Gil Weinberg Interconnected Musical Networks: Toward a

Weinberg

Gil WeinbergMusic DepartmentGeorgia Institute of Technology840 McMillan St.Atlanta, Georgia 30332 [email protected]

Interconnected MusicalNetworks: Toward aTheoretical Framework

23

This article attempts to define and classify the aes-thetic and technical principles of interconnectedmusical networks. It presents an historical overviewof technological innovations that were instrumen-tal for the development of the field and discusses anumber of paradigmatic musical networks that arebased on these technologies. A classification of on-line and local-area musical networks then leads toan attempt to define a taxonomical and theoreticalframework for musical interconnectivity, address-ing goals and motivations, social organizations andperspectives, network architectures and topologies,and musical content and control. The article con-cludes with a number of design suggestions for thedevelopment of effective interconnected musicalnetworks.

Interdependent Music Performance

Music performance is an interdependent art form.Musicians’ real-time gestures are constantly influ-enced by the music they hear, which are recipro-cally influenced by their own actions. In groupplaying, the interdependent effect bears unique so-cial consequences such as the formulation of lead-ers and followers or changes in individual players’dynamics and timing in correlation to group syn-chronization (Rasch 1988). Other manifestations ofinterdependent group routines can be found in a va-riety of musical genres such as Western chambermusic, Jazz, Gamelan, Persian music, and others(see details in Weinberg 2002). Performers oftenaddress the importance of interdependent group col-laboration and sharing in their music. Jazz per-former Milt Hinton noted, “I was pretty youngwhen I realized that music involved more than play-ing an instrument; it’s really about cohesivenessand sharing . . . I’ve always believed you don’t trulyknow something yourself until you can take it fromyour mind and put it in someone else’s” (Hinton

and Morgenstern 1988). Cognitive scientists, on theother hand, tend to address the perceptual aspects ofinterdependent group play. William Benzon (2001)discusses his experience playing Ghanaian Bells in agroup of four: “melodies would emerge that no onewas playing . . . it arose from cohesions in the shift-ing patterns of tone played by the ensemble. . . .Occasionally, something quite remarkable wouldhappen. When we were really locked together inanimated playing, we could hear relatively high-pitched tones that no one was playing.” Benzonuses this example to strengthen his definition ofmusic as “a medium though which individualbrains are coupled together in shared activity.”

But although acoustic-interdependent modelsprovide an infrastructure for a variety of approachesfor interconnections and interdependencies amongplayers, they do not allow for actual manipulationand control of each other’s explicit musical voices.Only by constructing electronic (or mechanical)communication channels among players can partic-ipants take an active role in determining and influ-encing not only their own musical output but alsothat of their peers. For example, consider a playerwho, while controlling the pitch of his or her owninstrument, continuously manipulates the timbreof a peer’s instrument. This manipulation will prob-ably lead the second player to modify his or her playgestures in accordance to the new timbre that wasreceived. These modified gestures can then be cap-tured and transmitted to a third player, influencingthis player’s music playing in a reciprocal loop. An-other example is a network that allows players toshare musical motifs with other members of thegroup. By sending a motif to a co-player who cantransform it and send it back to the group, partici-pants can combine their musical ideas into a con-stantly evolving collaborative musical outcome.

The shape of the composition in such systemsgrows from the topology of the network and its in-terconnections with the performers. Such an envi-ronment that responds to input from individuals ina reciprocal loop can be likened to a musical “ecosys-tem.” In this metaphor, the network serves as a habi-

Computer Music Journal, 29:2, pp. 23–39, Summer 2005© 2005 Massachusetts Institute of Technology.

tat that supports its inhabitants (players) through atopology of interconnections and mutual responsesthat can, when successful, lead to new breeds ofmusical life forms. These interdependent connec-tions can bring a wide range of new experiences intomusical group playing. For example, a soloist canguide his or her collaborators with a simple interde-pendent touch toward a musical idea in which thesoloist is interested, or change a supporting voiceinto a contrasting one so that a desired musical ideawill become clearer. Players can shape their peers’accompaniment line so they would lead toward anew direction when the current one is exhausted,send a motif to other players who can manipulate itand send their variations back to the group, have amusical response accentuated by the player whosent the original call, plant a musical “seed” thatwould be picked up by the group in various man-ners, etc. Musical networks, therefore, bear thepromise of using technology to enhance the socialcontext of music performance and enrich its socialritual roots.

Technological and Historical Landmarks

The development of interconnected musical net-works during the last half century is closely relatedto several technological developments that occurredduring that time. In particular, I see four major in-novations—analog electronics, the personal com-puter, the Internet, and alternate controllers—asprincipal enablers for the various approaches takenfor interdependent musical connectivity. Whenthese technologies became widespread and com-mercially available, they inspired musicians whowere looking for new ways to expand the vocabu-lary of socio-musical expression.

Analog Electronics: The Early Musical Networks

John Cage was one of the first to notice the expres-sive potential that lies in using technology to en-hance musical group interdependency by treatingthe then-recently invented commercial transistorradio as a musical instrument that could be used to

provide a sonic medium for interdependent proce-dures, rules, and processes. Cage’s compositions fortransistor radios allowed, for the first time, for anexternal entity (audio steams from a set of radio sta-tions) to generate and support evolving and dynamicmusical contexts, providing a first crude glimpse atthe concept of musical networks. Cage’s 1951 Imag-inary Landscape No. 4 for twelve radio transistorsplayed by 24 performers can be considered the firstelectronic interdependent musical network. Thecomposition’s score indicates the exact tuning andvolume settings for each performer but with noforeknowledge of what might be broadcast at anyspecific time or whether a station even exists at anygiven dial setting. Inspired by the Chinese book oforacles, the I Ching, Cage demonstrated his fascina-tion with chance operation, allowing players to con-trol only partial aspects of the composition, whiletechnology, chance, and performers together deter-mined the actual audible content. The role of Cageas a composer was narrowed down to setting thehigh-level blueprint of dial-setting instructions.

The interdependency in the piece was manifestedin two planes. First, there were the interdependentinteractions between the players and the network ofradio stations that provided unknown and dynamicmusical content. But the system also supported intra-player interdependencies, as for every frequency-dial player there was a volume-dial player whocould manipulate the final output gain, controllinga full continuum from complete muting to maxi-mum amplification. The volume-dial players, there-fore, had a significant impact on their peer’s musicaloutput, as they could control anything from render-ing their actions inaudible, through blending themsmoothly in the mix, to boosting them as a scream-ing solo.

The explorations of the transistor radio as an in-frastructure for interdependency opened the door tofurther experimentation with interdependency. InCartridge Music (1960), for example, Cage made hisfirst attempt at a musical network focused on tactilegeneration of sounds and intra-player, amplification-based interdependencies. Here, players were in-structed to pluck small objects (such as toothpicksor pins) that have been put into a gramophone car-tridge and to hit larger objects (such as chairs) that

24 Computer Music Journal

Weinberg

were amplified with contact microphones. Thesimple, intra-player interdependency was generatedowing to other players who controlled the ampli-fiers’ volume knobs, leading, again, to a wide dy-namic range of output from muting to soloing. OnCartridge Music, Cage remarked: “I had been con-cerned with composition which was indeterminateof its performance; but, in this instance, perfor-mance is made indeterminate of itself.” Althoughrevolutionary, the level of interdependency inCage’s early experiments were constrained by thecrude nature of the technology, where, in effect, theonly possible direct interpersonal connections werelimited to coarse gain manipulations.

More elaborate attempts at analog interdependen-cies were made by composers such as Stockhausen,who in Mikrophonie II (1965), for example, routedthe sound from four choruses and a Hammond or-gan to modulate each other’s spectra through asingle ring modulator, or David Tudor, who in Rain-forest IV (1968) instructed players to attach micro-phones and contact loudspeakers to various objectson stage to create resonance-based feedback loops.The analog synthesizer provided an inspiring infra-structure for experimentalists such as David Rosen-boom (1976) to use biofeedback methods forinterconnecting groups of players to generate syn-thesized sounds, as well for popular music groupssuch Tangerine Dream to interdependently manipu-late multiple synthesized sound parameters.

The Personal Computer: The Digital Advantage

The next significant breakthrough in technology fordetailed and controllable interdependent networkswas achieved thanks to the commercialization ofthe personal computer. One of the first commercialcomputers that was used for creating fine-tuned andconfigurable network topologies was the 1976 Com-modore KIM-1. The League of Automatic MusicComposers, a group of musicians from Oakland,California, that included John Bischoff, Jim Horton,and Rich Gold (later replaced by Tim Perkis) wasone of the first to use a number of such KIM-1s towrite interdependent computer compositions (seeFigure 1.)

Each member in the group was able to send andreceive data from and to his personal composition,which ran on his personal networked computer.The group named this new genre of music perfor-mance that allowed programmable and detailedmusical interconnections “Network ComputerMusic.” In their 1978 performance in Berkeley, Cal-ifornia, for example, the group set up a three-nodenetwork, mapping frequencies from one computerto generate notes in another, or mapping intervalsfrom one composition to control rests and rhythmicpatterns in another. On this performance, the groupwrote: “[W]hen the elements of the network are notconnected, the music sounds like three completelyindependent processes, but when they are intercon-nected, the music seems to present a ‘mindlike’ as-pect” (Bischoff, Gold, and Horton 1978). The Leaguecontinued to work until 1986 when it evolved intoan offspring group, The Hub, which employed moreaccurate communication schemes by using MIDIand central computers to facilitate the interaction.The Hub also experimented with more hierarchicalsystems, such as in Waxlips (1991), where a “leadplayer” sent cues to initiate new sections and tojump-start processes by “spraying” the networkwith requests for note messages. The Hub expandedtheir explorations to other areas such as remote col-laboration and audience participation. In their first1985 remote networking effort, the group was di-vided into two sites and communicated via tele-phone lines. Other pieces such as HubRenga (1989)attempted to involve the general public in on-line,remote, interdependent interactions. These earlymusical networks introduced the computer as a ver-satile and resourceful partner for interconnectedgroup interaction. But the technology at that timecould not yet support large-scale on-line interac-tions, nor was it designed to address novices with awide range of musical backgrounds. Scot Gresham-Lancaster, a Hub member, reflects on the Hub’s re-mote online experiments: “[T]he technology was socomplex that we were unable to read a satisfactorypoint of expressivity” (Gresham-Lancaster 1998).Regarding the Hub’s attempts to allow audienceparticipation, he writes: “The varying range of tasteand innate talent made for a pastiche that lacked fit-ness and cohesion, and despite the best intentions

25

of the contributors, the results were mixed.” In thenext two sections I present more recent approachesthat were taken by musicians and researchers to usenew technologies in an effort to allow for large-scaleonline musical networks as well as novice and audi-ence participation.

The Internet: Approaches for Online Networks

Advanced Internet protocols and on-line applicationstoday provide a faster and more reliable platform forlarge-scale interconnected musical networks. Thesedevelopments helped facilitate the recent prolifera-tion of on-line musical networks, offering a varietyof musical activities for diverse audiences. Severalrecent studies attempt to classify and categorize on-line networks based on concepts such as location,media, timing, physicality, and other factors (see,

for example, Barbosa 2003). Here, I attempt to mapthe field based on what I see as the central innova-tive concept of the medium: the level of intercon-nectivity among players and the role of thecomputer in enhancing the interdependent socialrelations. Based on these criteria, I have identifiedfour different approaches and have named themthe Server, the Bridge, the Shaper, and the Construc-tion Kit.

The Server Approach

This simple approach uses the network merely as ameans to send musical data to disconnected partici-pants and does not take advantage of the opportu-nity to interconnect and communicate amongplayers. Participants in such a client/server configu-ration cannot listen to or interact with their peers,and the musical activities are limited to the com-


Figure 1. An Early MusicalNetwork: The League ofAutomatic Music Com-posers. Photo by PeterAbramowitsch.

Weinberg

munication between each player and the centralsystem. A typical example of the Server approach isthe Sound Pool Web application, which is part ofthe interactive piece “Cathedral” by William Duck-worth (Duckworth 1999). Here, a Beatnik-basedJava applet allows individual players to triggersounds by “accidentally or randomly” clicking onhidden nodes on the screen. The interaction occursindependently in each player’s browser so that“each user can create his or her own unique experi-ence.” The original sounds in the piece were com-posed by Duckworth, but users can contribute theirown sounds to the mix. Because there are no con-nections between participants, the system can sup-port any number of users.

The Bridge Approach

The motivation behind the Bridge approach is toconnect distanced players so that they can play andimprovise as if they were in the same space. Unlikethe Server approach, musical collaboration can oc-cur in such networks because participants can lis-ten and respond to each other while playing.However, the role of the network in this approach isnot to enhance and enrich collaboration but to pro-vide a technical solution for imitating traditionalgroup collaboration. Aspects of bandwidth, simul-taneity, synchronization, impact on host computer,and scalability are some of the challenges that areusually addressed in this approach. A characteristicexample of the Bridge approach is the “DistributeMusical Rehearsal” project (Konstantas, et al. 1997),which focused on remote conducting. Using videostreaming and a three-dimensional sound system,an ensemble of six players in Geneva was connectedto a conductor in Bonn in an effort to rehearseDèrive by Pierre Boulez. The system’s goal was“giving the impression to the participants that theyare physically in the same room,” and the mainchallenges were minimizing transmission delay andaccurately reproducing the sound space by usingmultiple microphones and a dummy head. TheTransMIDI system (Gang, et al. 1997) addressed asimilar challenge, but instead of sending audio, thesystem used the more efficient MIDI protocol thathelped minimize latencies. By using the “Transis”

group communication system, TransMIDI also al-lowed easy arrangement of multicast groups so thata “conductor” player could determine exactly whateach participant heard at any time. Here, too, thesystem was aimed at bridging the distance betweenremote participants, allowing them to play, impro-vise, and listen to music in a way similar to a tradi-tional “jam session.”

The Shaper Approach

In the Shaper approach, the network’s central sys-tem takes a more active musical role by algorithmi-cally generating musical materials and allowingparticipants to collaboratively modify and shapethese materials. Although players in Shaper net-works can continuously listen and respond to themusic that is modified by all participants, the ap-proach does not support direct algorithmic interde-pendencies among players. One of the first attemptsat this approach was the Palette (Yu 1996), which al-lowed on-line players to control aspects such as“style,” “coherency,” and “energy” of MIDI eventsthat were generated based on input from other on-line players.

Another example of the Shaper approach can bedemonstrated by the Pazellian application (Pazel, etal. 2000), a Web-based application that uses “SmartHarmony,” an algorithmic mechanism that anno-tates each note with harmonic information anddetermines a set of harmonic constraints for thecomposition. Here, players could control parame-ters such as pitch range, volume, and instrumenta-tion as well as manipulate multiple individualparameters for all voices in the composition. Play-ers could hear and respond to the musical outputthat was generated by all the participants, but theycould not directly communicate with any specificplayer. The “Variations for WWW” project (Yamag-ishi 1998) took a similar approach. In this system, aMax patch was connected to the Web via the W pro-tocol so that remote users could manipulate param-eters in an algorithmically generated theme. TheMax patch sent MIDI commands to a MIDI synthe-sizer, which transmitted the audio output back tothe participants via a Real Networks audio encoder.The system’s interconnectivity was derived from its

27

ability to play the combined manipulation of allusers back to the participants, who could modifytheir musical contribution in response. A recentvariation on this approach is the PIWeCS project(Whalley 2004) in which participants can shape andmanipulate samples of Maori instruments using amulti-agent system. Here, too, the focus is not ongenerating original material but on modifying exist-ing musical content.

The Construction Kit Approach

This approach offers higher levels of interconnectiv-ity among participants, who are usually skilled mu-sicians, by allowing them to contribute their musicto multiple-user composition sessions, manipulateand shape their own and other players’ music, andtake part in a collective creation. Interaction insuch networks is usually centralized and sequentialas participants submit their pre-composed tracks toa central hub and manipulate their peers’ materialoff-line. Faust Music On Line (Jordà 2002) is a repre-sentative example of this approach. Here, a Web-basedsynthesis engine allows players to create musicaltracks and construct them into a composition thatthen can be downloaded by other participants. If thedownloaded composition is not complete (i.e., itstill has empty tracks) a participant can generatenew tracks locally, add them to the composition,edit them, and upload the full piece back to theWeb. Participants can also reprocess and distort anyof the previous tracks in the composition by using avariety of synthesis generators and modifiers.

The WebDrum application (Burk 2000) demon-strates a slightly different take on the ConstructionKit approach by basing the application on a tradi-tional drum-pattern editor in which users turnnotes on or off in a grid. Synthesized drum soundsare used to avoid downloading large audio samplefiles. Web users can play and listen to others partici-pants’ edits and add their instrument sounds totheir own pallets.

Several projects attempt to combine the Con-struction Kit and the Shaper approaches. The ISXproject (Helmuth 2000) allows users to algorithmi-cally change their peers’ sounds as well as to createnew tracks from scratch and shape them into a col-

laborative composition. The project uses Internet2as a wideband platform that can support the ex-change of large audio files. The more recent Auraclesystem by Max Neuhaus (Freeman, et al. 2004, seewww.auracle.org) analyzes vocal gestures from on-line players in real-time and uses these gestures tosynthesize new sounds that can be heard by otherusers and added to in a synchronous multi-track for-mat (see Figure 2).

Alternate Controllers: Gestures, Novices, and Real-Time Local Networks

Early musical networks tended to focus on complexinterconnections among participants, often leadingto “high-art” musical products that were not gearedtoward wide audiences. These networks posed highentrance barriers for players by requiring specializedmusical skills and theoretical knowledge in order totake part in and follow the interaction in a mean-ingful manner. The problem of interaction co-herency is accentuated in Internet-based musicalnetworks that cannot support clear real-time ges-tural performance cues. It is clear that a graphicaluser interface cannot replace the personal, unmedi-ated connection provided by tactile interaction


Figure 2. An online net-work: a screenshot fromAuracle.

Weinberg

with physical instruments in a local space. In an ef-fort to facilitate expressive and more conveyable in-terconnected musical experiences, researchers areattempting to address novices and general audi-ences by simplifying and restricting the interactionand by using physical controllers, sensors, and ges-tural input in local spaces. I categorize these localnetworks into two groups: small-scale systemsand large-scale systems. The two categories differin the design challenges they impose and the ap-proaches that designers are taking to address thesechallenges.

Small-Scale Local Systems

I define small-scale collaborative musical networksas those that support three to ten players in closeproximity, which allows for detailed and subtle in-terpersonal interactions not possible with large-scale systems. Toshio Iwai’s Composition on theTable (1998) is a representative example of an effec-tive small-scale collaborative musical network fornovices. The network is composed of four tableswith various controllers such as switches, dials, andsliding boards that players can manipulate to con-trol sound and projected images. In one of the appli-cations, a grid is projected on the table, and playerscan direct animated objects by setting arrows ateach node on the grid. When an object hits a node, aMIDI note is played so that interlocking loops andrhythms can be generated and controlled by the par-ticipants. The experience promotes collaborationwhen players follow each other’s gestures and try topredict the object’s movement, and therefore themusic that will be generated.

Chris Brown’s “Talking Drum,” a local-area-network music installation (Brown 1999), is a col-laborative system that attempts to address moreskilled players by promoting thoughtful and intricatemusical collaborations. Here, four computerizedstations are installed in a large room or outdoors.Computer players (using MIDI instruments or com-puter mice) as well as acoustic instrument perform-ers (playing to a microphone and an electronic pitchfollower) interact with software that uses a geneticalgorithm to create rhythmic units. The softwareresponds to various aspects of users’ playing (such

as timing, loudness, density, and pitch) by changingparameters in the algorithms so that the rhythmicunits are shaped by the players. The central systemin this network (which runs on one of the four sta-tions) coordinates timing and synchronization be-tween stations.

Another approach for a gestural, collaborativemusic performance is taken by Sensorband (Bongers1998), a group of three musicians that includes Ed-win van der Heide, Zbiniew Karkowski, and AtauTanaka. The group has built the “Soundnet,” a giantweb, measuring 11 × 11 meters, strung with thickshipping ropes. The trio performs on the instrumentby climbing it as a set of stretch sensors at the endof the ropes measure the movements and send thedata to control processing of recorded natural sounds.The instrument was purposely made too large forone person to master thoroughly, and the ropes“create a physical network of interdependent con-nections, so that no single sensor can be moved in apredictable way that is independent of the others.”Some recent experimentations have also been madeat a hybrid approach, which attempts to combine lo-cal and online networks. The reacTable (Jordà 2003),for example, is table-based multi-user instrumentthat can be played collaboratively both off- and on-line at the same time.

A set of local musical networks was developed bythe author as part of his research into music inter-dependency. The Musical Fireflies (Weinberg, Lack-ner, and Jay 2000) are wireless rhythmic instrumentsthat can be synchronized when an infrared link isestablished between them. Players can record rhyth-mic patterns and trade timbres and rhythms witheach other. The collaborative interaction introducesplayers hands-on to complex musical concepts suchas polyrhythm. The “Squeezables” instrument (Wein-berg and Gan 2001) presents a different approach fora local, interdependent musical interaction. The in-strument, composed of six squeezable and retract-able gel balls mounted on a small podium, allows agroup of players to perform and improvise musicalcompositions using a set of squeezing and pullinggestures. The activity level of the “accompanimentballs” is measured and mapped to influence themelody ball’s pitch and timbre. The melody playercan control the accompaniment players’ level of

29

influence through his or her own squeezing andpulling gestures. The Beatbugs network (Weinberg,Aimi, and Jennings 2002) features more advancedcollaborative rhythmic interaction. Here, playersuse instruments called Beatbugs to enter rhythmicmotifs that can be sent and shared by their peers.Receiving players can develop the motifs in realtime by ornamenting the rhythm and the melodiccontour through the manipulation of two bend-sensor antennae, which leads to the creation of aconstantly evolving, multi-player, collaborativemusical composition (see Figure 3). A more recentlocal network installation is “Voice Network”(Weinberg 2004). Here, players can record, trans-form, and share their voices in a group. A centralcomputer system facilitates the interaction as par-ticipants interdependently collaborate in developingtheir vocal motifs into a coherent musical composi-tion (see Figure 4).

Large-Scale Local Systems

I define large-scale musical networks as systemsthat are designed for more than ten participants.Here, the details and subtleties of individual contri-butions are often hidden by the large quantities ofparticipants. The central system, therefore, oftenfocuses on analyzing the large-scale group interac-tion patterns and coordinating the multiple inputsources into a meaningful musical outcome. One ofthe earliest attempts at creating a large-scale collab-

orative musical system for novices was Tod Mach-over’s Brain Opera (see brainop.media.mit.edu).In this project, audience members were able to ex-periment with a number of new instruments suchas the “Rhythm Tree,” which included dozens ofdrum pads wired to trigger percussive sounds andword fragments; the “Gesture Wall,” where visitors’body movements were captured to control the mu-sical output; and the “Singing Tree,” which manip-ulated MIDI-based accompaniment in correlation tothe “pureness” of participants’ singing. The physi-cal and intuitive operation of these instruments al-lowed almost any visitor, from children to seniorcitizens, to take part in an expressive interactionwith the electronic sound. The Brain Opera instru-ments, however, were not designed to communicatewith each other, and it was left for the players to co-ordinate their actions if so desired.

An example of a computer-coordinated, large-scale local musical network is found in Feldmeier,Malinowski, and Paradiso (2002). Here, players usea set of low-cost, wireless motion sensors that allowfor a large group of participants (up to hundreds) tocontrol and influence the music that they are danc-ing to. The system does not identify each performerbut measures and reacts to the characteristics of thegroup in general. Algorithms based on temporal andfrequency analysis of the data are used for detectinggroup behavior and mapping it to the generated mu-sical material. Aspects such as tempo, layers, rhyth-mic complexity, timbre, and register are controlled


Figure 3. Using the Beat-bug handheld controller,players create, share, andinterdependently modifyrhythmic patterns in agroup.

Figure 4. Voice Networks-participants record andinterdependently trans-form their voices to createcollaborative vocal compo-sitions.

Weinberg

and manipulated in correlation to the level of activ-ity of the group. Results show that groups are moreactive and synchronized when controlling the mu-sic as opposed to a non-interactive control group.

A different, more centralized approach for a large-scale musical network is Golan Levin’s Dialtones:Telesymphony (2001; see www.flong.com/telesymphony). Here, the musical material was pre-composed and generated by orchestrating thedialing and ringing of audience members’s mobilephones. The composer downloaded ring tones intoparticipants’ cell phones, registered their numbers,and set the participants in a grid of 10 × 20 mem-bers. During the concert, performers called eachother’s cell phones in an orchestrated manner, andplayers were asked to raise their cell phones whenbeing called to help represent the interaction andthe music to viewers and other participants. A pro-jected grid helped audience and performers followthe activity, which led to a coherent musical out-come. But to support this coherency, players in Di-altones were stripped of any meaningful musicalcontribution and were essentially used as a grid ofspeakers for the composer’s musical ideas. In thissense, the work demonstrates the difficulty in creat-ing a coherent musical outcome while still allowinga large group of players to meaningfully participatein the musical activity.

Toward a Theoretical Framework of MusicalInterconnectivity

Informed by the historical review and the classifica-tion framework proposed above, I here attempt toidentify the theoretical and practical principles thatare at the core of musical interconnectivity. I believethat the taxonomy that I propose may also be usefulfor musical network designers in thinking and com-municating their ideas about the field. My investi-gation is based on a number of anchoring questions,including “Why?” (what are the goals and motiva-tions for designing and participating in a musicalnetwork?), “How?” (what are the different socialperspectives, architectures, and network topologiesthat can be used to address these goals and motiva-tions?), and “What?” (what are the musical parame-

ters and interdependent algorithms that can be usedin the network, filling the architectural form withmusical content?). As part of this analysis, I will ad-dress the advantages and disadvantages of a varietyof approaches taken by musicians and designers andwill suggest a scheme for maintaining a well-balanced system that maximizes the benefits.

Goals and Motivations

The definition of “Interconnected Musical Net-works” as live performance systems that allowplayers to influence, share, and shape each other’smusic in real-time (Weinberg 2002) suggests thatthe network should be interdependent, dynamic,and function as a facilitator of social interactions.An investigation of the rationale behind designingand participating in such musical networks leadsto a further classification of these motivations intotwo major network categories: process-centerednetworks and structure-centered networks. Thisdifferentiation can be related to the tension thatemerged in the midst of the 20th century betweenthe radicalization of musical structure and com-poser control, practiced mainly by “avant-garde”and “post-serialist” composers such as KarlheinzStockhausen and Pierre Boulez on one hand, and theescape from structure toward “process music” aswas explored mostly by American experimentalistssuch as John Cage and Steve Reich. As opposed tothe European movements that emphasized com-poser control over almost every aspect of the com-position, process music came from the belief thatmusic can be a procedural and emergent art formand that there are many ways of handling formother than constructing structures in different sizes.In such procedural process-based music, the com-poser sacrifices certain aspects of direct control tocreate an evolving context by allowing rules (inclosed systems) and performers (in open ones) to de-termine and shape the nature of music. John Cageaddressed this tension referring to his own experi-ence: “I was to move from structure to process,from music as an object having parts to music with-out beginning, middle or end, music as weather”(Cage 1961).

31

The use of technology in musical networkspushes the tension between structure and processmusic further into an experience where predeter-mined rules and instructions, combined with im-provised interdependent group interactions, lead toevolving musical behaviors, giving a new meaningto Cage’s exploration of unpredictability, chance de-termination processes, accidents, and contextualmusic. In my analysis, I will accentuate the differ-ences between these categories for argumentativeclarity. It should be noted, though, that most inves-tigated networks combine process and structural el-ements in different levels and balances.

The focus in process-centered networks, as can beseen in most of the local networks for novices de-scribed above, is on the player’s experience, whetherit is social, creative, or educational (see Figure 5).For designers of such networks, the musical out-come of the interaction is usually less importantthan the process participants undertake while creat-ing this outcome. Some process-oriented networks,such as the Beatbug network, focus on facilitatingelaborate social dynamics between players; others,such as Composition on a Table, emphasize the ex-pressive and creative process for individual players;and still others, such as the Musical Fireflies, centeron providing a learning experience. The interactionin process-centered systems can be further classi-fied into two additional subcategories: exploratoryand goal-oriented interactions. Exploratory net-works do not impose specific directions or goals forthe participants. These systems are driven by moti-

vations such as the investigation of novel ways forplaying in a group or the creation of unexploredmusical crossbred offspring (as with the League ofAutomatic Music Composers). Goal-oriented inter-actions, on the other hand (such as in the Feldmeier’ssynchronized dancing network), are designed to en-courage players to achieve specific objectives, musi-cal or non-musical. Such activities can be designedto reward participants for their social skills, musicalcreativity, or learning achievements and can offertasks that are based on encouraging collaboration orcompetition.

In structure-based systems, on the other hand, themain goal of the interaction tends to focus on itsoutcome, whether it is the music or the perfor-mance (see Figure 6). For designers of structure-centered systems, the player’s experience is relevantonly in regard to this final outcome. Most of theearly networks, which stem from the “high-art”music world, fall under this category. Composersand designers of such systems are usually more in-terested in aspects such as artistic vision, composi-tional arrangement, and performance. Players insuch networks, on the other hand, are expected torealize the artistic vision of the composer. It isimportant to note that although most networkscombine process and structure-based elements, cre-ating a successful balance between these aspects isnot a trivial task, as many of the elements are con-tradictory in nature. For example, it would be adifficult task to design a musical network thatwould lead to the creation of an interesting, well-structured composition while players are trying towin a musical game.

Social Organization and Perspectives

Similarly to other social systems, musical net-works are based on social organizations, which canbe informed by “social philosophies.” The mainconceptual axes at play when designing a “social


Figure 5: Motivations for aprocess-centered musicalnetwork.

Figure 6: Motivations forstructure-centered musicalnetwork.

Weinberg

philosophy” for a network are the level of centralcontrol desired and the level of equality provided tothe different participants in the interaction. Central-ized systems, such as in the case of online networksfrom the “Shaper” variety, tend to be governed by acomputerized hub that is responsible for generatingthe musical output based on input from participants.In decentralized systems such as the Musical Fire-flies, players communicate directly with each otherthrough instruments or software applications thatare computationally self-contained. Under the cen-tralized and decentralized umbrellas, we can find awide range of approaches that are based on the lev-els of equality provided to participants in terms oftheir musical role (see Figures 7 and 8). Politicalmetaphors may be appropriate. For example, a“monarchic” system, such as in Golan Levin’sTelesymphony, demonstrates a centralized unequalsocial approach. Here, a leader—a human or insome cases the computer—controls and conductsthe interaction. This leader can provide temporaryfreedom to other players when desired, change andmanipulates the interconnection gates among play-ers, and take control over the interaction in general.

Although providing a composed and structuredinterdependent outcome, the monarchy approachusually fails in providing a true collaborative andcreative voice to players owing to the leader’s domi-nance. Such systems, therefore, would be more ef-fective in addressing structure-centered motivationsthan process-centered ones. Democratic networks,on the other hand (such as in the Squeezables net-work), can be more effective when process is em-phasized. Here, the centralized system provides anequal, or almost equal, role for each player in defin-ing the musical output. In goal-oriented democraticsystems, participants would have to collaborate tocreate a noticeable and significant musical effect.Often in such systems, only the collaborative act ofthe majority defines the final musical result. Differ-

ent participants in democratic networks might havedifferent roles and responsibilities (such as control-ling the harmony, melody, or rhythm), while indi-vidual players might temporarily receive a leadingrole from their peers or from the system.

In decentralized systems, on the other hand, in-teraction occurs directly between participants with-out a central control to govern the experience. Oneextreme example of a decentralized unequal musi-cal network is an “anarchic” network, which pro-vides minimum central control and maximumfreedom for players to generate and manipulatetheir musical material, such as in Tod Machover’sBrain Opera. An interesting hybrid approach is theconcept of Network Computer Music by TheLeague of Automatic Music Composers, which ex-perimented with synchronous decentralized butdemocracy-oriented networks. Another unique de-centralized approach is a rule-based network wherehigh-level musical patterns emerge from the inter-action between a large number of participants whofollow identical simple rules (see Resnick 1999).

Architectures and Topologies

The social organization of the network, an abstract,high-level notion, is addressed by designing and im-plementing the lower-level aspects of the network’stopology and architecture. Here too, the categoriza-tion into centralized and decentralized networks ispivotal. In this classification, centralized networksallow players to interact through instruments andcontrollers that do not have direct influence on eachother. Data from players are sent to a computerizedhub for analysis and algorithmic generation of themusical output (as in the case of most of the onlinenetworks, the Squeezables, Telesymphony, and oth-ers). In decentralized topologies, on the other hand,players interact directly with each other using in-

33

Figure 7. Centralizedsocial perspectives.

Figure 8. Decentralizedsocial perspectives.

struments or applications that have their own com-putational function (such as with Cage’s ImaginaryLandscape 4, the League of Automatic Music Com-posers, the Musical Fireflies, and others).

Centralized and decentralized topologies can beclassified into two additional subcategories: syn-chronous (or real-time) and sequential (or non-real-time). In synchronous networks, players modify andmanipulate the music of their peers while it is beingplayed. In sequential systems, players generate theirmusical material with no outside influence andonly then “submit” it to further transformation and

development by their peers. At its simplest form, acentralized synchronous network can be depicted ashaving a “flower” topology (see Figure 9a). The differ-ent input nodes in the network, which represent theplayers, are constantly connected to a computer hubthat is responsible for creating the interconnectionsamong the nodes. A centralized sequential networkcan be depicted as having a “wheelbarrow” topology(see Figure 9b). Here the inputs nodes, or players,are separated from each other in the time domain,as each new input stage builds upon the last one.

Figures 9c and 9d depict the decentralized versions


Figure 9a. Synchronouscentralized network:“flower” topology.

Figure 9b. Synchronouscentralized network:“wheelbarrow” topology.

Figure 9c. Synchronous de-centralized interaction:“star” topology.

Figure 9d. Sequential de-centralized interaction:“stairs” topology.

Weinberg

of the flower and wheelbarrow topologies and canbe titled as “star” and “stairs” topologies, respec-tively. Synchronous interactions are more likely tobe supported by local-area networks in which la-tency is less of a problem in comparison to remotenetworks. Synchronous networks are also morelikely to support constantly evolving and immer-sive musical experiences, although such an approachcan also lead individual players to lose the sense ofcoherency and causality if the network topology isnot designed carefully to maintain aspects of individ-ual autonomy. In sequential systems, on the otherhand, the interdependent interactions occur in anordered manner by turn-taking procedures. This ap-proach is more tolerant to latency, can be easily sup-ported by remote online networks, and is usuallysimpler to follow for the individual player. How-ever, in sequential networks, the real-time groupexperience might involve compromise, because allplayers are not always involved in music-making.

These generic depictions of synchronous, sequen-tial, centralized, and decentralized interactions donot represent the directionality and the nature ofthe connections among the nodes in the network. Amore detailed depiction can be seen in Figures 10–17.Figure 10 depicts simple one-way flower architec-ture. Here, data is sent synchronously from the play-ers to the hub, which is responsible for generating

the musical output based on algorithmic treatmentof musical and gestural input. The interdependentaspects in this simple network are derived onlyfrom the players, who listen to the musical outputfrom the hub and change their actions accordingly,as can be seen in the Pazelian application (Pazel, etal. 2000), for example. A higher level of interdepend-ency is depicted in Figure 11, which shows a decen-tralized star topology where players are connected

35

Figure 10. Symmetric, one-way “flower.”

Figure 11. Symmetric, in-terdependent “star.”

Figure 12. Asymmetric,weighted interdependent“flower.”

directly to each other and can interdependently ma-nipulate each other’s musical output such as in thecase of the FMOL project. Cage’s Imaginary Land-scape No. 4, on the other hand, can be seen as a syn-chrotrons anarchic decentralized network with asymmetric interconnection scheme.

Both Figures 10 and 11 depict symmetric topolo-gies, where all nodes are connected to the hub or toeach other. Such architectures would be best suitedto support an “equal” social approach. Figure 12, onthe other hand, presents an asymmetric (unequal)network architecture where connections are pos-sible only in specific directions and in between spe-cific nodes. Figure 12 also introduces the concept ofweighted gates that control the level of influence ateach intersection in the network. Normally, gateswould be open, providing a full level of influence,whether for algorithmic control or musical content.The gates, however, can also be partly (or fully)shut, allowing only a limited level of functionalityat each particular intersection. Gates can have dif-ferent weights as default values (depicted as num-bers next to some intersections in Figure 12) but canalso be changed and manipulated in real time basedon a player’s input, such as in the Squeezables net-work. This asymmetric weighted flower topology iscommon in democratic networks, as it provides dif-ferent roles and levels of importance to the differentplayers. An extreme version of this approach canlead to a monarchic network where one player cancontrol all the weighted gates in the system andtherefore gain full power in conducting the interac-tion, such as in Golan Levin’s Telesymphony.

Sequential networks have similar kinds of inter-connected topologies. In the simplest architecture,each node is only connected to the next one so thatevery player can manipulate the musical product ofthe previous player, such as in the case of the Musi-

cal Fireflies. Figure 13 depicts such a symmetric,one-level stairs topology. The arrows between stepsare bidirectional; the outgoing arrows represent themusical output that is sent to the next node, andthe incoming arrows represent the algorithmic ma-nipulation that is applied to it. More elaborate se-quential topologies allow players to transform notonly the previous player’s musical output but alsothe musical product of the other participants in dif-ferent nodes of the network and at different times.These transformations can be applied directly amongplayers or through a central hub. Such a symmetricmulti-level wheelbarrow topology is depicted inFigure 14. Similarly to synchronous networks,asymmetric sequential networks can facilitate inter-dependent interaction only in specific directions andprovide idiosyncratic roles to the different players.Here too, weighted gates can be assigned to the in-tersections (see Figure 15).

In practice, most elaborate musical networks com-bine synchronous and sequential elements in differ-ent balances. In such hybrid networks (such as theBeatbug network), part of the interaction is sequen-tial (as players have autonomous control over theirown music before sharing it with the group), whileother parts of the interaction are synchronous, whenplayers can influence their peers’s musical output inreal-time. In addition, certain sections of the inter-action can be centralized, and others can be decen-tralized. In general, these hybrid systems can be


Figure 13. Symmetric, one-level “stairs.”

Figure 14. Symmetric,multi-level “wheel-barrow.”

Figure 15. Asymmetric,multi-level, weighted“wheelbarrow.”

Weinberg

depicted in two manners: “Stairs of Flowers,” wheresynchronous interactions are ordered sequentially intime (see Figure 16), and “Flowers of Stairs,” wherea set of sequential interactions are synchronouslyconnected to the system’s hub (see Figure 17). Aweight system can also be assigned and controlledin real time to provide dynamic levels of influence.

Musical Content and Control

Making decisions about the motivations, socialperspectives, and the network architectures are es-sential steps towards setting the framework for aneffective musical network project. But of great im-portance for the final musical result are the lower-level decisions regarding the actual musicalparameters and transformation algorithms thatwould infuse such a framework with musical con-tent. Of unique importance are decisions regardingthe musical parameters and functions that would begenerated and controlled autonomously as opposedto those that would be controlled interdependently.This aspect of the design bears a subjective aestheticcore, as different composers and designers wouldhave different ideas, tastes, or artistic interests whendetermining the precise control parameters. Themusical content and transformation decisions areinformed by the higher-level design decisions. For

example, an exploratory, anarchic flower topologywould be best served by allowing players to gener-ate and manipulate a full gamut of pitches, timbres,timing, and expression aspects and by applying con-stantly changing random transformation algorithms.On the other extreme, a system that would limit thepossibilities for input and manipulation by allowingplayers to choose from a limited bank of presets inan effort to achieve a specific task is likely to sup-port a goal-oriented sequential structured topology.

Generally speaking, every musical parameter, suchas pitch, rhythm, timbre, or dynamics, is a candi-date for autonomous as well as interdependent con-trol. However, there are some rules of thumb thathave been proven to be effective for the creation of acoherent yet collaborative and immersive interac-tion. For example, allowing players to influence andcontrol parameters such as pitch or contour of apeer’s melody may lead to an incoherent experiencefor the peer who may loose control over some of thefundamental aspects of the melodic production.Such mappings may draw the systems into an anar-chic experience that is difficult to follow for partici-pants and viewers, even if the architecture supportsother social perspectives. On the other hand, grant-ing a player full autonomous control over his or herpitch content while allowing other players to con-trol ornamental and expressive aspects such as tim-bre or dynamics tend to lead to more coherent

37

Figure 16. Hybrid, generic“Stairs of Flowers.”

Figure 17. Hybrid, inter-connected “Flower ofStairs” with weightedgates.

experience for the melody player. Here, the melodyplayer can interdependently enrich his or her ownplaying experience while attempting to accommo-date the melody to the new timbres and dynamics.

Another important aspect for maintaining the in-teraction coherency is preserving the nature of themusical material as it was originally created. In se-quential networks in particular, it is easy to allowco-players to modify their peers’s music beyondrecognition. This might disconnect the originalplayers from the music they created, obscuring theirdetailed idiosyncratic contribution to the group.The interdependent transformation algorithms,therefore, can be more effective if they attempt tomodify surface elements and to maintain reversibil-ity, so that the original musical output would beperceivable and retrievable.

A Final Note

The field of interconnected musical networks hasmatured and is currently expanding to new audi-ences and new platforms such as wireless commu-nication devices (see, for example, McAllister et al.2004; Tanaka et al. 2004). If the field is to continueto grow, then composers, performers, and audienceswill require a solid theoretical framework of refer-ence when composing, designing, participating in,or listening to interconnected musical networks. Inthis article, I attempted to define the fundamentalaspects for such a theoretical framework by map-ping the field, addressing aesthetics and technicalconcepts and motivations, and providing design sug-gestions based on an investigation and analysis ofhistoric and current work. I plan to modify and ex-pand this theoretical framework in correlation tothe constantly evolving changes in technology andartistic focus in the field.

Acknowledgments

I would like to thank the Hyperinstrument Groupat MIT Media Lab and Tod Machover for his supportand advice.

References

Barbosa, A. 2003. “Displaced Soundscapes: A Survey ofNetwork Systems for Music and Sonic Art Creation.”Leonardo Music Journal 13:53–59.

Benzon, W. L. 2001. Beethoven’s Anvil: Music in Mindand Culture. New York: Basic Books.

Bischoff, J., R. Gold, and J. Horton. 1978. “MicrocomputerNetwork Music.” Computer Music Journal 2(3):24–29.

Bongers, B. 1998. “An Interview with Sensorband.” Com-puter Music Journal 22(1):13–24.

Brown, C. 1999. “Talking Drum: A Local Area NetworkMusic Installation.” Leonardo Music Journal 9:23–28.

Burk, P. L. 2000. “Jammin’ on the Web—A New Client/Server Architecture for Multi-User Performance.” Pro-ceedings of the 2000 International Computer MusicConference. San Francisco, California: InternationalComputer Music Association, pp. 117–120.

Cage, J. 1961. Silence. Middletown, Connecticut: Wes-leyan University Press.

Duckworth, W. 1999. “Making Music on the Web.”Leonardo Music Journal 9:13–18.

Feldmeier, M., M. Malinowski, and J. Paradiso. 2002.“Large Group Musical Interaction Using DisposableWireless Motion Sensors.” Proceedings of the 2002 In-ternational Computer Music Conference. San Fran-cisco, California: International Computer MusicAssociation, pp. 83–87.

Freeman, J., et al. 2004. “Adaptive High-level Classifica-tion of Vocal Gestures Within a Networked Sound In-strument.” Proceedings of the 2004 InternationalComputer Music Conference. San Francisco, Califor-nia: International Computer Music Association,pp. 668–671.

Gang, D., et al. 1997. “TRANSmidi: A System For midiSessions Over the Network Using Transis.” Proceed-ings of the 1997 International Computer Music Confer-ence. San Francisco, California: InternationalComputer Music Association, pp. 283–286.

Gresham-Lancaster, S. 1998. “The Aesthetics and Historyof the Hub: The Effects of Changing Technology onNetwork Computer Music.” Leonardo Music Journal8:39–44.

Helmuth, M. 2000. “Sound Exchange and Performance onInternet2.” Proceedings of the 2000 International Com-puter Music Conference. San Francisco, California: Inter-national Computer Music Association, pp. 121–124.

Hinton, M., and D. Morgenstern. 1988. Bass Line: TheStories and Photographs of Milt Hinton. Philadelphia,Pennsylvania: Temple University Press.


Weinberg

Iwai, T. 1998. Composition on the Table. Exhibition atMillennium Dome 2000, London.

Jordà S. 2002. “FMOL: Toward User-Friendly, Sophis-ticated New Musical Instruments.” Computer MusicJournal 26(1):40–50.

Jordà, S. 2003. “Sonigraphical Instruments: From FMOLto the reacTable.” Proceedings of the 2003 Conferenceon New Interfaces for Musical Expression, Montreal,pp. 121–124.

Konstantas, D., et al. 1997. “Distributed Musical Re-hearsal.” Proceedings of the 1997 International Com-puter Music Conference. San Francisco, California:International Computer Music Association, pp. 279–282.

Levin, G. (2001) Dialtone: A Telesymphony. Premier atArs Electronica Linz Austria. Available at www.flong.com/telesymphony/.

McAllister, G., M. Alcorn, and P. Strain. 2004. “Interac-tive Performance With Wireless PDA.” Proceedings ofthe 2004 International Computer Music Conference.San Francisco, California: International ComputerMusic Association, pp. 702–705.

Pazel, D., et al. 2000. “A Distributed Interactive MusicApplication Using Harmonic Constraint.” Proceedingsof the 2000 International Computer Music Conference.Berlin: International Computer Music Association,pp. 113–116.

Rasch, R. A. 1988. “Timing and Synchronization in En-semble Performance.” In J. Sloboda, ed. Generative Pro-cesses in Music. Oxford: Clarendon Press, pp. 70–90.

Resnick, M. 1999. Turtles, Termites and Traffic Jams: Ex-plorations in Massively Parallel Microworlds. Cam-bridge, Massachusetts: MIT Press.

Rosenboom, D., ed. 1976. Biofeedback and the Arts:Results of Early Experiments. Vancouver: AestheticsResearch Centre of Canada.

Tanaka, A., et al. 2004. “Enhancing Musical ExperienceThrough Proximal Interaction.” Proceedings of the

2004 International Computer Music Conference. SanFrancisco: International Computer Music Association,pp. 205–208.

Weinberg, G. 2002. “The Aesthetics, History, and FutureChallenges of Interconnected Music Networks.” Pro-ceedings of the 2002 International Computer MusicConference. San Francisco: International ComputerMusic Association, pp. 349–358.

Weinberg, G. 2004. “Voice Networks: Exploring the Hu-man Voice as a Creative Medium for Musical Collabora-tion.” Proceedings of the 2004 International ComputerMusic Conference. San Francisco, California: Interna-tional Computer Music Association, pp. 623–626.

Weinberg, G., R. Aimi, and K. Jennings. 2002. “The Beat-bug Network: A Rhythmic System for InterdependentGroup Collaboration.” Proceedings of the 2002 Confer-ence on New Interfaces for Musical Expression.Dublin, pp. 107–111.

Weinberg, G., and S. Gan. 2001. “The Squeezables: Towardan Expressive and Interdependent Multi-player MusicalInstrument.” Computer Music Journal 25(2):37–45.

Weinberg, G., T. Lackner, and J. Jay. 2000. “The MusicalFireflies: Learning About Mathematical Patterns inMusic Through Expression and Play.” Proceedings ofthe 2000 Colloquium on Musical Informatics.L’Aquila, Italy: Instituto GRAMMA, pp. 146–149.

Whalley, I. 2004. “Adding Machine Cognition to Web-Based Interactive Composition.” Proceedings of the2004 International Computer Music Conference. SanFrancisco, California: International Computer MusicAssociation, pp. 196–200.

Yamagishi, S. 1998. “Variations for WWW.” Proceedingsof the 1998 International Computer Music Conference.San Francisco: International Computer Music Associa-tion, pp. 510–513.

Yu, J. 1996. “Computer Generated Music Composition.”Master’s Thesis, MIT.

39

Documents

Gil Weinberg Interconnected Musical Networks: Toward a