This RC entry discussed two concrete approaches in composing the coupling of a feedback system with a specific site and situation, with the aim of making this relation central to the generative process that is experienced in the space. The two methods differ in conception and realisation, but what's more important is that they give access to different aesthetic and compositional affordances.
The piece in Reagenz, being designed to be extremely sensible to initial conditions, is intrinsically more open to deviations and less predictable. The generative process can wander in many directions, producing sound forms that extend into variegated textures, long gestures and articulations, rythmicalities, repetitions and so on. I think the piece has a good temporal balance, that makes state transitions quite perceivable without being too chaotic. I'm also quite satisfied with the variety which is generated out of small differences in the response of the glass: in the way I perceive it, this mainly influences the spectral region in which the system tends to move.
The Kunsthaus work is a bit different, because the states the system transitions to are much more pre composed. This site being a shared space, I wanted to have some more control on the sound, 'bounding' the system a bit to have a balanced alternation of densities and volumes. This, combined with the fact that a direct coupling to the meta parameters of the system didn't really work, created a piece that is much more predictable and that works more in the details. Nevertheless I think it is well balanced with the other three segments, and when listening close I think it is possible to perceive these small movement of adaptation of the algorithm to the staircase space.
{kind: paragraph, function: comment, keywords: [algorithm, coupling, affordances, composition, feedback, fdn, generative, emergence]}
in Reagenz
In transduction I attached two transducers to the window at the entrance of Reagenz, thus creating two oscillating membranes that project sound outside, and simoultaneously one pick-up microphone feeding vibrations back into the software to initiate sound processes. I wanted the non-linearity of the glass, and its ever changing response due to temperature and other conditions, to be central to the generative process. I thought I could you use the feedback network as a sort of 'magnifying glass' to amplify the response differences, and I had the idea to produce several short 'pieces', each being generated out of a single 'acoustic snapshot', representing the response of the window at a very specific time of the day.
So how the piece works is: every thirty minutes one transducer plays back a pre-recorded noise impulse that is simultaneously picked up by the microphone and then used as initial perturbation for the network. The glass acts as a filter and, due to its non-linearity, the impulse that is then passed to the network as initial perturbation is always slightly different from the previous ones. This difference in itself cannot be heard, and it is also not so easy to spot when looking at the waveform. I thought it was an interesting challenge to try to amplify these minute differences through the network, and therefore I worked with the aim of maximising its sensibility to initial conditions.
I selected some system states, with different behaviors, feeding the network with impulses recorded through the glass in order to observe which state was the most sensible to these specific variations. I found one that I thought had a good balance between stable and chaotic behavior, and that was very dependent on the initial impulse for its development (see video --> ). Then I fine-tuned the system manually to avoid distorsion due to the resonances of the glass.
This way I could achieve a coupling between the window and the feedback network that makes the non-linearity of the glass a central form generating principle: each sound piece that is produced is unique to a condition captured at an arbitrary time, and the sound forms that compose the piece are exclusively emerging out of the response of the system to that exact initial perturbation. In other words, the piece makes perceivable the reaction of the system to an 'acoustic snapshot' of the glass response at a very precise moment of the day, acting as a sort of magnifying glass that audifies these differences into sound developments.
{kind: paragraph, artwork: transduction, keywords: [algorithm, coupling, affordances, composition, feedback, fdn, generative, transducers, vibrations, pickup, differences, glass, window, filter, initial conditions, resonances, tuning, impulse, noise, network, sound forms, emergence]}
The general setup of Through Segments is composed of thirty-two loudspeakers and four microphones disposed all along the staircase. Each of us composed his/her segment for one microphone and eight loudspeakers, grouped into two clusters of four. Each cluster was then installed in a flight of stairs, so that eight speakers would cover one whole floor of the staircase (two flights of stairs + mezzanine).
Here I decided to couple the algorithm to the site in a way that could make the system adaptive, rather than just responsive (as it was in Reagenz). During the day many acoustic situations alternate in the staircase: depending on the office hours, cafe openings, visitors coming to the museum, this place can be very quiet but also very loud, and crowded at times.
I wanted my system to adapt to these different situations, so the idea was to couple the program to the signal coming from the microphone in a way that could make it transition through different states, with varying sound densities. I defined four quite heterogeneous states (ranging from quite and stable to chaotic and present) and then I tried different coupling to the microphone. I especially experimented with relating the rms of the staircase to the macro parameter I used to compose the four states, but I couldn't find a good balance: states transitions were either too fast or not happening at all, and the coupling I had in mind was not really perceivable in situ.
Therefore I decided to change strategy and to use the idea of adaptivity to articulate different behaviors in a single state, while scheduling state transitions independently from the microphone signal. In particular, I coupled the microphone to two main parameters: (1) delay lengths (2) duration of the zero crossing chunks. Perceptually this corresponds to differences in noisiness and density of the generated sound forms. This way I obtained a system that is adaptive at the micro level (the microphone signal influences the details of the sound forms), while the macro behavior is defined a priori (state transitions are scheduled at specific times).
(sound recordings here)
{kind: paragraph, artwork: through segments, keywords: [algorithm, coupling, affordances, composition, feedback, fdn, generative, staircase, adaptation, network, sound forms, emergence]}
This RC entry gathers some thoughts about how to relate sound
algorithms with specific sites or physical setups and possible
strategies to compose this relation. These thoughts I collected in
summer 2020 when working on two site-specific sound installations in
Graz: through segments, a collaborative sound work installed in the
staircase of Eisernen Haus in Kunsthaus and transduction, a work
conceived for the window niche at the entrance of Reagenz, space for
artistic experiments.
The two works deal with the question of how to couple a
generative process with a physical space and how to make the relation
between the sound synthesis algorithm and the physical setup an
essential form generating principle. I chose to use the exact same
sound algorithm for the two pieces, and to work esclusively on the
relationship between the algorithm and the site it is installed in in
order to make this aspect central to the compositional work. In other
words, the main difference between the two works is in the way the
algorithm is coupled to its site. I will discuss
how this dissimilarity constructs two conceptually different generative
relationships, and how these are crucial to the emergence of the sound
forms that are generated.
I'm portraying the compositional work in a diaristic form, discussing
my choices retrospectively and at the same time pulling out
of the experimental process some aspects that will be the base for
further reflections.
{kind: paragraph, function: introduction, keywords: [sound, algorithm, site, setup, sound installation, generative, relationships, composition, coupling]}
composing the algorithm
Since the coupling of the algorithm to the space is also related to
the affordances of the program itself, I'd like to briefly talk about
how it was composed and how it came to the 'final form' the two works
are based on. Indeed the algorithm - essentially a 16 channels
feedback delay network - was not pre existent but rather the result of
a development spanning some months, in which I alternated working on
the two pieces. In some sense, traces of both works are inscribed in
the structure of the program: its main form was mainly conceived when
I was working for the Kunsthaus piece, while refinements, tuning and
some small features happened when I was more focused on the Reagenz
work. The two ways of coupling the algorithm with the respective sites
were then conceived independently.
Thinking retrospectively about what I did, I can derive a sort of
idiosyncratic 'receipe' to work with feedback I used in the
compositional process. This could be summed up in four main steps. The
first three happend when working on the Kunsthaus piece, the
last one when I was composing for Reagenz.
(1) take a sound model (2) tweak the model until (3) analyse what
happened (4) derive marco parameters
development repo
final kunsthaus repo
{kind: paragraph, keywords: [algorithm, coupling, affordances, composition, feedback, fdn, process, workflow]}
coupling the algorithm
The resultant program is essentially a generative complex system based
on a sixteen channels feedback delay network. After being triggered by
an inital perturbation, the network enters a state of self-oscillation
generating emerging sound forms and developments in time. The system
is completely deterministic, meaning that for the exact same initial
conditions it will always produce the exact same sequence of sound
forms. In order to generate differences - to achieve a variety of
behavior when the piece is installed in a sapce - it is then possible
to take two different approaches: (1) couple the system to the space
so that the environment provides initial conditions that are never the
same (this is the approach I used in Reagenz) (2) couple the system to
the space in such a way that the structure of the system is altered in
time (this I tried in Kunsthaus). These two approaches differ on a
conceptual level because in the second case I'm creating a complex
system that is also adaptive. In the first piece the system itself
does not adapt: here the system never changes, the differences are only
happening in its environment (in the systemic sense) and are
provided to the system in the form of external perturbation. It is
worth noting that these kind of systems have an intrinsic openness
to the outside. The network I'm describing here, in the absence of any
external stimuli, does not produce any sound. It is when it gets
connected to an outside that interesting relations happen, that is
one of the reasons why I find interesting to work with these kind of
processes.
{kind: paragraph, keywords: [algorithm, coupling, affordances, composition, feedback, fdn, generative, self-oscillation, emergence, sound forms, interaction, environment, perturbation, outside, inside]}
(1) take a model
One of the first ideas I had when working for my Kunsthaus
segment was to try to relate the concrete space of the staircase with
a "virtual" space I would construct through sound processing. When
searching for a (reverberation) model I could implement in
SuperCollider I stumbled upon the Hadamard feedback matrix, that I
decided to try. My first implementation was a "room" whose dimensions slowly shrink or expand (in irregular pattern) according to the amount of energy measured in sixteen different frequency bands.
This resulted in a reverb evolving slowly over time with a "drony"
development and the overall impression is that of a room with
infinite echo, as I expected from the Hadamard model.
{kind: paragraph, keywords: [space, reverberation, hadamard, simulation, model], place: Kunsthaus}
space KH
The space in Kunsthaus is a four floor staircase that connects the Iron House to the "friendly alien". Each floor has a mezzanine, therefore the staircase is divided into eight flights of stairs of different length. It is made of concrete, which makes it a quite reverberant place. The staircase is a passageway for many visitors, since it connects various location inside the Kunsthaus (KH Cafe, offices, Camera Austria etc).
{kind: paragraph, function: contextual, keywords: [staircase, site, floor, staircase]}
(1) working on sensibiliity to initial conditions: differences happen outside the network and are passed in as external perturbations
{function: comment, keywords: [sensibility, initial conditions, feedback system, perturbation]}
(2) working on adaptivity: differences in the environment directly alter the structure of the network itself
{function: comment, keywords: [adaptation, feedback system, environment]}
(2) tweak the model
Then I played a bit with the model, to see if I could make it behave
differently. In particular, I was interested in finding a point at
which it couldn't be perceived anymore as a reverb. This was a rather empirical research, and I mostly experimented with different ways of using the outuput of the algorithm to regulate its internal
processes. Again, I messed around mainly by patching RMS values coming
different frequency bands, and I also introduced some way of chunking
the signal in the feedback loop by looking at zero crossing
occurencies. Some intermediate experiments:
{kind: paragraph, keywords: [mode, feedback, rms, zero crossing, experiment]}
(3) analyse results
In the meantime I was also porting part of the code into faust to
optimise the thing in order to run it on a Raspberry pi 4. I was
trying to port it as close as possible and soon I realised this
process was impossible really useful in analysing in detail what
I previously did in an istinctive way. In particular I could find some
small components I was not aware of, that were rather important
elements for the behaviour of the overall feedback system. By working
on these details I got to some more advanced experiments (8 channels to 2 reduction):
{kind: paragraph, keywords: [faust, porting, experiment, raspberry]}
Some initial impulses recorded during the exhibition period. Some minute differences can be heard. There is also some audible artifacts caused by the normalisation process. Since the system is deterministic I can seed the network with these recordings and replicate what happened on the window at a later time, in a different location (even tho missing the colouring of glass, transducers etc.)
{function: comment, keywords: [impulse, filter, glass, differences]}
(4) derive parameters
At this point I was satisfied with the variety of behaviors, nevertheless sound was
quite dense and complex and I decided to find a way of selecting and
exposing different parts of the process, instead of having everything
togheter all at once. I derived some 'macro parameters' that I could
tune in order to balance different parts of the system. This activity
was partly derived from the previous 'analysis' and partly again
empirical, since changing weights in a feedback system is always a
delicate step with rather unpredictable results.
{kind: paragraph, keywords: [tuning, parameters, feedback, system]}
space
The space in Reagenz is a street-facing, ancient window niche made of three thin glass surfaces and wood frames. The niche can be closed from the inside, giving the window a "showcase" look. Morellenfeldgasse is a quite calm street in the center of Graz, with a relatively small amount of traffic and noise.
{kind: paragraph, function: contextual, keywords: [site, window, glass, sound installation]}
