Sound in a nutshell:

Granular Synthesis
An overview of granular synthesis and the
techniques involved.

Written by Timothy Opie

If you have any queries regarding this article, would like
more information, notice anything that needs to be changed,
added, or modified, or just wish to contact me then please email
me:

tim_opie@yahoo.com

BA Honours (Music)
Supervisors:
Terry McDermott
Graeme Leak
Preface

This thesis is designed to give an overview of where granular synthesis

currently stands. It is an area which is still not well known. It is even omitted in
some recent computer music publications. I had personally heard very little
about granular synthesis 12 months ago (late 1998). Although I was first
introduced to the concept by Larry Polansky in 1996, when he gave a guest
lecture at La trobe University. It is hoped that after reading this thesis you will be
able to go and use these methods in compositions, with a full understanding of
what you are doing.

The thesis begins by going through all of the details and parameters that
form a grain. It then looks at how the grains are structured together as textures
and the control methods involved. It goes through the history of the
conceptualisation of granular synthesis, and how sound has come to be
perceived. The thesis considers some examples of compositions written using
granular synthesis methods. It finishes off by looking at areas which still need to
be explored, and gives some new ideas about the direction research in granular
synthesis could go.

Acknowledgements

I would especially like to thank Terry McDermott for all of his help and the
patience he displayed whilst I was still trying to learn the basics of an area
which has now become dear to me.

I would also like to thank Graeme Leak for his fun and practical ideas
about composition. I would like to thank David Hirst for getting me onto the road
of computer music technology in the first place.

Other people I would like to thank:

All the staff in the La Trobe University Music Department for their support
and willingness to keep on going, despite what is happening to the department -
namely the closure of the La Trobe Music Department.
 Nicole Bone for lending me computer equipment, Blake Stickland for
helping me find references, Marc Lawless for grammatical advice, and Wendy
Jones for her encouragement and proofreading. Lastly I would like to thank my
family.

Statement of Authorship

I certify that this thesis is my original work. No other person's work has
been used without due acknowledgement. This material has not been
presented by me for examination in any other course or subject at this or any
other Institution. Signed in the original document by Timothy Opie on 12
November 1999.

© Copyright 12 November 1999 Timothy Opie

This work was submitted to La Trobe University as part of my Honours Degree
In Bachelor of Arts.
Contents:

0.0 Sound in a nutshell: Granular Synthesis (Title page)

1.0 Introduction

2.0 The Grain

2.1 The Contents
2.2 The Envelope
2.2.1 Duration
2.2.2 Amplitude
2.2.3 Attack time
2.2.4 Sustain time
2.2.5 Decay time
2.2.6 Shape
2.3 Synchronicity of granular synthesis
2.3.1 Synchronous granular synthesis
2.3.2 Asynchronous granular synthesis
2.4 Controlling textures of grains
2.5 Streams
2.6 Grain Order
2.7 A look at a multidimensional texture
2.8 Summary of the Grain

3.0 A short history of granular synthesis

3.1 Isaac Beeckman
3.2 Changing Perceptions
3.3 Norbert Wiener
3.4 Jean Fourier
3.5 Dennis Gabor
3.5.1 Gabor's research on hearing
3.5.2 Gabor's machines for time distortion
 3.6 Springer
3.7 Martin Bastiaans
3.8 Iannis Xenakis

4.0 Granular synthesis in Composition

4.1 Iannis Xenakis
4.1.1 Analysis of Analogique A-B
4.2 Curtis Roads
4.3 Paul Lansky
4.4 Barry Truax
4.4.1 Analysis of Riverrun

5.0 Structuring the future

5.1 Software
5.2 Extending the use of granular synthesis

6.0 Conclusion

7.0 References

Appendix 1

Appendix 2

Appendix 3

Appendix 4

Appendix 5
1.0 Introduction

The technique of granular synthesis is becoming more common as a tool

in computer composition, yet it could be seen as more than just a basic tool.
The development in granular synthesis has now reached a point at which it is
possible, and relatively common, to create entire compositions using only this
technique.
Granular synthesis involves generating thousands of tiny sonic events
known as grains. These grains can be considered the building blocks of sound,
or atomic sound particles. They can be combined to generate new sounds,
timbres, or sonic textures. Granular synthesis is unique because it collapses the
time and frequency domains within the concept of the grain. This collapse in
domain allows the limits of sonic perception to be exploited.
The textures that can be created using granular synthesis are very
exciting as they allow for a soundscape that is totally controllable by the
composer. Not only is the composer in control of each timbre, each duration,
and each pitch within the piece, but also how these parts interact with each
other. This advantage of complete controllability of all aspects of the texture is
carried out at the macro, as well as the micro level. Unfortunately such a large
amount of control makes it difficult to keep control of all the levels. This is
overcome by using statistical forms of control.
Computer synthesis has been common for a long time now. There are
many forms of computer synthesis including additive synthesis, subtractive
synthesis, distortion synthesis, frequency modulation, amplitude modulation,
formant (FOF) synthesis, dynamic stochastic synthesis, graphic synthesis,
interpolation synthesis, table-lookup synthesis, time-varying wave form
synthesis, waveshaping synthesis, and window function synthesis (Roads
1996a). Granular synthesis could be seen as a form of additive synthesis but its
approach and sonic results are quite different from additive synthesis.
Until very recently the techniques and means associated with granular
synthesis were not available to many people. This was due to the nature of
granular synthesis, in that it can contain literally thousands of parameterized
events just to specify one second of sound. Processing such large amounts of
data required the work of mainframe computers equipped with a digital to
analog converter (DAC). For many years Bell Telephone Laboratory in the USA
was the only place capable of any type of sound synthesis (Roads 1996a: 87).
Later there were facilities available at IRCAM in France (Risset 1985:11),
University of California (Roads 1996a: 169), and Massachusetts Institute of
Technology (Roads 1996a: 547).
Even with such large computers the processing could take a week just to
get basic results. Until about 1990 the most prominent people involved with
granular synthesis were Iannis Xenakis, Curtis Roads, Barry Truax, Douglas
Jones and Thomas Parks. Xenakis actually completed his granular composition
without the aid of a computer. This was achieved by splicing and rearranging
magnetic audio tape, an even more difficult process (Roads 1996a: 169). Curtis
Roads spent 10 years on a number of mainframe computers creating a 10
minute composition called nscor (Roads 1985: 165-167).
Computers have come a long way since 1990. They can process data at
extremely high speeds. They are all equipped to perform sound synthesis
techniques including sampling synthesis. They are small enough to use in the
home. This step forward in computer technology has meant that granular
synthesis can now be realised everywhere, including some commercial synths.
As people come to learn about granular synthesis techniques there is a growing
understanding and popularity of it's usefulness.
2. The Grain
Like a nut, which has two basic elements, the shell, and the flesh
contained within, a grain has an envelope which contains the actual sonic
content. These two parts or parameters:

contents

envelope

make up an entire sonic grain. The sonic grain itself has a very short duration
and as a single entity would seem very insignificant, but once the grain
becomes part of a granular population, these two parameters make a big
difference to the sound. Let us begin by looking at the contents.

2.1 The Contents

The contents of the grain can be derived from any type of sound wave.
The purest sound wave is the sine wave. The sine wave though, is not a natural
wave, and it can only be produced synthetically. Barry Truax described it in
such a way:

If you remember the first time you heard a sine wave oscillator your ears
should have told you this is not a good sound, this is not an interesting musical
sound. (Iwatake 1991)[1]

The purity of the sine wave may sound unnatural, but it means it is
predictable when manipulating and distorting. Sine waves are also useful
because they are very simple and easy to generate. They have just three
parameters: amplitude, frequency, and phase, which means they can be
controlled with ease. Sine waves can be precisely added together through the
process of additive synthesis to create new signals. The additive process is
fundamental to granular synthesis.

The main problem with sine waves is that they theoretically have an
infinite duration.[2] The parameters of a sine wave allow the user to determine
the frequency, and the number of oscillations per second, but not how many
seconds to actually play the sine wave for (See figure 2.1). Using granular
synthesis the dimension of duration is added to the sine wave. By truncating the
sine wave it can technically no longer be called a sine wave. For better
understanding it may be called a sine grain (See figure 2.2).

ad infinitum.

Time -->

Fig 2.1: A portion of a sine wave

Fig 2.2: A sine grain (Dodge & Jerse 1997: 263)

Granular synthesis is not limited to the support of additive synthesis

techniques. Other forms of synthesis can also be easily incorporated into the
process. The contents of the grain can for example be comprised of signals
from any form of distortion synthesis such as FM synthesis. The contents could
also consist of white noise which is generated using a random signal generator.

Sound samples are used often in granular synthesis as grain contents.

Before the technique of granular synthesis was devised there was a technique
by which sound samples were broken into acoustical quanta as a means to
stretch sound samples.[3] This time stretching method involved breaking the
sound sample into grain sized acoustical quanta and then arranging them by
overlapping quanta to condense the sample, or by repeating quanta to stretch
the sound sample. This method alters the duration without altering the sense of
pitch. It can also be used to alter the sense of pitch without altering the duration
of the sample (Roads 1996a: 440-442).

There are two sorts of sampled sound in granular synthesis. They are
short samples, and long samples. That may seem rhetorical but they are used
differently. Short samples that have the same duration as a single grain, can be
used in their entirety as the grain contents. A longer sound sample can be used
in a series of grains, although not necessarily in a particular order. The principle
of time-stretching has been carried over to granular synthesis which involves a
very deterministic progression of the sampled grains, but there are also many
other ways to use sampled sounds. These include using some grains from a
sound sample in the same way one would use a sine grain. For example you
might use just three small sections from a large sound sample as the content
source.

A sound sample is created using an analogue to digital conversion (ADC)

process.[4] It can be represented graphically as such: (see figure 2.3).
A
M
P
L
I
T
U
D
E
FREQUENCY

Fig 2.3: A digitally encoded sound sample

A
M
P
L
I
T
U
D
E
FREQUENCY

Fig 2.4: The same sound sample after it has been formed into a grai

There are many different ways to create the contents for a grain. As more
content ideas are introduced, the variety of granular textures increases. Looking
at the sound sample after it has been turned into a grain it is obvious that the
shape of the sound sample has changed. This is due to the introduction of the
envelope, which will be discussed next.

[1] Toru Iwatake interviewed Barry Truax on August 7, 1991 at the Department
of Communication, Simon Fraser University.
[2] See chapter 3.4 for a details on Fourier frequency transform.
[3] Acoustical quanta is a term used by Gabor to describe the grains of sound
he created in his experiments with communication and signal reproduction.
These are also referred to as Gabor grains. See chapter 3.5 on Dennis Gabor.
[4] See appendix 1.

2.2 The Envelope

The sound envelope determines the duration, and the amplitude of the grain.
The process of adding a grain envelope is known as windowing. Windowing is a
type of amplitude modulation technique. The idea of windowing comes from an
analysis process where small amounts of sound information can be put into a
window frame, making it easier to analyse the sound. The idea of windowing
also has a more closely related link to granular synthesis because of the
practical nature in which it was first used.1 When the grain content is windowed
it imposes an amplitude change over the content as was seen in figure 2.2 and
2.4. The amplitude follows the shape of the envelope. Figure 2.5 shows the
most basic way to create a sonic grain.

The envelope
generador

The content

Figure 2.5 A method for producing a sound grain (Dodge & Jerse 1997: 263).
 This method involves multiplying the amplitude of the contents with the
amplitude of the envelope

Windowing may cause some slight convolution to the grain, but as a

sonic grain is so small it is not noticeable.2 If the grain is constantly repeated in
a synchronous manner the convolution will become noticeable. This suggests
that the parameters of the envelope can be changed to alter the spectral
composition of the grain and give extra control over the grain.

There are certain parameters that define an envelope. These are:

Duration
Amplitude
Attack time
Sustain time
Decay time
These parameters can be summarised using the multidimensional
'parameter':
Shape

Fig 2.4 Diagram of a simple envelope.

2.2.1 Duration

The duration of the envelope determines the entire duration of the grain.
Experimentation has shown that durations of between 10 milliseconds and 50
milliseconds are the best to work with.3 Obviously this is an extremely short
duration, therefore only the most salient features would be immediately
obvious.The duration determines the distinction of the contents of the grain.
Grains with a shorter duration have a more ambiguous spectrum. This is due to
the time-frequency relationship of sound. The shorter the duration, the less the
spectrum is defined. A short grain duration creates a broader bandwidth making
the grain noisier. This is very useful for creating a larger spread of frequencies.
Another feature of using a short duration is that it is more difficult for the ear to
decode the contents. This allows the composer to deceive the listeners
perception of sound. An isolated grain with a very short duration produces a
kind of buzzing or clicking sound. A longer duration causes less convolution in
the contents of the grain. An isolated grain with a longer duration sounds more
like when a woodblock is tapped on. Different durations become much more
significant when there is more than one grain involved, as discussed later.

2.2.2 Amplitude

The amplitude refers to the peak amplitude. It determines the height of

the envelope. This parameter may be applied directly by specifying the
amplitude, or it may be implied indirectly by the other parameters. It is also
influenced by the contents of the grain. The amplitude of the envelope dictates
whether the amplitude of the contents is able to reach full capacity or not. The
envelope amplitude as such, can be worked out as a percentage, 100% being
loudest.
2.2.3 Attack time

The attack time determines how quickly the envelope reaches full
amplitude. The attack is a necessary rounding tool. If a sound is immediately
played at full amplitude, then there will be a loud crackling noise at the
beginning. This occurs when a sound changes instantly from silence to full
amplitude and causes extra distortion, which is usually not the desired result.
Any sharp angles in the envelope will cause strong side bands in the spectrum.
The side band is caused by convolution between the envelope shape and the
contents of the envelope. Having no envelope is just like a vertical attack
gradient, with essentially no duration. As stated previously there is a
relationship between time and frequency. With an extremely short, or zero
attack time, the resultant frequency would contain an extremely large
bandwidth, resulting in the audible clicking noise. If the clicking sound is
undesired then a significant attack time must be applied.

2.2.4 Sustain time

The sustain time determines how long the envelop stays at peak amplitude. This
factor is also very important in determining the signal that is produced. For example if
the grain has a long duration, but the sustain is only very short, with a very steep attack
and release, such as with exponential curves, then it would treat the grain as if it had a
short duration resulting in a buzzing or clicking sound in isolated instances. Using a
shorter sustain is better suited to percussive type sounds. If the grain has a long sustain,
the content becomes more significant allowing the content to play a more active role in
the sonic effect. The sustain is not usually set as a parameter, it is normally a
consequence of the overall duration of the envelope less the attack and decay time.

2.2.5 Decay time

The decay time determines how quickly the envelope drops back to zero
amplitude. Different gradients of release will produce faster or slower cut off
times. The release does not have to be at the same rate as the attack. When
working with synchronous granular synthesis a different release time to the
attack time will impose a different spectrum on the resultant sound. This is
caused by convolution between the envelope and the contents. In an isolated
grain the effect is not really noticeable. In order to percieve the resultant sound
it must be repeated at a continuous rate, which is why it has less effect on
asynchronous granular synthesis.

1In the earlier experiments of Dennis Gabor, he used a small glass window with
the sides slightly shaded. The window was used in a film soundtrack (which is
visual) where light passes through the soundtrack and is then converted to
sound. See Chapter 3.5.2 for more information. (Gabor 1946: 446)
2See appendix 2 for more information about convolution.
3Gabor was the first to start experimenting and he came to this conclusion
which has been reinforced by each successive exploration into granular
synthesis. This will be dealt with in later chapters, starting with chapter 3.

2.2.6 Shape

As already suggested the envelopes do not need to be symmetrical. An

isosceles triangle will give a different sonic result to that of a scalene triangle.
The envelopes do not need to be linear. A more rounded envelope will result in
less disturbance of the spectrum of the grain's content. The most common
envelope shape used in granular synthesis is the Gaussian curve1(Gabor 1946:
435). It looks like a bell and is sometimes referred to as a bell shaped curve.

With all the parameters available for the envelope it is possible to create
hundreds of different types of envelopes. These could be categorised into three
basic shapes.
Straight-lined
Curved
 Complex
It must be noted that the shape is only a graphic representation of the function
of the envelope.

2.2.6.1 Straight-lined
The straight-lined envelopes consist mainly of triangular and trapezium
shaped envelopes. The Trapezium is the most common type of envelope
because it is simple to make and has all the parameters that are needed. An
attack, sustain and release. A triangular shape is even simpler to make,
however, many applications require a sustain. Linear shapes are useful
because they can be calculated more quickly, and achieve their purpose
efficiently. (See fig 2.5)

Amplitude

Time

Amplitude

Time

Fig 2.5: Straight-lined envelopes - Triangular and Trapezium shaped.

2.2.6.2 Curved

Curved envelopes consist of the probability curves such as Gaussian,

and Hanning, and also a large range of exponential shaped curves. These are
more useful because they have more control over each part of the grain. FOF
grains for example are designed by setting two exponential curves over a sine
wave. One for the attack, and one for the decay. Because curves can be
gradually sloped they are generally used to reduce spectral change induced by
windowing. They can also be used for very steep inclines and declines, which is
useful for introducing a large spectral change. (See fig 2.6)

Amplitude

Time

Amplitude

Time

Fig 2.6: Curved shaped envelopes: Gaussian (bell) and exponential.

2.2.6.3 Complex

'Complex' Envelopes are still very new to the idea of granular synthesis.
They consist of two envelopes. The first envelope is a sinusoidal type wave.
Over the sinusoidal wave is placed another wave, any type of linear or
exponential wave is adequate.

Amplitude

Time

Fig 2.7: Complex envelope shape

This type of envelope has, until recently, only been used with an analysis
technique called Wavelet Transform (WT). This analysis process was
developed by scientists at the University of Marseille in 1988 for physics and
acoustic applications. (Boyer & Kronland-Martinet 1989: 51) The wavelet is very
similar to a grain, but it's primary service has been analysis of magnitude of
frequency, and phase. This analysis technique is used by windowing a sound
sample with wavelets. (Roads 1996a: 581). Gordon Monro a mathematician at
Sydney University has used Wavelets as a compositional tool. He suggested
that this knowledge could be put into practice with granular synthesis. (Monro
1996) At the moment though it lies as a vast new area awaiting in-depth
research.

1A probability curve was suggested by Dennis Gabor, but others have referred
to it specifically as a gaussian probability curve. Whilst Gabor did not say
gaussian, his mathematics and diagrams imply gaussian distribution. (Roads&
Alexander 1996b: 3)
2.3 Synchronicity of granular synthesis

There are two distinct ways of using granular synthesis. These methods are
quite different from each another and produce different textural effects.

They are:
Synchronous granular synthesis
Asynchronous granular synthesi

2.3.1 Synchronous granular synthesis

The spacing within the grains in granular synthesis can radically change the
texture that is created. With synchronous granular synthesis the grains are all
seperated by the same amount of time, or with some linear relationship. (Roads
1996a: 175).

Synchronous granular synthesis is very closely related to Formant Synthesis,

more commonly known as FOF synthesis.1 FOF synthesis uses the addition of
sinusoidal (FOF) grains to create FOF textures. It is considered basic
synchronous granular synthesis, because of it's use of FOF grains which are
similar to sine grains. The main difference though is the way they are used.
FOF grains have very strict parameters. They use only sinusoidal contents, and
the envelopes have a very strict structure. It is the parameters of the envelope
which change the sound created, in conjunction with the FOF grain density.
This method is very measured and predictive, therefore desired results are
easier to achieve. A change in grain density affects the sense of pitch when
using FOF grains (Roads 1996a: 299). This rule also applies when using
synchronous granular synthesis, although it is not as predictive, due to the
variety of grains that can be produced.

Synchronous granular synthesis uses a number of control parameters which

are:

Amplitude
Grain Density
 Pitch
Inter-onset time
also attack and decay times.

2.3.1.1 Amplitude
The Amplitude controls the overall amplitude of the texture. This, like the grain
envelope, can be set up as an envelope over the duration of the texture.

2.3.1.2 Grain Density

The Grain Density is the single most important factor in creating granular
textures. Initially, it determines how densely the grains are distributed within the
texture, but the density affects a few different parts of the texture. Grain density
is measured as the number of grains per second. There is really no limit on
grain density, although the denser it is, the longer it takes to process the sound.
There are many side effects of grain density which will increase as the density
does. The Grain density affects the timbre, the amplitude and the pitch of the
texture. A sparse density of less than 30 grains per second will create a
rhythmic effect. As the density increases the distinction of any rhythm will be
lost and will give way to a pitch sensation determined by the grains. The higher
the density the greater the spectral complexity of the timbre. Denser textures
will yield more partials, formants and side bands in the texture making them
richer. The amplitude will increase as the density increases. The pitch will rise
as the density rises.

There is a direct relationship between grain density and grain duration. The
longer the grains are, the more they will overlap making the texture denser at
lower grain density rates.

2.3.1.3 Pitch
The sense of pitch can be set from the texture level. This can be done by
playing the grains at different sampling rates. Normally the sense of pitch would
be changed at the grain level, but it is possible to control it from a higher level
as part of a change in frequency. As mentioned above the sense of pitch can be
increased by increasing the grain density.

2.3.1.4 Inter-onset time

This function is associated very closely with the grain density and the
synchronicity. It could be considered an obsolete parameter, because it can be
fully determined by the density and the synchronicity. It is used mainly as a
measurement, rather than a parameter. The inter-onset time is the duration
between two adjacent grains.

2.3.2 Asynchronous granular synthesis

With asynchronous granular synthesis the grains are added to the texture
without the strict linear relationship of synchronous granular synthesis. The
relationship will contain random elements or there will be a very indistinct
relationship. Asynchronous textures can be created by randomly spreading
grains across the texture without worrying about timing.

The control parameters in asynchronous granular synthesis are:

Amplitude
Grain Density
Pitch
Inter-onset time

2.3.2.1 Amplitude
The amplitude functions in exactly the same way as with synchronous granular
synthesis.
2.3.2.2 Grain density

As with synchronous granular synthesis, a very low density will cause a

rhythmic effect. Once the density reaches a certain level the rhythmic effect will
become a pitched effect. The pitch does not change as the density increases.
The density is simply a parameter within which the grains are dispersed. For
example if the density is specified at 200 g/s then 200 grains will be randomly
(or statistically) distributed throughout that second.
Barry Truax used densities between 1-2300 for his composition called Riverrun
(Truax 1990: 120).

2.3.2.3 Pitch

Initially this is the same as synchronous granular synthesis except that it plays a
less noticeable role and it is not affected by density.

2.3.2.4 Inter-onset time

This is a measurement of the duration between the beginnings of adjacent

grains and as such would change constantly. Asynchronicity can be achieved
by setting up a random or statistical inter-onset parameter.

1 FOF is an abbreviation coming from the French term fonction d'onde

formantique. (Roads 1996a: 299)

2.4 Controlling textures of grains

When granular synthesis is heard it becomes a texture rather than a series of
grains (Zadorin 1997). This is a phenomenon that humans have experienced in
many different ways since the beginning of time. A prime example of this is a
pine tree. The pine needles are very small. When looking at a pine tree, all that
is really visualised is a green mass, or on the ground all that is seen is a golden
mass. The needles are still there, and they can be explored as such, but they
create a texture which is more relevant to the human senses. Whilst one can
analyse the grains in a granulated sound, the impression one receives is the
texture.

This idea of a controllable texture is a new way of considering music. Granular

synthesis provides the means to precisely control multidimensional parameters,
and therefore control the entire texture. It is unique in the way it creates a bridge
between the micro and macro level. The micro-grain is a trivial insignificant
sound object, but with the implementation of higher level structures it becomes
a sound texture on the macro level (Truax 1990: 123). The micro level consists
of lower control levels, the contents and the envelope of an individual grain. The
macro level of control deals with higher level parameters to control the flow of
grains, and to give the grains some kind of cohesion within the texture.

A number of control parameters have already been discussed, that is, the micro
level controllers, but there are still more parameters to consider. It is apparent
that, whilst it is good to be able to control so much information, it is becoming
increasingly more difficult because of the sheer number of events, especially if
there are over a thousand grains being processed per second.

One way to overcome this problem is to use algorithms to control different

parameters. This could be achieved by setting an upper limit and a lower limit
and commanding the computer to select random elements within the prescribed
field.
Another method used is statistical determination algorithms. The most common
algorithm in granular synthesis has been stochastic determination.1 Iannis
Xenakis spoke about this method in great detail in his book Formalized Music
(Xenakis 1971). The very first granular synthesis composition Analogique A-
B written by Xenakis in 1958-1959 uses this method. This composition will be
discussed in more detail in chapter 4. The fundamental principle behind
stochastic determination is an asymptotic evolution towards some
2
goal (Xenakis 1971: 4). The goal is reached through an iterative process,
which will finally stabilise around a certain point. The teleological nature of
stochastic determination makes it useful for giving granular textures some
direction or point.

Another distribution method is Poisson distribution.3 The fundamental principle

behind this method is that it is not a continuous distribution, but it is limited
within certain fields. Truax devised a computer program POD that uses this
method of distribution. POD has been updated many times as computer
technology has improved. POD uses an idea of tendency masks. Using the
theory of Poisson distribution being limited between different values, the
tendency mask is made of a lower limit and and upper limit. This can be set up
as a graphical representation of the limits (Truax 1988: 19). The tendency
masks consist of the different parameters that will be using the Poisson
distribution method. Figure 2.8 shows the Tendency mask used in Truax's piece
'The wings of Nike'.
 Fig 2.8: Tendency mask for Wings of Nike (Truax 1990: 128)

Tendency masks are useful because they are simply understood and much
more compact than a list of numbers. POD allows the parameters of the
tendency mask to be manipulated in a real-time environment by pressing set
keys on the keyboard whilst the mask is being synthesised (Dodge & Jerse
1997: 275). This allows the composer to react to the texture produced. The
piece Riverrun which is analysed in chapter 4, makes use of tendency masks.

2.5 Streams
So far all the methods involve a single group of grains. These populations of
grains have a single identity, in that they create a single layered texture.

Streams are an important addition if multiple layered textures are desired.

Streams are a way of labelling the different groups of grains. They can be used
to manage multiple sets of grain populations. (McAdams & Bregman 1979: 26)

Streams can be achieved by recording through different channels and then

mixing down together, however there is less interaction between the streams.
Streams can usually be easily followed in granular synthesis. They can be
thought of as the harmonies of the texture. Streams usually have some kind of
relationship, for example, stream1 could consist of grains created with sound
sample1, and stream2 could consist of grains created with sound sample2. See
figure 2.9.

^
^
Y

X >>

Fig 2.9: Two streams. It has two layers within the one texture.

1Stochastic determination is closely related to the Markov chain, created by

Markov, a Russian mathematician. It was used mainly as a valid linguistics tool
until Noam Chomsky proved otherwise. That which is not possible in language
though can still be realised in music. (Miller 1997, 8)
2The term stochastic is derived from the Greek word stóchos meaning an aim
or guess.
3Developed by the French mathematician Dennis Poisson, 1781-1840.
2.6 Grain Order

A very important factor in making granular textures is the order in which the
grains are placed. When dealing with sound samples, and the samples need to
be kept more or less in their original form, the grains must be placed in that
manner, running from the beginning to the end of the sample. Another method
is to randomly scatter the grains in an attempt to create some totally new
texture. This is another area where statistical distribution can play an important
role.

2.7 A look at a multidimensional texture

The following diagram (fig 2.10) contains a multidimensional texture that I have
created for analysis. It contains a number of different granular sounds combined
into one multidimensional texture. It is comprised of just 3 types of grain
contents:
Sine grain
Complex signal
white noise
There are just 4 types of envelopes:
Hanning envelope
Exponential impulse
Exponential impulse with a sustain.
The complex signal is also used as a grain.

Apart from the grain functions the texture is created by changing the parameters
of the texture, such as the density, the duration of the grains, the frequency and
the amplitude.
These parameters are all assigned different algorithms to determine how the
parameter conditions are met. The algorithms consist of different ramps,
envelope shapes, random numbers and complex signals.
Varying these parameters provided the entire texture.

Freque
ncy

(20-
22,000
Herz)

Time (0-295 seconds)

Fig 2.10: 'Where the Wind Meets the Sea'. Composed by Timothy Opie. On
Track 2 of the accompanying CD, or play below. Spectral graph made in
Cooledit96 for PC.

See appendix 4 for a picture of the original idea, and the Csound files that made
the texture. The Csound scores are detailed in the types of functions and tables
used.

The composition starts off with rectangular shapes in which random numbers
are distributed. The amplitude fades in and out, creating a high pitched whisper.

The lower section rumbles.

The first two streams of the composition contain a changing variety of sonic
effects. Other streams join in as the grain density increases. The grain duration
gets shorter, and the random pitches become more varied resulting in a large
spectrum where all streams meet. This quickly contracts to an extremely small
spectrum, of one stream. It then explores different shapes, and frequencies,
splitting again into two streams.

The main focus of the composition is on grain densities, grain durations, and
frequency. Each envelope is tried with each grain content providing a range of
different sounds. As can be seen on the spectral graph, very distinct shapes are
created within the boundaries of the mask (See Appendix 4). The mask is not
as detailed as those created by Truax, but it does show how masks can be
used.1

2.8 Summary of the Grain

It has been demonstrated that the grain consists of an envelope and contents.
Of course these things are inseparable. If one part is taken away it is no longer
a grain. The control methods involved in the contents and the grain have been
discussed. The control methods of the texture have also been discussed. The
problem of having too much control, and statistical ways to get around this
problem have been explored. As has the use of multiple layers in granular
textures. It is now time to look at the events that took place to allow the
conception of granular synthesis.

1If you look at the Csound score in appendix 4, you can go through it to see
what parameters where used where. It gives a basic idea of the range of
sounds that can be created in a texture with only a few grain sources.

3. A short history of granular synthesis

Granular synthesis might be thought of as a recent development in sound

synthesis, but it can be seen as a reflection of ideas about the nature of sound.
It is only since the advent of computer and sound recording technology that it
has been easy to apply this idea. Granular synthesis is based on a certain
notion of what sound is, in particular, the human perception or cognition of
sound. In order to be able to better understand granular synthesis, one must be
able to see how the theory of sound perception was developed.

3.1 Isaac Beeckman

Isaac Beeckman was from the Netherlands and lived from 1588 until 1637. He
was a scientist who had received no formal training and was never published
during his lifetime. However he was prolific in his scientific endeavours, which
also included investigations into sound. (Cohen 1984: 116-118)

Beeckman's studies in sound led him to the belief that sound is produced by a
vibrating source. He stated that the vibrating causes corpuscles to be cut into
the air. Beeckman never calls the corpuscles atoms, although he was a believer
of an atomic theory in which everything is made up of discrete particles. He
conducted an experiment that is identical to one performed by Galileo in 1638,
one year after Beeckman's death.1 Beeckman showed that rubbing the rim of a
glass filled with water caused globules to form in the water and the air where
the glass was being rubbed. These globules or sound particles he observed
were actually standing waves within the glass, but observing the results of this
experiment it is easy to see how they can be seen as particles. About a century
before but unbeknown to Beeckman, Leonardo da Vinci theorised that sound
travelled in waves. Galileo, Mersenne, Boyle and Newton studied this theory
intensely during the 1600's. Their findings became known as the wave theory.
The wave theory can be seen visually by placing a struck tuning fork, into water
and examining the waves that are formed. The waves travel through the air in
the same fashion. However Beeckman rejected any possibility of a wave theory.
Whilst rejecting this idea Beeckman did believe in sound being transmitted
through a process of expanding and contracting of particle density through the
air. This can be explained as part of the wave theory, but not in the context to
which Beeckman places it. (Cohen 1984: 121-123)

Whilst Beeckman is not wholly correct in all of his assumptions, he does make
some very useful observations. The most important is how the globules are
composed and how they function. Essentially theglobules are a grain of sonic
data. They contain corpuscles of sound. Beeckman also noticed that they are
not necessarily continuous. During certain phases the density would vary,
fewer globules would be dispersed. Even though the corpuscular theory
became a part of the study on light with the insight of Newton, it was all but
forgotten in terms of sound.

3.2 Changing Perceptions

In the 17th century Antonie van Leeuwenhoek developed high-powered lenses

which when used in his microscope was able to magnify up to 200 times without
distorting the image (Taylor et al 1983: 166). Direct observation through the
microscope showed that

a drop of pond water was a teeming mass of life suggestive of a crowded

city (Wiener 1964: 540).

This discovery introduced a whole new world to our own world, and led people
to believe there must be many things too small to see. This was the beginning
of change in the perception of the world. It led to question about how small
objects can be. This discovery led to many discoveries in medicine, and
eventually to light, matter, and the atomic theory.

In 1900 the study of quantum Mechanics began. Max Planck argued that light is
not continuous, it has a granular texture. This theory was more easily accepted
due to the widespread acceptance of the atomic theory. Einstein carried on
Planck's work. He showed that light is quantised, it carries specific quantities of
energy. These packets of energy became known as quanta, the light particles
within as photons. One man in particular, Norbert Wiener, pointed out the
relationship between his quantum physics studies and the characteristics of
sound.
1 The experiment conducted by Galileo was designed to prove a different point.
His experiment was used to show how sound travels in waves. He filled a glass
with water and rubbed his wet finger across the top so that the glass would
resonate. The resonance causes waves to form in the water. Galileo used this
as proof of the wave motion of sound.
3.3 Norbert Wiener

In 1925 Norbert Wiener gave a talk in Goettingen. His talk was on quantum
physics, but he decided to use music as the principle upon which to base it.
Wiener gave detail on the relationship between time and frequency. He
proposed that to get a precise time would mean vagueness in pitch, and vice
versa. This is because whilst a note may be played for a certain amount of time,
the pitch produced relies on a different time line. He proposed that if a 20Hz
note is played for less than 1/20th of a second it will produce no sound, just a
small puff of air. You can't play a jig on the lowest register of an organ. Things
can only get so small before they cannot exist. This argument was devised
mainly to disprove the notion that a better microscope would mean being able to
see still smaller objects within smaller objects 'ad infinitum'. (Wiener 1964: 544-
546) This theory though has many more applications to do with the aesthetic
nature of granular synthesis. Of course at this stage the term was not coined,
but this idea has since been introduced into the theory of granular synthesis.
One cannot have a grain that is smaller than the frequency it should contain.
The size of the grains and the number of grains would also determine what type
of sound could be produced.

3.4 Jean Fourier

At this same time Jean Fourier was working on different ways to define sound.
He designed a method whereby any periodic signal could be reduced to sine
waves and analysed as such. He examined the relationship between sound and
pure sine waves which can be defined as three variables for each sine wave:
frequency, amplitude and phase (relative to the fundamental wave). He
theorised that any sound (assuming it had an infinite duration) could be
explained, or analysed in a series of sine waves of infinite duration. The sine
wave must have an infinite duration because it is an oscillation that can never
vary.
Fig 3.1: The composite wave (c), can be broken up into the sine waves (a) and
(b), which can now be defined with just the three variables (although the phase
is zero in this case. (Dodge & Jerse, Computer Music, page 47)

By this definition of sound it was inferred that the converse was also true, that
any sound could be made up with the addition of various sine waves. This
would enable the reproduction of any sound or timbre (Dodge & Jerse 1985: 47-
49). The result would be a composite wave of a particular timbre over an infinite
duration.

3.5 Dennis Gabor

In 1946 the Physicist and Nobel-prize winner Dennis Gabor came up with a
system that used a grain system to reproduce sound. He documents his
findings and the mathematics to support it in an article called the "Theory of
Communication", which was followed up a year later with "Acoustical Quanta
and the Theory of Hearing". In this he discusses the problems with Fourier
analysis. Gabor says that whilst the mathematics are absolutely correct, it
doesn't allow for physical applications, especially common sounds such as a
siren, which has a variable frequency. It can only be used for creating one
sound at a time. The other problem is that in using sine waves it supposes that
the duration of the signal is infinite. (Gabor 1946: 431) Gabor presents the idea
of applying quantum physics methods to the sound signal. This would enable
the sound to be viewed as quanta. He proposed how the Fourier method could
be reduced into cells in a mathematical context. (Gabor 1946: 436)

3.5.1 Gabor's research on hearing

Gabor also gives an analysis of hearing. In his analysis he looks at the

threshold of the areas of discrimination. In doing this he examines how long it
takes to recognise pitches at different frequencies. For this analysis he used the
findings of two groups of scientists firstly Buerck, Kotowski, and Lichte and later
Shower and Biddulph. The results of the analysis indicated a number of things:

Frequencies between 500Hz and 1000Hz need to be played for a duration of at

least 10 milliseconds before they are discernible as tones rather than noise.

An analysis of frequency showed that if a frequency changes, then it takes more

than twice as long to discern the change, than it did to recognize the first
frequency. This duration in recognising frequency changes is dependent on the
frequency. At lower frequencies a human can discern quicker.

An analysis of change in amplitude showed that if an amplitude was changed it

would take a duration of at least 21ms to discern the change.
Gabor went into the mathematics behind the results of the experiments and
concluded the threshold area is determined by the frequency width of the first
signal and the duration of the second.(Gabor 1946: 442). Gabor analysed these
results further and discussed how it could be possible to deceive the ear. He
suggests that the results of the experiments could be more useful with a curved
change in variables, rather than an angular ramp. He also suggested that using
fixed time-constants would also aid in the deception of the ear. (Gabor 1946:
444)

Gabor said that the ear is remarkable and the only way to perform tests properly
is to do them directly and then make decisions based on the results. He used
his analyses to develop a number of machines that utilised the stated ideas
about sound quanta, duration, and frequencies. Gabor investigated practical
ways of expanding and compressing time scales in recorded sounds

3.5.2 Gabor's machines for time distortion

 Fig 3.2: sound film projector converted into an experimental frequency
converter. (Gabor 1946: 44)

Gabor built machines using a compression-expansion idea for frequency

conversion. His machines operated with mechanically moving parts. He called
them the kinematical method of frequency conversion.1The second type of
system he proposed did not require any mechanically moving parts. He called it
the electrical method. (Gabor 1946: 445)

Gabor never actually made an electrical method frequency converter, but he

suggested the mathematics behind the idea. His main reasoning being that the
waves can be added together without using any mechanical means. (Gabor
1946: 454)

The kinematical method was developed using a film projector (see figure 3.2)
(Gabor 1946: 452). Gabor ran a sound film through the projector. The film would
move at a constant velocity past a slit through which light would be projected to
a photocell. The photocell would convert the light to a sound signal. Gabor
decided to replace the light slit with a rotating drum that contained many slits.
He added a window between the sound film and the photocell. He decided to
add more slits to the drum so that the next slit would start going past the
window just as the previous slit left the window. Gabor realised that this would
produce a loud crack at the start of every cycle over the window, so he graded
the window so that it was clear in the middle but darker to opaque on the sides
so that the frequency would fade in and out without any clicking. (Gabor 1946:
446) See figure 3.3.
 Fig 3.3 kinematical frequency converter with sound film.(Gabor 1946: 446)

The film would pass over the spinning drum at different speeds. Using this
method Gabor was able to adjust the pitch of the sound. Each time the slit
passed over the window it produced just a small segment of sound. The
segments were then added together to reproduce the source signal at a
different frequency. Gabor noted that for some frequencies he got a nearly
faithful reproduction, yet for others he got strong beats. (Gabor 1946: 447) This
becomes very clear with a graphic analysis. See figure 3.4.
 Fig 3.4: A diagram of the sound segments separate and added(Gabor 1946:
447)

Gabor discusses many things that could be changed to improve the quality of
the pitch being reproduced. These deal with things such as the speed the drum
is spinning, the number of slits and the rate of the film that goes through the
projector. Gabor discovered that the number of slits makes the biggest change.
If there are too many slits the reproduction will be of lower quality and very
noisy. If there are too few slits the reproduction will be very good in some
places, yet non-existent in other places. Gabor stated it works best if the next
slit starts when the previous slit is half way across the window. (Gabor 1946:
448)

Gabor made another similar machine in which he adapted a 16mm camera to

rotate the drum at different speeds. This produced different sized grains. He
determined that the grains need to be between 20 and 50 ms.

Gabor also made another such machine this time using a tape recorder. He
designed it so that the tape was in a big loop. After reading each section of the
tape it would be erased and something new could be recorded before that
section of the tape got read again. This way it could be used infinitely without
running out of film, and it could be constantly updated. The tape recorder read
the tape using a pick-up coil that Gabor inserted inside a drum. The drum
consisted of a non-magnetic material with iron rods embedded into it. The pick-
up coil could only read the magnetic tape when the iron rod was between the
tape and the pick-up. The drum was covered with an oil film so that there would
be reduced friction and scraping from contact with the tape. (Gabor 1946: 453)
See figure below.

Fig 3.5: frequency converter with magnetic tape (Gabor 1946: 453).

3.6 Springer

The German company Springer made a device called the Tempophon. The
device was similar to that designed by Gabor (Figure 3.5). This time/pitch
changing device was installed in a number of analogue electronic music studios
around Germany (Roads 1996a: 441). It was the first time granular methods
were used in a commercial environment. In 1963 Herbert Eimert used the
Tempophon for an electronic music composition called Epitaph für Aikichi
Kuboyama (Wergo 60014). There were a number of similar machines built after
the Springer Tempophon (Roads 1996a: 441).

1Kinematics is the branch of physics that deals with pure motion, that is, without
reference to mass or the causes of motion.

3.7 Martin Bastiaans

In the 1980's Bastiaans went through all the mathematics from the two articles
that Gabor published in the 1940's. He verified that the mathematics Gabor
used had been correct. Bastiaans discussed Gabor's sliding window analysis of
a short-time Fourier analysis. He reduced this idea to a function with just two
variables - a discrete time index and a continuous frequency variable. These
two variables form a rectangular lattice. The rectangular cell produced by the
window can only focus on one variable at a time, i.e. The finer the sampling
frequency the courser the sampling in time and vice versa. (Bastiaans 1985:
873)

Fig 3.6 The first diagram shows time as the dominant domain, in which the
frequency becomes coarser. The second shows frequency as the dominant
domain, in which the time becomes much coarser.
3.8 Iannis Xenakis

In the 1960's, before Gabor's theory had been verified by Bastiaans, Xenakis
came across this idea. He was the first to look at Gabor's theory of
communication and use it in a compositional format. Using an integration of the
grains described by Gabor, Xenakis composed with the aid of stochastics.
Xenakis said each grain has a threefold nature: duration, frequency and
intensity. He proposed that all music is composed of a large number of
elementary grains. (Xenakis 1971: 43) Xenakis also goes into detail about how
these grains can be used together. This was represented in a three dimensional
context using frequency, time, and grains as the three axis. He compared them
to galactic masses, in which all parts move but one must decide if they want to
concentrate on the movement of the whole galaxy, or one star. (Xenakis 1971:
46-50). This idea of a galaxy of sound makes one think of sound as a three
dimensional object, whereas before it had mostly been considered two
dimensional. It is this idea that allows sound signals to become textures,
sounscapes and galaxies. These terms imply that sound must exist in a three
dimensional context. It allows the composer to think of music in another way of
abstraction from the conventional forms of abstracting sound in order to
compose music.

The next chapter will discuss the compositions produced by Xenakis and those
that followed him.

4.Granular synthesis in Composition

4.1 Iannis Xenakis

The first musical composition written using the method of granulation was
Xenakis' Analogique B. It consisted of hundreds of splices of tiny fragments of
magnetic tape. (Robindore 1997: 11-12)
Xenakis has always been very diligent at keeping up with research in the
sciences. He spends time reading the articles and understanding the
mathematics behind the theories. He strongly advises all composers to do the
same. Xenakis read the articles written by Dennis Gabor amongst other articles,
and as a result wrote the book 'Formalized Music'. The book is basically a
collection of the scientific ideas he had studied at that time. It contains
information on this topic and also some other related ideas to do with perception
and compositional ideas with which they can be incorporated (Xenakis 1969).

4.1.1 Analysis of Analogique A-B

Xenakis, Iannis (1997)

Analogique A-B being the first composition written using granular techniques is
still wary about it's place in the world of music. It hides behind traditional
instruments. Maybe it is creating a cross-over bridge between acoustic and
electronic music. Analogique B is the granular synthesis part of Analogique A-B.
It was designed to be played simultaneously with Analogique A. Analogique A
was written for orchestra and was derived using stochastical methods, and then
converting them into musical notation. Analogique B consists of four tracks and
is played through 8 speakers, which are set out around the concert hall in a
particular manner (Xenakis 1959).
 Fig 4.1 Speaker layout for Analogique B (Xenakis 1959: 3)

Analogique A-B uses a very old form as it's compositional base. It uses the call-
response method. In the beginning of the piece the orchestra calls out.

The granular synthesis responds to the orchestra, by mimicking the orchestra. It

is not a simple repetition of the orchestral part. It sounds like Xenakis is creating
a machine that can process sound infinitely faster than humans, but it is still
learning. It spills out a jumble of sounds. The orchestra then calls back, much
more slowly and patiently. The machine listens but is still unsure. It responds in
a quick jumble of sounds. This is how the song progresses. A slower
instrumental part followed by a quicker tape part, ending on an instrumental call,
with no response. The call by the instruments is always much longer than the
response by the machine.

Orchestra - 35 second Call Machine - 22 second Response

Orchestra - 23 second Call Machine - 15 second Response
Orchestra - 7 second Call Machine - 4 second Response
Orchestra - 40 second Call Machine - 7 second Response
Orchestra - 81 second Call Machine - 37 second Response
Orchestra - 37 second Call Machine - 22 second Response
Orchestra - 31 second Call Machine - 14 second Response
Orchestra - 79 second Call Machine - No response
The last non-response of the machine could be a look into the future of
technology. The machine whilst being faster at computation than humans still
has greater limits.
Analogique B is a true response to the instrumental part in that it is was written
after Analogique A. Analogique A was written in 1958, and Analogique B was
written in 1959. It seems Xenakis physically uses the call to create the
response. Using the same stochastic methods as he used to write the
instrumental music, he divided the analogue tape into tiny fragments. Each
fragment also had some kind of envelope placed over it. These grains where
then stuck together in a carefully created new order. The machine phrases may
have been made shorter using a couple of different granular methods. Xenakis
may have used time shrinking methods, without changing the sense of pitch. He
may have shortened the duration of the grains and brought them closer
together, he may not have used every grain, he may have used a combination
of these things. Spectrally there is not really a lot to examine. It looks about the
same with the instruments as it does with the granular synthesis. This is
because Xenakis was not trying to create something with the spectrum. It piece
has more to do with the instruments, and the effect of granulation on them, not
on different granular sounds. The granular sound that is created in Analogique
B has a very light density and the grain duration is fairly long. This gives a very
bubbly kind of sound. There is also a type of smouldering sound effect created
by a more unpitched sound in the background, like someone exhaling very
loudly.

This piece would have been very difficult to produce and one can see why the
techniques associated with granular synthesis were not really used until
computers became an option.

4.2 Curtis Roads

Curtis Roads built on the ideas of Xenakis and his work has led to a lot of the
current work based on granular synthesis. One of Roads' first compositions
was nscor. It has a duration of 9 minutes, and was released in 1987, although
his first version came out a decade previously. There are different versions of
this piece that were created over the period of a decade, due to the long period
over which it was created there is no full score to nscor (Roads 1985: 142). The
different versions came about as advances in technology allowed Roads to do
more with his composition. This piece was written using a number of computer
programs such as Project 1 written by G.M. Koenig, the Sound Synthesis
Program (SSP) written by Paul Berg, and Music 11 written by William Buxton.
(Roads 1985: 148,158) nscor creates a texture that explores different sounds,
frequencies, timbres and durations.

Roads is also responsible for a computer program called Cloud Generator that
is designed to teach the user about granular synthesis. It creates large clouds of
grains, which could be seen as small textures, due to the limited control the
user has over the parameters. (Roads & Alexander 1996b)

4.3 Paul Lansky

Paul Lansky wrote a series of compositions released in 1994 called the Idle
Chatter Series. Lansky uses three different computer techniques in producing
these pieces. They are Linear Predictive Coding (LPC),1 granular synthesis,
and stochastic mixing techniques. It takes a minimalistic approach. This piece
involves many voices chattering unintelligibly. (Dodge & Jerse 1997: 274)

1 Linear Predictive Coding was originally a process for analysing sound, but has
since been used as a compositional tool.

4.4 Barry Truax

Barry Truax has written a number of granular synthesis pieces such

as Riverrun, The Wings of Nike, and Tongues of Angels. (Truax 1990:
120) Riverrun is Truax's first venture into granular synthesis, and is another
landmark piece in granular synthesis. This is the first composition to rely fully on
real-time granular synthesis techniques. Xenakis started off with an integration
of instruments and granular synthesis. Roads used an integration of granular
synthesis with a number of other computer music methods, which were
impossible to create in any real-time environment. It took more than enough
processing time without having to replay instantaneously as it was.

4.4.1 Analysis of Riverrun

(Truax 1988b)

Barry Truax described Riverrun as representing the flow of a river. As with

water, each countless grain of sound flows away, yet the river itself remains as
it was. (Truax 1988b) Like a drop of water each grain on its own is trivial and
powerless, but when it accumulates it becomes like a strong flowing wave of
immense power.

Riverrun was entirely written using real-time granular synthesis techniques

mixed onto 32 tracks. It consists of grain densities between 1 and 2375. All of
the grains used in this piece were created using either the basic sine wave
method or by using FM Synthesis as the grain content. (Truax 1990: 120) The
envelopes used by Truax were limited to linear envelopes. He used Triangular
and Trapezium shaped envelopes. They were all linear because it allows for
much faster processing, which is crucial in any real-time environment. (Truax
1988: 14)

In Riverrun each grain has been defined with an individual frequency and
duration. Truax said "When the grains all have similar parameters, the result is
a pitched and amplitude modulated sound,but when random variation is allowed
in a parameter, a broad-band noise component is introduced." (Truax 1988b)

Looking at the spectrum of this piece there is a lot to see. There are some very
distinct streams that are going on in the piece. See figure 4.4.
Freque
ncy

Time

Fig 4.4: Spectral analysis of Riverrun. Created with Cooledit96 (Shareware).

This piece ranges from 20Hz to 22KHz. It is on a linear scale. The duration is
just under 20 minutes.

It looks as if the song was built by putting large chunks of ideas together. It was
created using tendency masks to control all of the parameters, which may
account for the blocked appearance. Despite the look the texture is very
continuous. There is always something flowing through between the different
ideas. The sections in the high frequency range look like they fluctuate a lot, but
this is not actually the case. This is because at such high frequencies the sense
of pitch doesn't change very much at all, therefore a large movement in a high
frequency may only sound like it is moving a couple of tones. The entire piece
spans 20 minutes.

This piece starts off very calm with occasional bubbly noises. The bubbly noises
are created by long duration grains with a very low grain density 1-20 grains per
second. These grains can be seen on the three dimensional chart (appendix
3:1). Slowly the bubbles become more frequent, resulting in a gradual
accumulation of sound. The sound becomes very intense as the texture builds
up in density. As the density increases the duration of the grains is reduced
(see Appendix 3:2). Then a much higher pitched group of grains move across
the top of the piece, moving a little but not changing a lot (see Appendix 3:3).
They fade out and the lower granular noise also fades leaving a clapping noise
very similar to horse hooves (see Appendix 3:4). This clapping type noise must
have been made using extremely short duration grains with a very broad
spectrum at a very high density. It is interesting to note that Truax slowly
increases the duration of the grains within the clapping sound, so that they lose
their spectrum and with it their ambiguous pitch. This gives us a better look
inside the sound. We can see that it is made by having a large spread of grains
across the frequency, but with a very short duration this spread gets lost and
the entire bandwidth gets filled. Truax has made sure the lower frequencies of
the spectrum are especially dense to give it the extra thudding sound.

The next section of the piece goes into more randomly generated clouds of
sound (see Appendix 3:5). This sounds more like a chattering type of sound, not
dissimilar to Paul Lansky's piece Idle chatter, except that Truax generated this
using sine waves, rather than sampled vowel sounds. It sounds like thousands
of tiny voices all trying to talk or shout at once. They are obviously randomly
pitched because the texture slowly evolves to a single pitch. It then boils to a
very hollow type of bubble texture, within multiple frequencies. It is as if there is
a definite sense of pitch within the bubbles, but then they go very quiet. This
may be derived through using pairs of triangular envelopes over waves, so that
it appears to give an attack and then release with nothing in the middle,
whereas it is two separate grains.

The piece seems to concentrate more in pitch areas from this point on. For a
while it sounds like Truax has created a very deep cello, lower than ever
possible for an acoustic instrument. The low sound moves slowly up and down
in pitch (see Appendix 3:6). The last idea in the texture is a sustained sound
that changes in timbre, and moving harmonics.

In summary, Truax explores a number of different textures, and the ways that
can be moved between them. His main method of changing the texture is to
bring in random pitches that are either very bubbly, or hollow types of sounds.
The sounds he explores are very varied, from imitating claps to changing
timbres and frequencies.
5. Structuring the future

So far we have seen exactly what granular synthesis consists of, how grains
can be created, and combined with other grains, the thought process that led to
granular synthesis, and the compositions that have been created using granular
synthesis. This now leads us to another area. Doing our own work with granular
synthesis.

There are two things to consider:

What software to use
Areas in which to extend the use of granular synthesis

5.1 Software

The amount of software available for granular synthesis is small but it is

growing. Most of the software available is shareware, and available on the
Internet. There is also specialist software that was designed by composers for
their own exclusive use. This is generally not available. There are also
sequencing programs such as Cakewalk for the PC that allow the user to
perform time stretching, but apart from this one function are not designed for
granular synthesis. Table 5 in the appendix gives a summary of the most
available shareware granular synthesis programs.
Csound, written by Barry Vercoe, is for general synthesis use. It has been tried
and tested in various incarnations for at least 20 years now. It just requires time
and experimentation to get used to the interface. It has a semi-real-time
interface in that it does allow the sound to be played whilst it is being compiled
by the program. This is not true real-time synthesis though because the creation
process of a Csound texture is done on a text editor. The text file must then be
loaded into the compiler as a finished document before it can make any sound.
Being a text interface it offers much more flexibility and control than any other
program. There are not many functions in the way of granular synthesis, but it is
possible to make your own instruments (Ekman 1998).
The two granular synthesis commands it does have are grain and granule.
Grain is used for creating grains with similar contents. It has parameters for
amplitude, pitch, density, grain duration, and a number of random offsets. It
cannot be use for large samples though as it is not designed for time stretching
or such granular activities that require a sequential order of grains.
Granule on the other hand is designed with this purpose in mind. It has the
same options as grain, but it has extra control for sound samples. It allows the
user to specify how much of the sample to use, and where to begin the
sampling process, with associated random offset settings. It also allows for
multiple voices so you can layer different pitches on top of each other that will
play at the same rate (Ekman 1998).

The main problem with these commands is that they receive all their data from
a function table, which must be defined previous to compiling the program,
which means that the user cannot create FM signals and send them straight to
the grain or granule command.
The song Where the Wind Meets the Sea (Track 2), uses the grain command
exclusively to create it's texture.

5.2 Extending the use of granular síntesis

There are many areas that still require looking into in the world of granular
synthesis. Barry Truax has suggested looking for new ways of defining grain
contents, and organizing granular textures. (Truax 1988: 25) This area has
been explored for some time now, but due to the atomic structure of granular
synthesis, this area is of near infinite proportion. A question arises when
investigating this area. Are new contents and algorithmic ideas actually
producing anything new? This is debatable. Essentially the sine grain is the
most pure impulse of sound. Using this it should be possible to create or
recreate any type of sound or texture using the parameters available in granular
synthesis. In reality the technology available is not able to do this yet. This
means we must go beyond the sine grain to create certain textures and sounds.
This limitation has already been realised, which is why the search for new
sounds continues. As control over granular textures improves though the need
for new contents should diminish, then again by the same token it may
increase, due to ease and creativity.

The use of wavelets as a compositional tool, as mentioned in chapter 2, lies as

a vast new area awaiting in-depth research. Indeed any new idea for changing
the shape of a granule of sound would provide a large platform for further
investigation.

As for algorithms, they are the means of gaining better control over granular
textures and as such need to be explored. Rather than exploring algorithms for
grain order though, it may be more useful looking at clusters and how to
organise grains into clusters that produce certain effects. This could begin by
building a set of clusters or gestures that can imitate the phonetic alphabet,
such as is done with FOF synthesis, but with a little more reality. As with
chemistry, one must start with the basics and build up. If granular synthesis can
be compared to the atomic theory then it would make sense to create molecules
or clusters with the atoms before one tries creating anything bigger. This is best
achieved through devising algorithms as the control would be too fine for a
human to control. Of course there are many types of clusters that could be
created, the alphabet is just a starting point. Then there are higher level
algorithms that can be created to control the clusters. Using the alphabet again
as an example, it may be useful to create algorithms that sort the clusters to
create words. This would probably involve using synchronous granular
synthesis algorithms for creating the clusters and asynchronous granular
synthesis algorithms to lay out the clusters.

Another area that needs work is within the real-time use of granular synthesis.
For the general computer user, the option of real-time is still very limited.
Computers are much faster now and can process phenomenally faster than
imagined. This has meant that many people can work at home on Csound and
other synthesis programs creating granular textures. But many programs cannot
work in real-time, the ones that do are largely unreliable (Appendix 5).

Working in real-time is not essential, but it does improve the conception and
realisation of textures. This is due to the practical nature of working in real-time.
When working in real-time it is possible to change a parameter and receive
instant feedback. This will allow the user to become used to the controls very
quickly. It will mean that different textures are quicker to create and recognise.
When working in non real-time, making changes to algorithms and grains is a
mechanical process which then needs to be executed before the results are
even known. The action, delay, and reaction process that evolves through such
a process interferes with ideas. Results that may have been achieved could be
substituted for a lesser quality.

Barry Truax spoke on real-time granular Synthesis:

....Curtis Roads had done it in non-real time, heroically, hundreds of hours of

calculation time on mainframes, just to get a few seconds of sound. He had
done that, but it remained a textbook case. As soon as I started working with
it in real time and heard the sound, it was rich, it was appealing to the ear,
immediately, even with just sine waves as the grains. Suddenly they came to
life. They had a sense of what I now call volume, as opposed to loudness.
They had a sense of magnitude, of size, of weight, just like sounds in the
environment do. And it's not I think coincidental that the first piece I did in
1986 called Riverrun which was the first piece realized entirely with real-time
granular synthesis.... (Iwatake 1991).

Unfortunately whilst there have been real-time granular synthesis programs

since 1986 such as the PODX system, they have not been widely available. The
main reason is due to that fact that most home computers still can't handle that
kind of processing. Hopefully with the new wave of computers that are coming
out, there will become programs available that can handle granular synthesis in
a real time environment. This will excite more people into realising the potential
of granular synthesis, because lets face it, composers and musicians don't get
excited by numbers on a page, they get excited by what they hear. This is
something that has become integral in music for two centuries now. It is what
motivated the imaginative revolution, known more commonly as the romantic
era. Nietzsche described music as a force to seize hold of the heart and
introduce the listener and indeed the composer to crystalline ecstasy (Wilson
1987: 1). Music is not music until it heard. Using a real-time environment would
also increase the potential to make sound clusters, and experiment further.
That last thing that should be mentioned is texture. This idea has already been
looked into previously in this thesis, but like grain contents and algorithms this
area is also vast. It needs to be examined as a large field of infinite proportion
that is capable of producing anything. All the previously mentioned algorithms
and clusters can be implemented in the exploration of this territory. Granular
synthesis has for a long time been divided into two separate sections, the
orchestra and the score, or the sound and the structure (Truax 1990: 120). This
is just following the original conventions of music. Composers write a structure,
whilst the musicians make the sound. Using a texture as a musical terms
requires changing the way one thinks of music. The texture is the score is the
orchestra. It is a single entity. It has many parameters which have been defined
in granular synthesis, maybe there are others that still need defining or at least
refining.

6.0 Conclusion

Sound signals the size of a grain may seem insignificant, but they hold more
potential than would have been thought possible fifty years ago. Despite the
potential of granular synthesis it is still vastly ignored by composers, due to its
complexities and a general lack of knowledge.
Through the course of this thesis the underlying concepts of granular synthesis
have been explained. Granular Synthesis has many control parameters, and
whilst these may seem daunting, they allow for vast textures to be created. Of
course there are many algorithms and distribution methods that can take care of
control problems, and help add complexity to the texture itself.

The historical background and the perceptions in sound that led to the
conception of granular synthesis demonstrate the thought process that had to
be developed in order to think in granular concepts. In many ways this idea is
still not fully understood, but it is slowly becoming more common. Xenakis has
brought sound into a three dimensional space which is much more expansive
than any two dimensional sound space that has been previously envisioned.
Composers need to look out towards the expanse of space and think musically.

There have only been a few major works written using granular synthesis. The
tools are now available, technology is finally adequately developed to be able to
create granular textures. There are many areas in granular synthesis left to
explore, areas that have rarely been considered before now. The most exotic
textures to explore through granular synthesis techniques in composition are
now at hand.

This is an exciting place in musical history to be right now.