A compositional methodology based on data extracted from natural phenomenaSkip other details (including permanent urls, DOI, citation information)
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License. Please contact firstname.lastname@example.org to use this work in a way not covered by the license. :
For more information, read Michigan Publishing's access and usage policy.
Page 00000001 A compositional methodology based on data extracted from natural phenomena Natasha Barrett, NoTAM, Pb. 1137-Blindern, 0317 Oslo, Norway. ABSTRACT This paper illustrates, through practical examples from the composer's own work, methods to base compositional structure upon natural phenomena by extracting data from real-world processes, or from the pattern a process leads to. Obtaining control data from observational techniques has been chosen in contrast to the more popular use of mathematical models, or aural analysis of acoustic sound recordings. Examples show how data have been extracted from two contrasting sources: a rain-forest acoustic environment, and a system displaying self-organised criticality, and how these data have been used to control the spatial, temporal and spectral organisation of musical structure. INTRODUCTION As for a great number of composers, aspects of the natural world inspire my composition. I apply a number of compositional techniques with the ideal goal of musically evoking my perceptual and emotional reaction to these aspects. In earlier works, these techniques consisted of the aural compositional methods used traditionally in acousmatic composition, including recording environmental and other acoustic sound sources, transforming these materials with various degrees of surrogacy, and mixing them in a manner to evoke a musical or extra-musical concept based on aural judgement. I continue to use such methods, to various degrees. The search for effective methods to express concepts in sound, such as ideas ranging from simple kinetic activity to complex landscape description, in a way that is balanced on both musical and extra-musical levels, seems infinite. Through searching for a method to more clearly and accurately articulate natural phenomena, it was clear that the use of numerical data would prove useful. Nature's formations are created by, or are part of, complex processes about which the sciences are only approaching an understanding. In many cases, understanding is aided through numerical modelling techniques, or by accurate observational methods. Numerical models create data in an attempt to simulate some aspects of a process or pattern. This is in contrast to observation, where data are obtained directly from the subject. Historically, common methods of numerical control over musical structure have involved data borrowed from scientific models. Cellular automata have been used to model growth processes (Brown & Rothery 1993), Markov chains have been used to describe geological strata (Davis 1986), stochastic processes used to model genetics (Brown & Rothery 1993) and the behaviour of gases, and all have been appropriated for musical use in some shape or form. It is not within the scope of this paper to investigate why some composers have chosen to use such control methods, nor to analyse or judge their degree of success. Nevertheless, rarely have data been applied with the aim of evoking specific physical phenomena, and often a lack of appropriate musical mapping has been clear in the result. If numerical data from either modelling or observational methods are to be used in a music composition, there are several questions that need asking: Which aspects of the subject do I want to capture in the music? Which aspects of macro and micro musical structure should be controlled, and what aspects of the data are musically most appropriate? Do I need to re-scale data sets such that the listener can perceive a musical 'mapping'? Such questions are difficult to answer, and are different in every context. To investigate these simple questions it is necessary to have at hand reasonably complete sets of data describing the pattern or process. Often, data that describes a strong visual trend is not the most appropriate for musical mapping. For example, a numerical description of the shape of a landscape (commonly achieved by manipulating a fractal algorithm) may be less musically appropriate than a description of the process of erosion. In this instance, the description of the pattern has less importance than the description of the process. When one begins to look closer at how a natural phenomenon occurs, often the complete picture is much more involved than one expects, and an algorithm will rarely provide sufficient numerical data to begin answering such questions. For example, although Xenakis used algorithmic descriptions of gas particles to define some aspects of the musical material, significant editing was necessary to approach a coherence in musical structure (Xenakis 1971), and one could argue that the source implication weakened in the musical result. There are, however, advantages to using algorithmic descriptions, in that one can "tweak' parameters such that the most contextually pleasing musical control data are obtained. Through wanting to avoid appropriating an algorithmic process to create musical materials and structures, I decided to obtain control data from real-world observation. In the following I give examples from two different projects: one where I collected and analysed my own material (through lack of sufficient existing data), and one where data and a subsequent numerical model were supplied by a third party. REAL-WORLD OBSERVATION: THE SPATIOTEMPORAL DISTRIBUTION OF ANIMAL VOCALISATIONS IN A TROPICAL WET FOREST. Though there have been many studies of the acoustical behaviour of individual animal species in tropical forests, the complete, complex acoustical ecology of this environment is a little researched area. By sampling the whole community, the distribution of animal signalling in time, space and frequency could be assessed. The project was carried out with the help of Oyvind Hammer at the University of Oslo. Two locations were tested, only one of which was used for musical application. This location was the biological field station of La Suerte, located in the lowlands of northeastern Costa Rica. The forest of La Suerte has a very high density of vocalising frogs, mantled howler monkeys, oropendolas, white-collared manakins, tinamous, and insects. Sound recording An array of four microphones (Acoustical localisation system (ALS)) was set up in a location of partly disturbed wet forest, and 14 hours of sound were recorded over a 24 hour period. Three condenser microphones were positioned in a 20 meter triangle one meter above ground. The fourth microphone was
Page 00000002 hoisted into a tree, five meters above ground, in order to improve vertical accuracy of the array (figure 1). The ALS method is emerging as a robust, non-intrusive way to track vocalising animals in space (McGregor et al. 1997). By measuring the differences in sound arrival times between all pairs of microphones, the position of the vocaliser can be determined with an accuracy that depends on the distance from the array. The distances between microphones should be large to improve accuracy, but not so large that a sound will be received in only some microphones. The relative positions of the microphones were found using measuring string and compass, with an estimated accuracy of 20 cm. There was no human presence in the forest apart from ourselves when changing tapes, thus ensuring that behaviour was not significantly disturbed. long-term energy distribution in different frequency bands. In addition, temperature variation, rainfall, and daylight times were noted. The amount of vocalisation activity was noted through a simple aural analysis of animal diversity and sound intensity. Only a brief selection of results is presented here. Spatial and temporal information Differences between arrival times at the four microphones were estimated by finding the peaks in the time-domain cross correlations between each of the six pairs (McGregor et al. 1997). To calculate the sound location we chose to do a simple exhaustive search of the space, finding the position where the differences between calculated and observed arrival times were minimised. The array was set up close to a group of howler monkeys, which moved out of the area during the recording period. At the beginning of the first recording vocalising activity was high. Three howlers could be differentiated and localised as shown in Figure 1. Howler A, a male 11 meters above ground, was very active, giving barks and occasional roars. Howler B, a male 30 meters above ground, vocalised less frequently, but using very loud roars. Howler C, 21 meters above ground, gave frequent soft calls (chuckles) of low amplitude. Figure 2 shows contour plots of the arrival time error functions for howlers A-D. The position of the animal is taken to be the one that minimises this function. It is clearly seen that distance accuracy decreases further from the array, yet the direction remains clear. This is most appropriate for musical mapping where our perception of direction is more accurate than our perception of distance. The time instances of sounds were logged manually from the recordings, along with an indication of the character of the call. c SD I0 FIGURE 1. Spatial relationships during first recording at La Suerte. Microphones are represented by circles, birds by crosses, howler monkeys A, B, C and D are marked with letters. Grid size 10 m. Four-track digital recordings were made using two DAT recorders with rechargeable battery packs. The two recorders were synchronised using a click-producing switch, with a better than 0.6 ms accuracy. Speed differences between the two recorders were also measured. Each acoustic sampling lasted for the life of the batteries; about 1 hour and 50 minutes. Two sets of rechargeable batteries allowed us to start a new recording every 3 hours. Air temperature was measured at the beginning of each recording. These readings were interpolated in order to estimate the speed of sound at any time, using a linear approximation (Rossing 1990, p. 40). Test sounds (human voice) were produced at seven known locations around the array, and used to validate the results from the acoustic localisation procedure. The error ranged from 0.2 m at the centre of the array up to 4 m at 30 m distance. Even though the absolute values of the positions may be inaccurate, the variance is small, so that relatively close vocalisers can be separated and small movements detected. Data analysis: methods and results The recordings were analysed to provide the following information: (a) the location of animals and the direction in which they were moving, (b) the temporal placement of sounds, (c) the 150 M. i~ 140[ 120 110 E 100 ýoc D 0..................................... 50 20 30 40 50 60 70 80 90 100 110 120 m FIGURE 2. Contour plots of error functions for the howler monkeys A-D. Contours at 1-7ms error levels. The position of the vocaliser is supposed to be at the minimum of these functions. Accuracy decreases away from the array, but direction can still be precisely determined. With this method, spatial and temporal locations for monkeys and birds were obtained. Figure 3 shows an extract of these data.
Page 00000003 Time Description Position 0'03.3 howler C soft calls start -8, 9, 11 Howler C goes on almost continuously, at least for the first minute. 0'04.2 howler A, 4 quiet loud calls 0'06.3 howler B, 5 quiet loud calls 0'12.0 howler A, 2 quiet loud calls 0'20.0 howler A quiet loud call 0'32.6 howler A, 3 quiet loud calls 0'37.7 howler B loud calls 0'42 bird - not localisable 0'42.7 howler B quiet loud calls 0'45.4 howler A, 3 quiet loud calls 0'50 bird, 7 kHz 0'55 bird, 7 kHz 0'55.3 howler A, 4 quiet loud calls 1'12.0 howler C soft calls stop 1'15 bird. 600-1000 Hz. -49, 17, 11 -20, -24, 35 -49, 17, 11 -49, 17, 11 -49, 17, 11 -20, -24, 35 -20, -24, 35 -49, 17, 11 -7, 16, O0 -7, 16, O0 -49, 17, 11 -8, 9, 11 -81. 90. 37 FIGURE 3. Data extract from analysed sound recordings showing spatial and temporal information. Coordinates given from the centre of the microphone array (0,0,0). Often, interesting temporal information was heard in animal calls not loud enough to allow location identification. In these instances, only the time information was extracted. Long-term energy distribution Continuous vocalisations produce background sound levels equal on all microphones. Over the short term, these sound levels produce a constant drone in relation to louder animal calls. Over the long-term, the 24-hour energy distribution can be obtained. An analysis of the energy content in seven one-octave bands over the first 12 hour period is given in Figure 4. The peaks in the 150-320 Hz band are due to aeroplanes. Loud calls from howler monkeys are mainly located in the 320-670 and the 670-1400 Hz bands, with particularly high levels at 0 minutes when the group was close to the array. Wind and rain are the two most prominent contributors in the 1.4-2.8 kHz band. The peaks at 100 and 190 minutes correspond to gusts of wind, while the peak at 620 minutes is a short rain shower at night. The 2800-5500 Hz band is dominated by frogs during the day, and insects at night. The activity in this band has a falling trend throughout the day, but increases substantially to a stable, high level at night. At 620 minutes, a small dip indicates that the insects are inhibited by the rain. The two uppermost bands are occupied by insects and some birds. The dusk chorus around 430 minutes shows prominently in the upper bands of the frequency analysis. Insects are major participants in the dusk chorus, along with birds and howlers. Musical Application Two composition projects were undertaken: one work for tape, and one work for ensemble, tape and live electronics. Below are examples of how data were mapped into musical aspects. In addition to using the numerical data, the recordings were used as source-sound materials. Sound examples refer to locations in the tape composition "Viva La Selva" (Barrett 1999). (a) The long-term energy distribution in different frequencybands over the 24-hour period was scaled to a 20-minute version, and used as the macro structure for both tape and ensemble works. The overall variations in sound activity, temperature, rainfall and daylight were also included in the structure. Clear examples of this can be heard at the opening of "Viva La Selva" (0'00 - 1'00) corresponding with the beginning of the recording period (10.40am), and from 9'25, corresponding with the night period and a sharp increase in insect activity. (b) The spatial and temporal locations of prominent callers were used to locate specific sounds in the tape work. As the tape piece was to be in stereo format, the three-dimensional co-ordinates needed to be reproduced in the stereo picture, with the illusion of the third co-ordinate via reverberation and filtering. In many instances I would begin with original animal calls as sourcesounds, and then gradually implement increasing surrogacy, or substitute non-animal sound materials. In this way the music maintained the spatial and temporal interactive attributes observed in the recordings. Sound extract 10'00-10'30 illustrates the increasing surrogacy and sound-source substitution method. Original temporal information was taken from three interacting "tink frogs''. The material continues over the next one minute, where the "tink frog" sound, and respective surrogacies, are heard with other materials. Sound extract 11'25-12'30 illustrates how the temporal and spatial organisation of the sound materials (human voice) are mapped onto the calls of three interacting howler monkeys (numerical data extract given in figure 3). In this sound extract, the "tink frog" material is also heard. USE OF A NUMERICAL MODEL: SELF ORGANISED CRITICALITY The stereo tape format used in Viva La Selva restricted the application of interesting spatial data. The second example presented here uses ambisonics spatialisation techniques, and uses modelling, as opposed to observation, for obtaining the numerical data. The initial aim was to find a way to accurately and perceptually control three-dimensional spatial trajectories as an integral part of the musical structure. In other words, the spatial interplay would be the structural force behind the music. It was necessary to use a process or a pattern found in nature that was clearly evident to our perception, and valid on both macro and micro structural terms. For this purpose I used a selforganised critical system. Self-organised critical systems evolve into complex critical states, after which, the size and frequency of disturbances follow a 1/f type distribution (Bak 1996). Such systems can be found everywhere in our natural and social world where periods of equilibrium are punctuated by infinite scales of disturbance. In 1996, at the University of Oslo physics department, experiments were carried out on piles of rice, to test the self-organised critical behaviour of avalanches, and the approximation to 1/f pattern. After initially intending on using their observational data, it was clear that the observations were very similar to the results of simple computer simulation. The experiment and simulation are clearly described in Bak (1996). I used a small program operating on a 40x40 grid, onto which virtual grains were randomly dropped (Poisson process), and subsequent avalanches of different sizes created. The advantage of using the model instead of the observed data was that I could select interesting information, and control parameters like density of drops, and speed of avalanche propagation.
Page 00000004 11-22 kHz 0 I 100 5.5-11 kHz 0 40 40 I2.8-5.5 kHz 0 o * o n ^ ^ ^ ^ ^ ^ ^ ^ ^- ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ - ^ ^ ^ ^ ^ - ^ ^ ^ ^ ^ ^ ^ ^ aD E I 1.4-2.8 kHz < 01 100 670-1400 Hz 200 320-670 Hz 400 150-320 Hz 0 100 200 300 400 500 600 70 Minutes FIGURE 4. Mean amplitudes in seven one-octave bands over the first 700 minutes. 0 Musical Application The computer simulation created a large data set. The drop rate was adjusted until a suitably dynamic set of data was achieved. From a simulation which would result in 30 minutes of sound material, one third of the data was selected containing approximately 50,000 time-location entries. A 40x40 'grid' of sounds were created by transposing and filtering single input sources between two extremes containing 40 increments (the x-axis controlling transposition, the y-axis controlling filtering, both procedures carried out with csound). Each sound has a unique spatial location. Separate sounds were then selected by the numerical data, along with their respective time and spatial positions, and mixed into an ambisonics b-format sound file with the software Vspace (Furse 1999). Filtering and transposing each original sound source 1600 times has the advantage of producing musical variation, while maintaining a "frame' within which the variation can operate. The separate 1600 sounds in the 40x40 grid ranged in duration from 300ms to 3000ms, and it was clear that either some decimation of the numerical data was necessary, or for the playback rate to be decelerated. The latter would result in a slower propagation of the "avalanche trajectory', and so decimation was selected as the preferred method of data reduction. After a 90% decimation, 5,000 time-location entries were used to create a 10 minute ambisonics b-format sound file. This process was repeated with different sounds as input sources to the 40x40 sound grid, and in some instances, 40 input sounds were used. The results were decoded from b-format to a loudspeaker set-up, and mixed to give further variation, purely for musical reasons. The final composition is called "Turbiditic Flow" (Barrett 2000). CONCLUSION In both compositions a reasonably complete composition methodology was based on data extracted from real-world patterns (first example) and processes (second example), even though it was sometimes necessary to stray from accurate translations of data organisation, and rely on aural judgement. In both I have attempted to capture attributes of the natural world in the short and long term organisation of the musical material. In working with the materials, careful selection of source data and mapping methods were necessary to avoid ignoring long term musical parameters at the expense of short term structures, and vice versa. By composing with sound materials inherent to the real-world source from which the data are extracted, one can describe this technique as an attempt to combine both the mimesis and the impressionistic compositional methods in one tidy parcel. REFERENCES Bak, P. 1996. How nature works. New York: Springer-Verlag. Brown, D. & Rothery, P. 1993. Models in biology: Mathematics, statistics and computing. New York: John Wiley & Sons. Davis, J.C. 1986. Statistics and data analysis in geology. New York: John Wiley & Sons. Furse, R.W.E. Vspace software. http://www.muse.demon.co.uk/vspace/vspace.html McGregor, P. K., T. Dabelsteen, C. W. Clark, J. L. Bower, J. P. Tavares & J. Holland. 1997. "Accuracy of a passive acoustic location system: Empirical studies in terrestrial habitats." Ethology, Ecology and Evolution 9: 269-286. Rossing, T. D. 1990. The science of sound. Massachusetts: Addison-Wesley Publishing Company. Xenakis, I. 1971. Formalised Music. London: Bloomington. Music references Barrett. N. 1999 Viva La Selva. Acousmatic composition. Unpublished. Barrett, N. 2000. Turbiditic Flow. Acousmatic composition. Unpublished.