings, a position that even a per-
[2] W. B. White, “The human element in piano tone produc-
former is usually not able to listen from. First, it is in- tion,” Journal of the Acoustical Society of America, vol. 1,
teresting that finger–key noises are so salient in the audio pp. 357–367, 1930.
signal recorded close to the strings (compare findings re-
[3] H. C. Hart, M. W. Fuller, and W. S. Lusby, “A precision
ported in [8]). And second, it is remarkable that the dif-
study of piano touch and tone,” Journal of the Acoustical
ferent touch conditions were not identified more correctly Society of America, vol. 6, pp. 80–94, 1934.
given how salient those noises were in the stimuli. This
might be explained by the absence of other cues, as e.g., [4] C. E. Seashore, “Piano touch,” Scientific Monthly, New
York, vol. 45, pp. 360–365, 1937.
visual or hapto-sensorial information. We expect that es-
pecially hapto-sensorial feedback to the player influences [5] M. Cochran, “Insensitiveness to tone quality,” Australian
considerably his/her aural perception of a played tone, Journal of Psychology, vol. 9, pp. 131–134, 1931.
since this feedback changes substantially with the type [6] J. G. Báron and J. Holló, “Kann die Klangfarbe des
of touch (cf. [14]). However, we expect that the listen- Klaviers durch die Art des Anschlages beeinflußt wer-
ers’ accuracy in this test would improve considerably, if den?” Zeitschrift für Sinnesphysiologie, vol. 66, no. 1/2,
they were instructed to listen to the touch noises and with pp. 23–32, 1935.
training. [7] J. G. Báron, “Physical basis of piano touch,” Journal of
the Acoustical Society of America, vol. 30, no. 2, pp. 151–
4. Conclusion 152, 1958.
[8] A. Askenfelt, “Observations on the transient components
This study confirms that the difference between two of the piano tone,” in Proceedings of the Stockholm Mu-
equally loud piano tones due to type of touch lies in sic Acoustics Conference (SMAC’93), July 28–August 1,
the different noise components involved in the keystroke 1993, A. Friberg, J. Iwarsson, E. V. Jansson, and J. Sund-
[8, 9]. These noise components (i.e. finger–key noise) berg, Eds., vol. 79. Stockholm: Publications issued by
are audible when the key is struck, and absent when it the Royal Swedish Academy of Music, 1994, pp. 297–
is pressed down. This study provides a first systematic 301.
perceptual evaluation of whether musicians can aurally [9] G. W. Koornhof and A. J. van der Walt, “The influence
identify the type of touch that produced an isolated pi- of touch on piano sound,” in Proceedings of the Stock-
ano tone, independently of hammer velocity. Our results holm Music Acoustics Conference (SMAC’93), July 28–
suggest that only some musicians are able to distinguish August 1, 1993, A. Friberg, J. Iwarsson, E. V. Jansson,
between a struck and a pressed touch using the touch and J. Sundberg, Eds. Stockholm: Publications issued
noises as cue, especially the finger–key noise that char- by the Royal Swedish Academy of Music, 1994, vol. 79,
acterises a struck attack, whereas others could not tell pp. 302–308.
any difference. Without those touch noises none of them [10] M. Podlesak and A. R. Lee, “Dispersion of waves in pi-
could tell a difference anymore. When they could not ano strings,” Journal of the Acoustical Society of America,
hear the touch differences, they tend to rate louder tones vol. 83, no. 1, pp. 305–317, 1988.
as being struck, and soft tones as being pressed. We can [11] W. Goebl and R. Bresin, “Measurement and reproduction
only speculate about how the present findings generalise accuracy of computer-controlled grand pianos,” Journal
to a real-world concert situation (including pedals, rever- of the Acoustical Society of America, vol. 114, no. 4, pp.
beration, reflections, and the listener at a certain distance 2273–2283, 2003.
away from the piano). In the light of the present results, [12] A. Askenfelt and E. V. Jansson, “From touch to string vi-
we consider the pure aural effect of touch noises (exclud- brations. II. The motion of the key and hammer,” Journal
ing visual and other cues) a rather small one. of the Acoustical Society of America, vol. 90, no. 5, pp.
2383–2393, 1991.
Acknowledgments [13] W. Goebl, R. Bresin, and A. Galembo, “The piano action
This research was supported by a START Research Prize by as the performer’s interface: Timing properties, dynamic
the Austrian Federal Government, administered by the Austrian behaviour, and the performer’s possibilities,” in Pro-
Fonds zur Förderung der Wissenschaftlichen Forschung (FWF, ceedings of the Stockholm Music Acoustics Conference
project No. Y99–INF) and by the European Union (Marie (SMAC’03), August 6–9, 2003, R. Bresin, Ed. Stock-
Curie Fellowship, HPMT-GH-00-00119-02, the Sounding Ob- holm, Sweden: Department of Speech, Music, and Hear-
ject project (SOb), IST-2000-25287, and the MOSART IHP net- ing, Royal Institute of Technology, 2003, vol. 1, pp. 159–
work, HPRN-CT-2000-00115). The Austrian Research Institute 162.
for Artificial Intelligence acknowledges basic financial support [14] A. Askenfelt, A. Galembo, and L. L. Cuddy, “On the
by the Austrian Federal Ministry for Education, Science, and acoustics and psychology of piano touch and tone,” Jour-
Culture, and the Austrian Federal Ministry for Transport, In- nal of the Acoustical Society of America, vol. 103, no. 5
novation and Technology. Many thanks to Simon Dixon for Pt. 2, p. 2873, 1998.
helpful comments and to all participants.