3.3.1. Environmental Sensing
As discussed in
Section 2.2, within Di Scipio and Sanfilippo’s approach, the environment is generally considered in terms of its ability to disturb and affect actual—or potential—acoustic phenomena. We build on their work by expanding the role of the environment in its ability to perturb aspects of the various autonomous systems that function within it by collecting data that is not restricted to the sonic domain. This was explored in the most technically challenging part of the installation: an ecosystemic network of 18 loudspeakers, each positioned in 18 raised garden beds among vegetables and plants. The beds were separated into three groups, each outfitted with its own system of microphones, speakers, and environmental sensors. In total, three Belas were used to run a variety of sensors in this area of the garden (see
Table 1 and
Figure 5 for details).
Using custom software written in Pd and loosely based on Di Scipio’s work on autonomous feedback systems and emergent sonic structures [
18], self-adaptive feedback loops were created between the microphones and nearby loudspeakers. Essentially, these comprised a dynamic bandpass filter which would allow more or less frequencies to pass through depending on the amplitude of sound being received through the microphones. Importantly, no sound other than what was picked up by the microphones was used as source material.
The Larsen effect [
20] was induced by placing the microphones proximally to the loudspeakers. Tamed by the adaptive filter, this sound was further processed using delay lines, granular synthesis, comb filters, and bandpass filters. In additional to the audible processed feedback, sounds from both humans and animals entered into the network, including significant activity from birds and crickets. The environmental sensors, which included measures for soil, wind, movement, and light, were used to inject additional disturbances into the feedback computation (see
Figure 6). The resulting sound was the result of several self-organizing systems coexisting within the same environment.
3.3.2. Composed Relations
Being the result of the coupling of autonomous systems, the work itself was not composed, though compositional decisions could be said to reside in the mappings and relationships between sensor input and computational processes. Following several visits to and observations in the gardens, sensors were placed within the beds primarily on the basis of which sensor was the best fit for the bed’s conditions. This might depend on the location of the bed within the site, or the density and type of vegetation within it. The data from the environmental sensors was then analyzed in real-time and normalized, scaled, and mapped to control various parameters of DSP on the Belas. The result was a curated collection of simple relationships between organic matter, computational processes, and the environment, combined to produce complex emergent behaviors. The aesthetic and experiential considerations were determined through repeated and ongoing listening and observing within the site on multiples days, and at different times. Below are a few examples of such composed relations:
Wind: Along with the sea, wind is arguably one of the only naturally continuous excitations [
37]. Its continuous nature made it a useful resource to harness within this work. Wind speed readings from an anemometer (
https://www.adafruit.com/product/1733) were used to control how quickly sounds were spatialized throughout the garden, creating a poetic connection between the movement of the wind, and the diffusion of sound within space. Minimum and maximum wind speeds were calculated continuously over the duration of the installation. When a new input value rose above a previous maximum (or below a minimum), the respective input parameter was adjusted, and the wind speed was calibrated accordingly. Due to the design of the anemometer—comprising three hemispherical cups mounted via arms onto a vertical shaft—wind speed is always measured as an average. This provided sufficiently useful wind-related data, which did not require further smoothing.
The relationship between the wind speed and sound spatialization was directly correlated: the faster the speed of the wind, the faster and more chaotic the movement of sounds. Slower wind speeds resulted in a stillness in sound content and diffusion. The spatialization was implemented using constant power panning, which was controlled by the wind speed. Timbral changes were achieved by mapping the rate of change of the scaled wind speed value to the density parameter of the granulator. This varied the number of grains of sound being played back within the software, leading to changes in perceived sound density.
Movement: In the same bed, an accelerometer (
https://www.adafruit.com/product/163) was used to measure the movement of a swaying burlap sunshade. The three-dimensional data was normalized based on the minimum and maximum of values received over a sliding time-window. The method of rescaling the data was again used to account for the changing environmental conditions. In this case, the movement data was used to vary timbral aspects of the sound by modulating the speed of certain control waveforms within the granulator. The audible result was almost gestural, visually linked to the movements of the cloth, which allowed spectators to gain some insight into the processes at work.
Temperature: In another portion of the system, a soil sensor (
https://cdn-shop.adafruit.com/datasheets/SLHT5.pdf) was used to measure the temperature and humidity of the soil. The initial soil state was read at the launch of the system, and subsequent fluctuations in these values modulated the overall tuning of several bandpass filters within the feedback system. The center frequencies of these bandpass filters were initially tuned to another audible source within the garden (see
Section 3.4). The soil sensor values were scaled and mapping to these center frequencies. As they slowly changed over time, the resulting shifts in harmonicity led to an almost indiscernible process of detuning and retuning. Of course, in a climate more susceptible to rain—or if the bed had been watered—these changes would become more immediately perceptible to observers.
Ambient Light: Located in number 19 of the raised garden beds, a further system was developed, which involved several sunflowers. These were embedded with photoresistors (
https://www.adafruit.com/product/161) to measure levels of ambient light (see
Figure 7). The photoresistors were used as an affordable, lightweight solution to measure the amount of movement of individual flowers. This was accomplished by measuring the change in light over each subsequent frame. The sensor inputs of the Bela are sampled at audio rates, which allows for smooth alignment with audio data [
38]. To avoid encumbering the movement of the plants, the photoresistors were strewn together using conductive thread.
Following the movement of each flower, fluctuations in light gathered by the photoresistors were mapped to the amplitude of several subtractive synthesis engines. These comprised white noise passed through eight resonant bandpass audio filters. Each filter was tuned to a different harmonic (positive integer multiple) of the audible drone of a nearby dairy factory (see
Section 3.4 for details). As with the nonlinear nature of audible feedback systems, the movement of each flower is also complex, being influenced by environmental forces and its physical structure, as well the movements of adjacent branches. By giving several flowers their own voice, this
imitation [
39] allows for further insight into the complexity of their movements, bringing attention to subtleties of the environment by emphasizing the collective rhythm of moving plants.