A NON-LINEAR SOUND SYNTHESIS METHOD

Jnatas Manzolli
1,2
,Furio Damiani
1,3
, Peter Jrgen Tatsch
1,3
, Adolfo

Maia Jr.
1,4
1
Interdisciplinary Nucleus for Sound Studies
2
Fine Art Institute, Music Department
3
DSIF/School of Electrical and Computer Engineering
4
Applied Mathematics Department
400 Albert Einstein Av., 13081-970
UNICAMP University of Campinas, SP
BRAZIL
[email protected], [email protected], [email protected], [email protected]
ABSTRACT
This paper presents a new method for sound synthesis based upon non-linear dynamics.
The procedure is compact and can be interactively controlled in real-time. The sounds
thus synthesized have dynamic characteristics and rich spectra. It is an economical
method, for it uses only a cyclic buffer to control the audio output and a non-linear map
to generate the waveforms.
INTRODUCTION
This method, FracWave, uses a parametric control of non-linear maps for sound synthesis
(1,2)
. It
explores the structural control of non-linear dynamics (NLD) in a computer-based environment.
It applies recursive sonic constructs, integrating the macro and micro levels of the sound
synthesis process, thus enabling a fast generative method.
NLD in the acoustic domain has been investigated by Lauterborn
[3]
, has been used for
micro-structural control of sounds and connected with Granular Synthesis
(4,5)
.
In the domain of the synthesis of sounds using digital simulation of acoustic instrument
models, Lindeman
[6]
describes a physical model for a woodwind instrument reed, showing that a
simple model could exhibit sonic responses varying from harmonic to chaotic.
NLD and physical modelling belong to two different classes of digital sound synthesis: the
first uses abstract algorithms and the second makes use of traditional musical instruments
models. The latter generally expends much calculation time to generate the output; the first is
dependent on the algorithm: if it is less calculation intensive, it can be used for real time sound
production, as for instance, the FM
(7)
.
It is necessary to formulate new models for timbral design, taking advantage of the new and
cheap breeds of Digital Signal Processing (DSP) hardware and of new user interfaces
(8)
enabling
real-time gestural control of synthesis parameters.
1. OVERVIEW
Waveforms of natural sounds and acoustic instruments are time dependent
(9)
. FracWave uses
numerical sequences produced by simple non-linear equations to create continuously
transforming waveforms. It samples a circular buffer in which numerical sequences are stored.
This buffer is a dynamic lookup table, which we will call dynamic wavetable, or DW.
To produce complex sounds, but also to create degrees of sound complexity, a linear
interpolation after a sub-sampling process was applied to foster the control of the frequency
spectra. This has to be done, because if one would directly use the sequence of numbers
produced by the non-linear map, the sound produced would always possesses too rich a spectra,
with intense energy in high frequency components.
One can find similarities between FracWave, the Karplus-Strong Algorithm
(10)
, and the
LASy Technique
(11)
. A diagram with the basic blocks of the FracWave process is shown in Fig 1.
2. NUMERICAL ORBITS AS WAVEFORMS
A DW stores the values produced by a process in which a non-linear map is recursively
iterated. The basic idea is, given a starting point x
0
and a non-linear map F(x
k
), a sequence of
values x
k,
also called a numerical orbit, is obtained by the expression x
k+1
= F(x
k
).
A sequence x
k,
generated by the non-linear map, is mapped into the interval [-1,1]; then it is
followed by a linear interpolation, generating a new sequence of samples x
n
. To obtain suitable
numerical orbits, we studied a number of non-linear maps, extending those found in the NLD
literature, such as the Logistic and Henon maps or the Barry Martin algorithm
(12)
.
Fig. 1 FracWave structure diagram.
Calling K the total number of points in the non-linear map and M the number of points in
the DW, they should satisfy the relation M > K. We have been using 2 100. Thus, the
y
m
(sound output)
N points interpolation
j
n
(reading increment
M points
Non-
linear
mapwith
y
k
x
k
h interpolated
points are inserted
betweenx
k
and
x
i
m
(writing increment
sub-sampling rate F
sub
on the non-linear map, to obtain a CD quality waveform with F
s
=44,100
Hz in the sound output will be given by F
sub
=44,100 Hz/.
The set of equations used in this work are listed below:
x
k+1
= y
k
sgn(x
k
)*sqr(abs(B* x
k
-C)) Eq. 1
y
k+1
= A x
k
x
k+1
= y
k
sgn(x
k
) + sqr(abs(B* x
k
-C)) Eq. 2
y
k+1
= A x
k
x
k+1
= y
k
sin(abs(B* x
k
-C)) Eq. 3
y
k+1
= A x
k
x
k+1
= y
k
sin(B* x
k
-C) Eq. 4
y
k+1
= A x
k
where A, B and C are parameters and all equations start with x
0
=0 and y
0
=0. The equations
above can be described by the following general formula:
x
k+1
= y
k
perturbation(B,C) Eq. 5
y
k+1
= A x
k
Equations 1 to 4 define two-dimensional maps in which a set of parameters {A, B, C} and
initial conditions {x
0
, y
0
} control a delay factor and a perturbation produced by a non-linear term
as described in Equation 5.
The orbit phase space {x
k
,y
k
} representation shows behaviours varying from quasi-periodic
to chaotic. This will give out timbres with a defined pitch and quasi-periodic spectral (small)
variations: these are natural sound characteristics.
To analyze the non-linear maps one can use the phase space plot x
k
versus y
k.
On such
plots, clusters of points mean that there is an attractor; cluster distribution regularity identifies
quasi-periodic numerical behavior; irregularity typifies a chaotic one. Plots showing the phase
space behaviour can be seen in Figs. 2 and 3.
Fig.2 Eq. 1 non-linear map phase space diagram, with A =- 0.54 , B = - 3.9, C = 1.4.
Fig. 3 Eq. 3 phase space non-linear map diagram, with A =- 1.4, B =- 3.9, C =1.4
2.1 DW CONTROLLERS
The reading and writing increments, j
n
and i
m
, are defined by:
j
n
= n rmod(M) with n =1,2,..N Eq. 6
i
m
= round(m ) mod(M) with m=1,2M Eq. 7
where
= M(Fr/Fs)
= M(Fw/Fs)
and
N is the number of interpolated x
k
values from the non-linear map, M is the number of
samples in the DW (M=1024, M=2048 or M=4096), Fr, Fw and Fs are the reading, writing
and sampling frequencies (in Hz) respectively. The operator rmod gives the remainder of a
floating point division and the operator mod gives the remainder of an integer division;
round returns the closest integer.
2.2 INTERPOLATION AND SUB-SAMPLING
Linear interpolation is applied to smooth the sequence x
k
generated by the non-linear map and to
improve the precision of the reading increment. Given two consecutive values, x
k
and x
k+1
, they
will be inserted into the DW at constant sub-sampling intervals h, given by:
h = int (F
s
/F
sub
), with F
sub
<<F
s
Eq. 8
where F
sub
is the sub-sampling frequency; the operator int truncates a number, returning an
integer value.
The algorithm for the linear interpolation and sub-sampling is defined by:
FracWaveInter(n, k, h )
When (n mod h =0) make
begin
x
k
= x
k+1
x
k+1
= f(x
k+1
)
x
k+1
= g(x
k+1
, min, max)
end
Return [(x
k+1
- x
k
)*n)/h + x
k
]
Note that the function f(x
k+1
) returns a new value produced by the non-linear map iteration and
the function g(x
k+1
, min, max) maps the value x
k+1
between [-1, 1] linearly. The interval [min,
max] can be found experimentally, applying a large number of iterations over f(x
k
) and verifying
the upper and lower limits of the resultant attractor in the phase-space horizontal axis.
Fig. 4 Diagram of the Linear interpolation applied to smooth the sequence generated by the non-
linear map. In this example, values from the vertical axis Y
k
are used to draw the Waveform.
Note that it is the same procedure for the horizontal axis X
k
.
The reading increment j
n
defined in Eq. 6 above, a floating-point value, is used to interpolate
two consecutive values of the DW. The procedure below performs it:
FracWaveInter(n, k, h )
A1 =int(j
n
)
A2 =(A1+1) mod M
Return W(A1) + frac(j
n
) (W(A2) W(A1))
where W(...) is a DW and the function frac returns the fractional part of a number.
(Xo,Yo)
(X
1
,Y
1
)
(X
2
,Y
2
) (X
3
,Y
3
)
h h h h
Phase Space Interpolated Waveform
Sub-sampled values
The following steps are used to start FracWave sound generation and subsequently iterations:
STEP 0
fill the DW with x
k
without interpolation
STEP 1
dynamically refresh the DW using the increments i
n
to write values and j
m
to read
them
subsequent call of the FracWaveInter( ) and FracWaveInter( ) functions
STEP 2
Repeat STEP 1 until the end
3. GESTURAL CONTROL
Due to its compactness, FracWave is well suited for live performances. A number of user
interfaces can be used to control the parameters. We used a mouse on a Microsoft Windows
environment GUI to enter the parameters A, B and C. We are developing, a small portable
gestural interface that is worn on a finger by the performer and gives out a temporal sequence of
the finger spatial coordinates
(8)
to establish the values of the relationship F
r
/F
w
in real-time.
To produce the necessary parametric changes we used a mathematical tool, which we
developed, called Sound Functors
(13)
. For this application, curves generated with the gestural
interface are used to transform sound parameters.
Fig. 5. General idea of the gestural control applied based on a vectorial field to be applied in
FracWave
3.1 IMPLEMENTATION
A sequence of points (x
p
,y
p
) generated by the gestural interface, is used as a control device as
showed in Figure 5. We take the sequence of points (x
p
, y
p
) to be equivalent to (F
r
(p),F
w
(p)). In
this way we define a discrete curve D in the plane F
r
F
w
, where F
r
and F
w
are the reading and
(x
p
, y
p
)
Vectorial Field
(F
r
(p),F
w
(p))
Change values of
and defined in Eqs.
6 and 7
Control Space
FracWave Parmeter Changes
writing frequencies of FracWave. Let v
p
=(x
p
, y
p
) the vector defining the position of a point on
D. Now suppose we have a vector field u defined on D. We can assume, without loss of
generality, u to be a normalized vector field, that is, ||u||=1.
At each point v
p
=(x
p
, y
p
) we have a corresponding vector u
p
=(a
p
, b
p
). A deformed curve D
is then defined by a new position vector
v
p
=v
p
+
p
u
p
Eq. 9
where
p
is a real number which measures the strength of the deformation.
In order to generate variations of an initial curve D we must define the deformation vector
field u and strength parameter . Taken the points (x
p-1
, y
p-1
), (x
p
, y
p
) and (x
p+1
, y
p+1
), we
define an approximated tangent vector field on D as follows
t
p
=(x
p-1
-

x
p+1
, y
p-1
- y
p+1
) Eq. 10
In extension, a normal vector field on D can be defined as
n
p
=(-y
p-1
+y
p+1
,-x
p-1
+x
p+1
) Eq. 11
and we further define a normalized normal field as
u
p
=n
p
/||n||. Eq. 12
As discussed above, we are interested to produce small deformations in the original D curve.
We can restrict our parameter to a continuous, bounded function defined as follows:

p
= (x
p
, y
p
) =1/1+sqr(x
p
2
+y
p
2
) Eq. 13
In Non-linear sound synthesis in real-time, this is particularly interesting because the
performer can draw any curve, listen to the result and interact with the curve. It is similar to play
a musical instrument, but in this case the performer controls a synthesis method.
After obtaining an interesting result, the performer may want to re-produce a family of sounds
with the same behavior. If the curve D is complex, it is difficult to re-draw the sequence of
points.
Having this in mind, we developed a model with the following features:
an intuitive control curve D drawn with the portable gestural interface
a family of control curves D( ).
CONCLUSION
The aim of this investigation was to produce sounds with a dynamic and rich spectra. We
obtained interesting sounds with colourful timbre. It is possible to imagine a new generation of
sound synthesis devices using gestural control to produce sounds based upon non-linear
dynamics.
This investigation provides a theoretical framework, which expands the techniques for
digital sound synthesis. Instead of imitating known sonic behaviour, the goal is to produce
unique sounds. The user experiences a deep interaction with sound material, an immersion on
soundscapes produced by his own intuition. It is necessary to formulate new models for timbral
design, for there is a continuing need of alternative sound sources for Art Music.
ACKNOWLEDGMENTS
This project was realised through the support of the Foundation for Research of So Paulo State
(FAPESP) gave to the Gesture Interface Laboratory in which FracWave was developed.
REFERENCES
[1] J . Manzolli, Musical application derived from FracWave sound synthesis method, Preprints
of the AES Convention, Berlin, 1993.
[2] F.Damiani, J .Manzolli and P. J .Tatsch, A non-linear algorithm for the design and production
of digitally synthesized sounds, ICMP'99 Proceedings, Campinas, Brazil, 1999.
[3] W. Lauterborn and U. Parlitz, Methods of Chaos physics and their applications to acoustics,
J ournal of the Acoustic Society of America, 84(6):1975-1993, 1988
[4] B. Truax, Chaotic non-linear systems and digital synthesis: an exploratory study,
Proceedings of the 1990 International Computer Music Conference, 1990
[5] A. Di Scipio, Composition by exploring of non-linear dynamic systems, Proceedings of the
1990 International Computer Music Conference, 1990
[6] E. Lindeman, Routes to chaos in a non-linear musical instrument model, Proceedings of the
1998 Audio Engineering Society Convention, 1988
[7] J . M. Chowning, The synthesis of complex audio spectra by means of frequency modulation,
J ournal of the Audio Engineering Society, 21(7):526-534, 1979
[8] F.Damiani and G.Mendes, Interface gestual ultrassnica, IV Iberchip, Mar del Plata, Ar,
1988
[9] M. V. Mathews, and J . C. Risset, Analysis of musical-instrument tones, Physics Today
22(2):23-30, 1969
[10] Karplus and Strong, Digital synthesis of plucked -string and drum timbres, Computer Music
J ournal 7(2):43-51, 1983
[11] Chareyron, Digital sysnthesis of self-modifying waveforms by means of linear automata,
Computer Music J ournal 14(4):25-41, 1990
[12] K. Becker and M. Dorfler, Dynamical systems and fractals, Cambridge University Press.
1989.
[13] J . Manzolli and A. Maia J r., Sound Functors Applications, Proceedings of the V Brazilian
Simposium on Computer Music, Belo Horizonte, Brazil, 1998.