July 15 - 24, 1992
CERL Sound Group
University of Illinois
Urbana, Illinois
Reviewed by Donna McCabe
for The Computer Music Journal
Oakland, California USA
For ten days in July the CERL Sound Group at the University of Illinois offered an intensive workshop in sound computation to ten invited participants from all over the world. Heading the workshop was Carla Scaletti and Kurt Hebel. Workshop participants included composers, computer programmers, and persons interested in sonification of data.
Scaletti lead the workshop with basics of digital audio, timbre exploration, real-time software sound synthesis techniques, and Kyma principles. The days were filled from 9-12 with lecture demonstrations, 1-4 was lab time, and from 4-5 were special guest seminars. From 6 until the last person left the lab, participants were invited to work on assigned machine problems, or conduct independent research. CERL Sound Group members were available at all times for questions and provided ideas which aided participants in choosing and realizing projects. With their help, after only one day, people who had never used a Kyma system were exploring sound with a new powerful tool, and on the road to reaching sonic goals in exciting ways. Although we had only begun to explore, the system allowed for immediate sound synthesis, and later proved to allow the programmer extremely fine control; an excellent environment for teaching students.
In addition to Scaletti, the CERL Sound Group intensely divulged their programs and research for much of the workshop. Lippold Haken demonstrated his Lime music notation program. It can compete with the best in print quality, yet is easier to use than most commercial programs of comparable power, and cutting notation and pasting as a graphic into another program is possible. Lime notated music can be imported directly into your Kyma system and interpreted through a score program.
Kelly Fitz and Bryan Holloway presented their spectral analysis and editing program, Lemur and LemurEdit respectively. In a very understandable format, Fitz explained McAulay - Quatieri Analysis and how the MQ model is non-harmonic and can analyze a broad class of sounds. Once the sound has been analyzed and stored in the MQ format, Lemur picks peaks in the spectra. Since the MQ model assumes that the sinusoidal components have time varying characteristics, Lemur then matches these peaks from one frame to the next. It is this connection of peaks that causes the birth and death of tracks which show the time varying behavior of the sinusoidal components of the signal.
It is through Holloway's LemurEdit that graphical editing and manipulation of the Lemur tracks can be done. LemurEdit uses an x and y axis to represent time and frequency respectively, and shades of grey to represent amplitude. This is contrary to the popular x, y, z axis representation for most MQ analyses which is often difficult to read. LemurEdit allows for fundamental extraction, any harmonic or group of harmonics extraction, noise extraction etc. These edited files can then be resynthesized through Lemur.
Andrew Horner showed his research on genetic algorithm spectral matching with his program GaGA. GaGA is a useful way to gain high level control over sampled sounds. Currently, Horner's work involves synthesis techniques such as wavetable and frequency modulation synthesis to model real instruments. For Horner, the genetic algorithm proved to be an optimized routine for finding basis spectra from the original sampled tone spectra. The matching procedure tries to find the correct harmonics at the correct amplitude, which requires a sampled sound input to be harmonic for accurate resynthesis. The results were convincing, and proved much less memory consumptive than storing samples.
Bill Walker and Sal Martirano showed us their newest work in progress, the ImprovisationBuilder; an interactive composition tool. Walker has done research on conversation patterns, psychology and linguistics to design a system of musical improvisation as conversation that incorporates signal and event level control over sound. The goal seems to be to allow the user to have a system as sensitive as the much earlier Sal-Mal Construction (weighing 1,500 pounds with a 24 channel speaker system under automated and performer control), but more portable (the Macintosh and Capybara weigh less than 100 pounds). The ImprovisationBuilder uses a conversation-based framework for improvisation, where both performers (the musician and the computer) must follow the basic requirements for improvising. Determining these basic requirements for improvising are what Walker is currently working on.
Richard Baraniuk presented a rousing lecture on time-varying spectral analysis including STFT and Wavelets. Baraniuk presented his research on the time varying frequency content of a signal and how it evolves over time. Baraniuk was quick to point out how an FFT looses temporal information relating to a signal. With a STFT he is able to get an instantaneous spectrum at any time during a signal. By taking numerous STFT's and stacking them along the time axis, an extremely accurate spectogram of the signal is obtained. Due to certain resolution tradeoff's (short windows produce accurate time information, but frequency information gets distorted; long windows have good frequency information, but less accurate time information), wavelet transforms became the next part of his research. With wavelets, time-frequency resolution properties are frequency dependent. Wavelets produce poor time resolution but good frequency resolution at low frequencies, but good time resolution and poor frequency resolution at high frequencies. Between both analysis techniques, Baraniuk notes that there is no best way to analyze a sound. Neither wavelets nor STFT can provide simultaneous good time and good frequency resolution. Baraniuk cleared up a lot of the mysteries of wavelet transforms and the differences between wavelets and STFT.
With the power of the Kyma system just beginning to be tapped, Kurt Hebel delved into SmallTalk-80 object-oriented programming and Motorola 56001 assembly language for the hard-core programmers in the group. This allows the Kyma system to be flexible and expandable, yet learnable in small increments. Through SmallTalk-80 the user is allowed to write algorithmic programs into "aScoreLanguague" which can control any subsound; ie. a timbre, a pitch, a duration. These "scores" parameter values can be any object and they can pass information into other score programs, making for complex algorithms which can be broken down into smaller programming steps. I found this helpful in debugging and allowed me to expand upon my initial ideas quite easily.
The DSP programming allows the Kyma user to get at the microstructure and optimize the processing tasks to the 56001. Lippold Haken showed his ongoing Sum of Sines (SOS) work on resynthesis from Fourier analysis. In order for his program to compute hundreds of sine waves at the same time, he has interleaved his assembly code and has the 56001 computing as many as three instructions at once through excellent register assignments.
Julie Baca, Brian Evans, Richard Misenheimer and Stuart Smith gave sonification examples and had an open discussion of concerns and problems with the discipline. Although sonification has not been widely used, it was apparent from the demonstration that the sonification of some data could allow for a clearer representation than with visuals alone. If this complex data were clearly represented, people other than the scientists that collected it would have a better understanding of the research. The Kyma system allows mapping of data streams to any parameter of a Kyma sound which makes it a powerful sonification tool.
Alan Craig, in the Virtual Reality Lab, showed he and Scaletti's real time sonification project with the Kyma File Interpreter. A sensor, attached to the top of a person's head, sends out laser homing signals. The data from this sensor was being fed into Kyma and mapped to a basic sound. These beeps were then heard instantaneously by the subject via headphones, which alerted them how close to the receiving pole they were. In many ways it acted like a Geiger counter measuring radioactivity; perhaps the most well known sonification application. The beginnings of an interesting project which the two continue to work on.
As part of the ongoing Monday night concert series put on by the Music Department of the University, several invited composers from the workshop had pieces performed on the Composer Forum concert. Tape pieces were heard from Brian Belet, Bruno Liberda, George Todd and Dick Robinson; with a tape and two piano piece from Donna McCabe.
Invited guests who presented special seminars were Sever Tipei, Chris Kriese, Sal Martirano, John Melby, James Beauchamp, Herbert Brun, Keith Johnson and Alan Craig.
Final projects were as varied as the participants, demonstrating the versatility and far ranging approaches available through the Kyma system. From sonic tic tac toe, to compositions, to live performance environments, workshop participants spent ten days at Sound Computation Boot camp, which opened a new world of questions and explorations. Next year promises the Third Annual Intensive Sound Computation Workshop. An excellent job by Carla Scaletti, Kurt Hebel, and the CERL Sound Group.