Proceedings of the Stockholm Music Acoustics Conference, August 6-9, 2003 (SMAC 03), Stockholm, Sweden

THE 'VIRTUAL FLUTE': ACOUSTIC MODELLING

AT THE SERVICE OF PLAYERS AND COMPOSERS
Andrew Botros, John Smith, Joe Wolfe
School of Physics
University of New South Wales, Sydney
J.Wolfe@unsw.edu.au www.phys.unsw.edu.au/music

ABSTRACT In The Virtual Flute, an expert system relates features of

impedance spectra Z(f) to the perceived musical behaviour of
Among the difficulties facing woodwind players are (i) the instrument. A waveguide model, developed and tested on a
awkward fingerings for rapid passages, (ii) intonation defects large database of measurements, predicts the Z(f) for every
in instruments, and (iii) performing exotic effects such as possible fingering, determines the pitch and 'playability' of
microtones and multiphonics. In many cases these may be all possible notes and lists possible multiphonic
ameliorated by alternative fingerings. combinations of notes. A musician-friendly interface allows
We report the construction and use of a web service for musicians to search the resulting database according to
the flute. An expert system was trained by an experienced appropriate criteria and constraints.
flutist to determine playability from features of 957 minima The Virtual Flute is currently used several hundreds of
in measured acoustic impedance spectra Z(f) for 76 selected times per day. Here we choose a small number of problems and
fingerings. Measurements on successively more complicated their solutions to illustrate some effects of acoustical and
acoustic systems yielded an accurate waveguide model of Z(f) biomechanical interest and some ways in which the musical
of the flute, which generated the minima of the 39,744 possibilities of the instrument have been extended.
different acoustic configurations of the flute. The expert
system, coupled with the waveguide model, produces a large 1.1. Alternative fingerings
database of alternative fingerings, microtone and multiphonic
All instruments are imperfect compromises. When a good
fingerings. The database is accessible to flutists and
player rehearses a single phrase many times, s/he may be
composers via a user-friendly interface that includes user
seeking ways to correct the pitch of one note, to control an
determined constraints on key combinations and ranks
inappropriate loudness or stability, or just finding ways to
fingerings by playability or pitch. This service, located at
execute or to avoid awkward finger movements. A particular
http://www.phys.unsw.edu.au/music/flute is used hundreds of
combination of keys depressed is called a fingering. Most
times per day. We report some of the service's uses and
players have a repertoire of alternatives, which are used in fast
discoveries.
passages or to produce more appropriate pitch and timbre in
1. INTRODUCTION different circumstances.
For the flute, there are either 13,248 or 26,496 different
One of the applied aims of music acoustics is to use the fingerings, depending on whether the lowest note on the
scientific understanding of music and instruments to help instrument is B3 or C4. We know of no previous attempt to
musicians. The Virtual Flute (TVF) is one such application make a complete study.
[1]: it is a widely used web service that supplies flutists and
composers with solutions to a wide range of technical 1.2. Microtones, timbres, multiphonics
problems. These include awkward transitions between notes, Solo and chamber music since the 1950s increasingly calls on
intonation and timbre difficulties. It also gives fingerings woodwind players to play notes that fall between those of
(key combinations) for microtones (non standard pitches) and standard temperaments, notes with varying timbres and
multiphonics (chords). We describe briefly the strategies and chords [2,3]. Only a subset of the possible combinations of
techniques used to construct TVF. We also report examples of notes are playable on the flute and we know of no previous
its use. The interest in such examples is that they illustrate attempt to list them all. Dick [4] gives an extensive
the sometimes subtle difficulties faced by musicians and the collection of fingerings for these exotic effects but it is far
steps that may be taken to remove them. These in turn from complete.
illustrate some of the details that must be addressed by The two lowest registers of the flute are usually played
scientists aiming to help them. with simple fingerings: nearly all of the tone holes
downstream from a particular point are open, while all of

SMAC-1
 Proceedings of the Stockholm Music Acoustics Conference, August 6-9, 2003 (SMAC 03), Stockholm, Sweden

those upstream are closed, except for register holes. In cross The quaternary playability ratings (0-3) were used as
fingerings, some tone holes downstream from the first open inputs to Cubist, the continuous version of C5.0. Its output
hole are closed. This often increases the end effect and so gives playability as a function of the parameters listed above,
flattens the note, creating the possibility of a microtone. The but again it is not very informative because of the correlation
inertance at the tone hole affects higher resonances more among the parameters.
strongly, so the resonances cease to be harmonic. Higher To obtain playing pitch from the frequency of an
resonances thus contribute less to the vibration regime, impedance minimum measured or calculated at low temperature
which creates the possibility of a darker timbre. See [5] for a and humidity is complicated in principle, but here is
detailed discussion. performed by a simple empirical relation determined by
Further, the impedance mismatch at a single open hole comparing played frequencies with those of the minima. An
may partially reflect a travelling wave. The transmitted average embouchure effect is already included in the
portion may be reflected at a subsequent open hole. Thus a measurements [8]. This method introduces errors, but they are
cross fingering can be considered as a set of resonators with not greater than the variations among players and
different frequencies. In general they are not in a simple instruments.
harmonic series, which creates the possibility of the One dimensional waveguide models to calculate Z(f) have
superposition of two or more standing waves to produce a been made by various authors [9,10]. The parameters for our
multiphonic. model were determined by measuring successively more
complicated combinations of bores, finger holes and keys. It
2. STRATEGY AND IMPLEMENTATION was then tested on the database of measurements [8,11]. Z(f)
was then calculated for all 39,744 fingerings, and the
Because they are open to the air at the embouchure, flutes
frequency, magnitude, bandwidth and harmonicity of all
operate near minima in the acoustic impedance spectrum Z(f).
minima were determined. From these, the expert system
The ease of playing the note associated with a minimum in
constructed a database of about 150,000 possible playing
Z (f) depends on properties of that and other minima. In
regimes. Inharmonically related minima are used to construct
principle, one might hope to understand this in terms of
lists of multiphonic possibilities.
properties of the jet-bore interaction and a knowledge of the
The web service itself was constructed according to
extent to which players control the jet. This is a difficult
recommended principles [12,13], and the user interface
question. Fortunately, flutists know how hard it is to play a
designed after consideration of the needs and knowledge of
note. So we asked an expert flutist to attempt to play a note
flutists and composers, the principal users.
corresponding to each of 957 minima identified on the
measured Z (f) data. When the note was playable, it was 2.1. The web service
assigned a playability from 1 to 3, with 0 for unplayable
notes. Three tools are offered. One allows the user to enter a
We expected that the playability of a note associated fingering via a graphical interface and then returns the
with a particular minimum might depend on its depth, its predicted playable notes with their predicted playability and
bandwidth, the proximity and magnitude of nearby extrema, pitches and the multiphonic possibilities. The second allows
and whether or not higher minima were harmonically related input of a note name to search the database for all suitable
to it. These were quantified and used as inputs. Three methods fingerings and ranks them by intonation or playability,
were tried to determine playability from these input data. which are included in the displayed output. This tool also
Neural networks were unacceptably slow, even when the input allows the user to constrain the search by excluding (or
parameter set was reduced. Linear regression was unhelpful and including) any keys that would inconvenient to use (or not to
uniformative, partly because of strong correlation among the use, respectively), in the circumstance. The third tool invites
parameters. A decision tree method, using the C5.0 algorithm input of two or three notes and searches the multiphonic
suite, an artificial intelligence technique developed by database.
Quinlan [6,7], proved successful. The first tree used only the
3. APPLICATIONS AND SIGNIFICANCE
binary data, 'playable' or 'unplayable'. Training the tree on
subsets of the data and testing on others enabled the rejection
3.1. Awkward passages and trills
of unimportant input parameters. The final tree ranks a
minimum as unplayable (i) if Z > 1.35 MΩ or (ii) if 1.35 MΩ Flutists write fingerings using the numbers 1,2,3 for the keys
> Z > 0.68 MΩ and if the next higher minimum is more than usually operated by the long fingers of each hand (left given
35% lower, or (iii) if 0.68 MΩ > Z and if the next higher first), Th for the left thumb key, and then individual names for
minimum is within 261 Hz. (1 Ω = 1 Pa.s.m-3) the other keys. A vertical line separates the two hands. For

SMAC-2
 Proceedings of the Stockholm Music Acoustics Conference, August 6-9, 2003 (SMAC 03), Stockholm, Sweden

instance, the standard fingering for F6 is written Th 1 - 3 |

1 – – D#, with the D# operated by the RH fourth finger.
A trill between the note F6 and A6 is rather awkward
using the standard fingerings. The fingering for A6 is
Th - 2 - | 1 - - D#. (The reader can appreciate this by
attempting a rapid alternation between 1 – 3 and – 2 – with
the left hand.) The flutist needing to accomplish this trill
sought an alternative fingering for F6, specifying that all the
keys used for A6 be included (this ensures that all fingers will
move together on the trill). TVF returns
Th 1 2 3|1 - - tr2 D#, which gives a comfortable trill in
which three fingers move in the same sense.
In trills and rapid passages, less than perfect intonation
may be tolerated: if a note that lasts 0.1 s, its frequency
cannot be resolved to much better than ±10 Hz .

3.2. Stable transitions

In rapid transitions, there is insufficient time to adjust the

embouchure optimally for each note: for a rapid trill, no
attempt is made, and for a rapid passage, the embouchure can
only change to follow the overall shape of the phrase. A rapid
alteration over a large interval can lead to the possibility of
'splitting': the production of an unwanted transient between
desired notes. (The possibility of splitting is almost as bad as
the effect itself: lack of confidence that the notes will sound
properly distracts musicians from interpretive and other
issues.)
Flutist Jane Cavanagh reports this example: Stravinsky's
Firebird requires an acciacatura from B5 to E6 to B5. Figure 1
shows why this is difficult. The standard fingering for B5
(also used for B4) will comfortably play F#6, the third
harmonic of B4. That for E6 will also play G#5 and A6. For a
slow transition, the flutist would shorten and/or quicken the
jet to play the E6. How much? Too much to play G#5 but not
enough to play A6. S/he would then lengthen and/or slow it
enough to play the B5. To play this acciacatura (or a trill),
there is no time to adjust embouchure: one simply forms an
intermediate embouchure and lets the flute fingering select the
note. The danger of the split here is that, if the embouchure
compromise favours the high note, one risks playing B5-E6-
F#6-B6, because the transient unwanted F#6 (the 'split') is
only a tone above the higher target. If the compromise
embouchure is too low, one will play B5-G#5-B5, or B5-
nothing-B5, because the E6 is not a particularly stable note.
This player requested an alternative E6 fingering and
found Th 1 2 - |1 - - D# tr1 tr2. The note is stable and
easy to sound, although slightly flat. According to TVF (and Figure 1: Z(f) for the standard fingererings for (top) E6 and B5
players), the minimum immediately below that which (middle), and for the alternate E6 fingering
supports E6 (Fig 1c) is unplayable, so there is little danger of
the E6 'dropping down'. Consequently, the embouchure may 3.3. Awkward high passages
be compromised more towards B5 and less to E6, which
High passages are awkward for several reasons. First, the
minimises the chance of sounding F#6. It is a comfortable,
resonances of all wind instruments are inherently weaker
safe solution.

SMAC-3
 Proceedings of the Stockholm Music Acoustics Conference, August 6-9, 2003 (SMAC 03), Stockholm, Sweden

(extrema in Z(f) are less pronounced) because of visco-thermal

damping in the bore. In the flute, this effect is exacerbated by 3.6. Further work
the Helmholtz resonance in the head [11,14]. Second, the Application to other instruments has begun. The clarinet has
individual fingerings are complicated and smooth transitions a geometry that is almost as standardised as that of the flute's
between them more so. Simple fingerings do not work but it is complicated by having very many more fingering
because either the end effects in the array of open tone holes possibilities. And the bassoonist among the authors
put them out of tune, or because the wave propagates too far earnestly desires an application to the instrument in which
into the array. Indeed, for notes above the tone hole cut off alternative fingerings are most used—and needed.
frequency, all notes are overtones of the whole length of the
pipe, and several open holes along the length act merely as Acknowledgments. We thank Jane Cavanagh, our expert
register holes or tuning perturbations [11,14]. Third, the flutist, John Tann for technical assistance, and the Australian
resonances are closer together, so the danger of undesirable Research Council for support.
transients, like that discussed above, is greater. Some of the
standard fingerings are difficult to play softly, so players seek 4. REFERENCES
stable soft fingerings (effectively, those with deeper minima 1. Botros, A., Smith, J. and Wolfe, J., "The virtual Boehm
in Z(f)). They also seek sequences that are less awkward. The flute—a web service that predicts multiphonics,
job is even harder when composers call for microtones. microtones and alternative fingerings", Acoustics
Kathleen Gallagher, who specialises in the contemporary Australia, 30: 61-65, 2002.
repertoire, cites two examples. The British composer Chris
2. L. Berio, Sequenza, Suvini Zerboni, Milan, 1958,
Dench, in Closing Lemma, writes a final flourish of high
notes landing on a sharp E7. This pitch is well above the tone 3. B. Bartolozzi, New Sounds for Woodwind, Oxford Univ.
hole cut off frequency, and well into the 'shorting' range of the Press, London 1967.
Helmholtz resonance, so easily played fingerings are rare. 4. Dick, R. The Other Flute, Multiple Breath Music, 1989,
Further, the composer wants a microtone. TVF obliges with 5. Fletcher, N. H., "Mode locking in nonlinearly excited
12 - | - - 3 D# tr2. inharmonic musical oscillators", J. Acoust. Soc. Am. 64:
Welsh composer Richard Barrett, in What Remains for 1566-1569, 1978,
flute, bass clarinet and piano calls for a slurred, fast passage
6. Quinlan, J. R., C4.5: Programs for Machine Learning,
sharp C7, sharp B6, E7, flat D7, flat E7, D7, Eb7, Bb6. This
Morgan Kaufman, San Mateo, 1993.
player uses standard fingerings for Bb6, D7 and flat D7, but
found the rest of the fingerings from TVF, thereby creating a 7. Quinlan, J. R. "C5.0: An Informal Tutorial", Rulequest
solution to a bar that has produced anxiety and performance Research. http://www.rulequest.com/see5-unix.html
approximations for many flutists attempting this work. 2002.
8. Wolfe, J., Smith, J., Tann, J. and Fletcher, N.H.,
3.4. High notes and soft high notes "Acoustic impedance of classical and modern flutes"
Journal of Sound and Vibration, 243:127-144, 2001.
The range of the flute ends somewhere near the middle of the
fourth octave, and notes near the end of the range are often 9. Plitnik, G. R. and Strong, W. J., "Numerical method for
difficult to play softly, and sometimes difficult to play in calculating input impedance of an oboe", J. Acoust. Soc.
tune. So flutists are keen to seek improvements: minima in Am. 65: 816-825, 1979.
Z(f) that are deeper and closer to the right frequency. TVF 10. Caussé, R., Kergomard, J., and Lurton, X., "Input
offers a fingering for F7 whose minimum is about 30% lower impedance of brass musical instruments - comparison
than that of the standard fingering, and which allows even between experiment and numerical models", J. Acoust.
flutists of modest ability to play this very high note, and Soc. Am. 75: 241-254, 1984.
even play it softly. 11. Music Acoustics. www.phys.unsw.edu.au/music
3.5. Multiphonics 12. Greenspun, P., Philip and Alex"s Guide to Web
Publishing, Morgan Kaufman, San Francisco 1999.
Multiphonics are rare in the low range of wind instruments
13. Nielsen, J., Designing Web Usability, New Riders,
because an open tone hole acts as a shunt. In the high
Indianopolis, 2000.
registers they abound. Composers and flutists routinely use
The Virtual Flute to find multiphonics and to find how to play 14. Wolfe, J. and Smith, J. "Cut off frequencies and cross
them. fingering in baroque, classical and modern flutes",
Journal of the Acoustical Society of America, accepted
for publication.

SMAC-4