|Serveur © IRCAM - CENTRE POMPIDOU 1996-2005.
Tous droits réservés pour tous pays. All rights reserved.
IRCAM's specially designed building has just been completed : however, since computers need much preparation before being operational, we have already worked for two years to implement computing facilities in a temporary site.
The computing facilities have been developed to make them available to music research in all departments in IRCAM -- in particular in the field of sound analysis, synthesis and control. We had to take into account several requirements :
Requirements 1) and 2) are somewhat contradictory if one envisions only one computer. Requirements 2) and 3) were influential in our choice of the main computer, a DEC 10 with a number of terminals, some of which could be used as stand-alone computers. The stand-alone computer can be most easily equipped so as to satisfy the real-time control mentioned in requirement 1).
We shall enumerate the equipment selected, and the special hardware that was built around it. We shall also indicate some software changes made to the monitor and to some utility programs. We shall then enumerate some of the presently available user programs, with emphasis on additions made to the Music V sound synthesis program.
A computer with good time-sharing facilities was considered desirable to facilitate access. A medium-size computer could provide a means of communication between users, fast editing and good conversational possibilities. Both for its intrinsic possibilities and to ensure compatibility with Stanford's Center for Research in Computer Music and Acoustics, we selected a DEC 10 computer as the main computer. In the IRCAM final building the computer will be accessible from-some 25 terminals, most of which will be alphanumeric display consoles with some graphic facilities. Three terminals will in fact be PDP 11 computers usable as stand-alone computers with graphic and digital storage capabilities.
Here follows the list of our present computing equipment :
One can notice that this list comprises no card reader or card punch.
We are in the process of upgrading this configuration. The core memory will be expanded to 256 K words. We shall add 2 more RP-03 disk drives and 9 more DATAMEDIA display terminals. The most important addition will consist of two DEC 11 computers, which will be linked to the 10 but which will have substantial computing and storage capabilities as stand-alone computers. These computers will be helpful as dedicated computers for the different departments in IRCAM, used for such operations as real-time sound control, on-line psychological testing and computer control of IRCAM's experimental hall (2).
Analog to digital and digital to analog conversion facilities have been added. Initially 12 bit DACs and ADCs were installed by Joseph Zingheim. Higher performance converters have been installed by David Cockerell, who designed a scheme not calling for a sample and hold circuit. At present two 16 bit DACs are operating satisfactorily (at full amplitude, the intermodulation distortion products are more than 85 dB down). Two more channels will soon be added, as well as 4 16 bit analog to digital converters : we shall thus have 4 channel high quality conversion both ways. Desampling filters have also been installed : eventually the choice of different sets of such filters will be software controlled.
The standard 10-11 link could not handle satisfactorily even simple graphic communication. A medium speed data link between the DEC 10 and the PDP 11 computer has been designed and built by David Cockerell. Data is transferred as 16-bit words under program control. This link can achieve rates of approximately 105 bauds.
David Cockerell has also designed and built several devices to be attached to the PDP 11 computers, to provide on-line manual control : keyboards (both conventional organ and unconventional), linear slide potentiometers, 3 dimensional joystick.
We mention here Di Giugno's "4 A" digital synthesizer, which can be considered a special processor for sound synthesis, as well as the new "4 B" model Di Giugno and Alles are presenting at this meeting.
Peter Eastty has designed special modifications to the Datamedia terminals, which will enable them to function as low-resolution graphic terminals for input and output.
The standard DEC monitor was substantially modified to make it more appropriate or even usable for the envisioned operation. A number of these modifications were adapted from the Stanford facility. We can classify the changes according to three categories :
Two handy features for searching and for circular searching were added to the TECO editor. A major effort to provide powerful text-editing facilities was made by Brian Harvey in cooperation with Martin Frost of Stanford University the SAIL display software has been made exportable so as to be adaptable to terminals with "insert-delete" capabilities (that is, the ability to insert or delete characters and lines without retransmitting all the following text to the computer). The Datamedia terminals are endowed with this ability : the E text editor can be used with them -- which permits viewing the text modifications as they are done. However, if the E editor is too popular with users, it could heavily load the computer.
The servicing of digital to analog and analog to digital conversion is done by programs called PLAY and REC.
Software for the Versatec graphic plotter was contributed by Larry Johnson of Michigan State University.
We can use a number of important programs, mostly adapted from Stanford University. We acknowledge here the help of the Stanford group, specially Andy Moorer, Loren Rush and Leland Smith. We share the same conventions for the headers of the binary sound files, which provide vital information on these files : sampling rate, number of channels, packing mode, etc... Among the Stanford programs, let us mention S (See), a package for display and Fourier analysis of a sound file ; MIXER and MXSND for digital mixing ; NEWMUS (formely MUS10) which is Stanford's augmented dialect of Music IV ; programs for predictive coding and the simulation of phase vocoders, and other programs for the processing of sound files.
Leland Smith of Stanford University adapted his SCORE input language so that it can be used with either NEWMUS or MUSIC V. This is a powerful input language with nice motivic capabilities, which is efficient for music coding (for subsequent analysis or printing) as well as for writing scores to be synthesized by computer.
David Wessel and Larry Johnson brought or adapted programs operating at Michigan State University on a PDP 11 computer : Music 4B for sound synthesis a program for digital filter design ; programs for graphic display of multidimensional analysis.
Bennett Smith wrote programs to concatenate sound files, and to have the computer play sound files selected by typing a terminal's key. These programs are specially useful for David Wessel's self testing procedures, described in that meeting. Smith has also written service software for the 10-11 interface, following principles suggested by Johnson.
Giovanni de Poli, from the University of Padova, adapted his input language for music (MUSICA) to the PDP 10 and to the Music V program. The language permits an easy and compact coding of conventional scores.
James R. Lawson and Max Mathews will present at this conference the software projected for Di Giugno and Alles' digital synthesizer.
While keeping the basic logic of the Music V program, John Gardner had performed substantial changes and additions to the initial program. This adaptation incorporates some of the U.C.L.A. modifications, in particular the merging of the initial Pass I and Pass II, and an increased amount of error checking analysis, most useful for inexperienced users. Also a feature was added to enable the length of each stored function to be specified as a number 32, 64, ... , 4192. The D and G arrays were combined in a single 4000 word array. Control variables (such as sampling rate, number of channels) were made distinct from freely usable variables.
The program was modified to operate best language in a time sharing system. The unit generators were mostly coded in assembly language -- where possible, the unit generators executed in the ACs on the PDP 10 for maximum speed. The notes were executed in instrument number order. An assembly 1/0 routine was written.
The logical AND was used for the oscillator : it did not require new unit generators allowing for negative frequencies.
Numerous additional unit generators were coded, including some taken from the Marseille-Luminy version of Music V (operating on a 16 bit 32 K mini computer), and some suggested by the INA-GRM Group in Paris. Among these, we shall mention single unit generators performing Chowning's frequency modulation unit generators making the output a prescribed function of the input ; several low pass, high pass, band pass filters, comb and all-pass reverberators ; a unit generator inputting digital sound files into input-output blocks, hence permitting application of Music V transformational capabilities to any sound, recorded or synthesized logical ; unit generators previously selected by the composer for execution -- this alteration could be subject to conditions similar to those encountered in Fortran conditional GO TO statements. So we can now perform conditional branching inside the instrument -- and for example compute new control parameters at a rate slower than audio rate.
Numerous small features have been added. These include the P? feature for defining scratch fields without keeping track of them, or their possible omission altogether, and the ability to define constants directly in the instrument definitions.
An important change of structure gradually evolved. The incentive was provided by both Gerald Bennett and Leland Smith. The size of various arrays in the Pass III execution are controlled by constants located in the IP array. These constants can be dynamically controlled using information gleaned from the Pass I-II phase of the program. Thus many of the limitations concerning the number of note cards playable at one time, the number of F functions, B arrays, etc... could be removed. A final step was to dynamically allocate core for the I array in Pass III, thus making these limitations a direct function of the core available, rather than a function of the constants compiled into the program. Thus the size of the functions and blocks can be changed in the score. There is no limit to the number of simultaneous voices, provided enough core is available. Also Pass I will now accept expressions both in instruments, functions and note definitions, thanks to a Polish parser (expression evaluator).
Gerald Bennett has designed a conversational tutorial for Music V taking advantage of the dialogue feature of the system. This makes the program much less intimidating to the un-initiated user.
Kip Sheeline has written a tutorial for the use of the IRCAM version of Music V.
The computing facilities have already been used for research in psychoacoustics and for producing music. However they are still in a very preliminary stage, especially as far as usage is concerned. We are aware that many modifications will have to be made as we get more experience and feedback from the users. (We already know that one must watch carefully the operation of the computer, to make popular programs efficient -- and perhaps inefficient programs less appealing. Also keeping the documentation up-to-date is important -- and hard to achieve).
Nevertheless our limited experience indicates that the facilities are useful : they allow fairly rapid response from the computer for small jobs, which is helpful for the new user and also for trial and error procedures. They provide convenient users archives, editing facilities, extended software and documentation files. Whereas the large computer permits easy access and interaction, the small computers are essential for genuine real-time processes.
Computers are general-purpose tools, but it is well known that there is no universally optimal configuration. Although the above enumeration tried to be unspecific, it is clear that some research options underlie our choices. To state them in a general way, we are fairly skeptical about letting the computer take over high-level advanced computer languages ; we believe that present artificial intelligence techniques are much to crude to be of real significance to music, a highly complex domain. It is a worthwhile challenge, however, to design interactive aids to composition, so that the computer can usefully amplify the composer's ideas or respond to them in a stimulating fashion. We believe it is important to involve the composer in the programming, so that he can tailor the program to his own needs and have other musicians benefit from his ideas.
We believe that the computer used as a tool can open new avenues to music. In particular it affords an unprecedented refinement of control of sound : although this use of the computer does not seem so ambitious, we believe that considerable potential resides here. For the first time one can envision composing the sounds themselves and not merely putting together a sequence of preexisting sounds. One can produce new material susceptible to new musical architectures. However these possibilities are far from being well explored yet. The full exploitation of this powerful potential needs substantial research at various intertwined levels involving music, science (specially psychoacoustics) and technology. We moreover think that computer sound processes will soon be widely spread thanks to the development of microelectronic technology. Hence it is essential that these processes be investigated thoroughly. Only thus will we learn to take advantage of their potential richness in musically significant ways. The computing facilities we described aim primarily at making the computer a responsive tool for these musical investigations.
Server © IRCAM-CGP, 1996-2008 - file updated on .
Serveur © IRCAM-CGP, 1996-2008 - document mis à jour le .