Sever Tipei
The term "manifold composition", introduced in 1989, defines all actual and potential variants of a musical work generated by a computer which:
1. runs a program containing elements of indeterminacy
2. reads essentially the same data for each variant
A unlimited number of compositions belonging to the same equivalence class can be produced in this way. The members of a manifold composition are variants of the same piece; they share the same structure and are the result of the same process, but differ in the way specific events are arranged in time. A manifold composition is somewhat similar to the serigraphs produced by a visual artist, except that its individual members may be more distinct from one another. The global effect depends on the coherence with which the control parameters are chosen. For example, the members of a manifold composition may share certain fixed sections, while other sections diverge in different degrees.
Composed with Manifold.f, a computer-assisted composition program, the ANL-folds exemplifies this situation. This 2:30 minutes piece, starts with a rich, low sound followed by an irregular, ascending arpeggio made out of its own partials. The low (60 Hz) sound is always the same but the exact order and duration of the notes in the arpeggio varies from version to version. The ending of the piece mirrors the beginning with a descending arpeggio and a low 60 Hz tone; this time too, the order and duration of sounds is slightly different each time. In the middle of the piece, a seven sound pattern, the Argonne Chime, a musical spelling of the word ARGONNE (A; R (re) = D; G; O (sol) = G; N,N = non pitched, percussive; E) is always heard exactly the same and exactly at the same moment. In all the other sections however, pitches, durations, attacks, and spectral changes follow stochastic distributions and every variant produced has its own distinct profile. Even broad characteristics such as density of texture, overall dynamics, or the use of particular effects may vary. These variations derive exclusively from the use of a new seed for the random number generator each time.
The ANL-folds tapes were generated with DIASS, a Digital Instrument for Sound Synthesis designed by Sever Tipei and Christopher Kriese. DIASS consists of an editor through which the composer inputs data (implemented by Christopher Kriese, David Ralley, Cheryl Herdon and Arun Chandra) and an M4C instrument. The instrument and portions of the M4C code required by it were parallelized by Dave Blumenthal, Mario Lauria, Thomas Lawrence and Scott Pakin and run on the IBM Scalable POWERparallel System (SP) of the Argonn National Laboratory.
Manifold compositions such as the ANL-folds represent idiomatic ways of using computers in music composition by mass producing slightly different and, at the same time, unique versions of the same archetype. A version cannot be performed in public more than once, thus stressing the ephemeral quality of any musical activity.
Some of the ANL-folds variants were performed in the CAVE, both at the Argonne National Laboratory and at National Center for Supercomputing Applications of the University of Illinois. The CAVE is a room size virtual environment where images are projected in stereo and users wearing stereo glasses can see in three-dimensional space. A set of ANL-folds were also presented at the November 1996 Supercomputing '96 conference in Pittsburgh, PA. on the IMMERSADESK, a scaled down, two-dimensional version of the CAVE.
In the demonstration, images are made to correspond exactly to sounds through a one-to- one mapping between visual attributes and sound qualities. First of all, partials are represented as spheres; their size is proportional to loudness, their height corresponds to frequency and an optional grid in the background shows the octaves divided into twelve equal increments. A sound's position in the stereo field determines the placement of the sphere between far left and far right in the room. Visual objects display a shimmering when tremolo or vibrato is applied and their color varies from deep blue to light pink when reverberation is present. In a separate and slightly less rigorous rendition, a cloud made out of a large number of confetti-like objects changes color and degree of internal activity according to the same musical parameters.
Although more sophisticated demonstrations could be devised, this experiment in the visualization of music is a significant attempt at creating both aural and visual images from the exact same data in stead of using sound in a kind of a loose commentary to strictly derived visuals. In the present case the sound is richer than the visual domain and reflects more degrees of freedom encoded in the data. But the M4CAVE application (implemented by Dave Blumenthal and Max Levchin) which produced it allows both the changing of the mapping between parameters (as shown by the two representations described above) and the inclusion of new such degrees of freedom.
These features coupled with the availability of a supple and resilient digital instrument, DIASS, make possible the next logical step: the analysis of complex scientific data through sonification as a complement to visualization. In the case of the ANL-folds, a data set, representing a piece generated with a computer-assisted composition program, was successfully translated into both aural and visual images. The same tools could be used to do similar translations into sounds and images of data resulting from scientific computations.
Both the parallel version of DIASS_M4C and the M4CAVE program were written as class projects for Music 320 D, "Musical Applications on Supercomputers" offered every sememster. A discovery course, Music 199 T, "Music, Science, and Technology", scheduled for Fall 1997 will discuss sonification, visualization of music and manifold compositions among other topics.