Page  00000001 Recent work around Modalys and Modal Synthesis Francisco lovino Ren6 Caus s Richard Dudas Institut d& Recheiche et Coordnaion Acoustiqie Musique 1, place Stravinsky, 75004 Paris,, ddas @ircafmfr Abstract The originality and flexibility of the physical modlling synthesizer Mochlys is being continously validted by an ever-growing community of users mad up of composeis, acousticians andmusicologists alike In this paper we will present the main lines tha have orientedleent ieseaich aroundModtlys. Thesecan be summaized (1) Exploration and tentative dassification of the musical possibilities of Mochlys. (2) Use of modcl synthesis as a meeting point between signal synthesis and physical modIlling synthesis. (3) Poit of Modhlys to a real-time platform 1 Background 1.1 How Modalys works Recent work around Modhlys has been developed on the hypothesis that sound synthesis d&sign can be oonsidled a fundamental deement of musical omposition [1]. The operation principle of Mochlys is physical modelling by mochl synthesis [2] which consists of solving the vibratoy equations of the involved physical structies on a mochl oordnate basis. A mode of vibration is an eigenvalue (flepency and loss) and an eigenvector (modcshape) of the characteristic equation of a physical system. The user inteface of Mochlys is a set of primitives extendng the Scheme language[3]. The principal chta structues handled by the progran are objects (physical stuctures), connections (inteactions between objects), and controlleis (time-varying parameters). Two oonsecutive phases are necessary to synthesizea soundwith Mochlys: (a) Instrument construction - instantiating objects and assembling them together via onnections, (b) Instmument execution - sending ontroller information to onnections to make the instlument vibrate. Mohdlys offeis a way to operate on sound which is substantially dffeent from signal sound synthesis methods. The user simultaneously plays the role of luthier, oomposer and intepreter. It can be argued that these roles exist implicitly in signal synthesis methods, but the fact tha in Mochlys each role must be deadly d&find, implies a new relationship between the ommposer andthe soundmaterial. Fulthtmmoe, the complexity that each of the two synthesis phases can have, with the many dffeent levels of abstraction tha manifest themselves a the time of sound dsign, can lead to a situation in which the richness of possibilities may confuse and overwhelm instead exciting the imagination A tentative dfinition of a franework which would contain a set of suggestions for musical applications of Mochlys, with each suggestion being described with its motivations potential problems, would be more than useful, indedindspensable Before proceeding, it is impoitant to remember the particularities of Mochlys. 1.2 The issue of synthesis control Although the first exanples of sound synthesis by physical modelling weie iealized 25 years ago - almost simultaneously with the fist experiments of dgital FM synthesis!- and some of the equations bdscribing vibratory movement of musical instruments chte from many cbacbs, we ascetain that tochy, physical modelling synthesis is far from being exploted to its full potential and that a substantial refeience lepeitory of compositions using thesetechniques cbes not yet exist. The lack of use of physical modelling is explainednot only by the heavy algorithms that implement the vibratoy equations and the practical dfficulty to ontrol them, but also by the fact that, even if progiess in haidware technologies has brought high-qiality ommmcial physical modclling synthesizes, market reasons constrain them to be more oriented toward eprochcing existing acoustical instruments rather than providng open enviuonments whereimaginative instruments can be conceived Neveithdess, when compaed with signal synthesis techniques, physical modelling offeis two precious advantages tha cannot be

Page  00000002 undrestimaed: (1) causality, or the possibility tha human perception associates the synthesized sound with some kind of vibrating staucure, and (2) expressivity, or the possibility tha the control mechanism of the synthesis technique is diecly related to the gestural information contained in the sound In addtion, modal synthesis offers the paiticular advantage of mocdlarity as the modal representation of physical strucures is uniform (a modeshape matrix, a vector of modal fr~quendes, and a vector of modal losses), the user can indfferently assemble sub-structures and connections to build any imagined instriment. Yet the posibility of buildng an arbitrary instrument raises inmedately the problem of controlling the synthesis parameters: even if the instrument's constuction phase may seem stlaightfolward and intuitive, of the execution phase(findng good values to sendto the input parameters of the connections) can oftenbe complex and laborious. As formal training in sound synthesis has always essentially consisted of the study of signal synthesis techniques, in which spectral operations have a cbminant staus, composeis may at first glanoe find physical modelling synthesis as a rupture with tradition. Foitunately, modal representation of a physical structure is diecily related to the spectral content of the sound produced [4,1]; the user has to benefit the most from this advantageous situation. A principal focus of our ieseach has been the use of Modhlys as a place where signal and physical modlling synthesis techniques convege 2 Composing with Modalys: experiences and suggestions Now that we undeistandthe particularities of the synthesizer, we can concretize ideas and propose a franework for musical applications of Modblys to the user. We present five possibilities for musical applications of Modhlys; note that these are non-exclusive ad may therefore be usedin oonjunction with one another. (i) Variations on an object. Transfobming some physical parameter of an object or changing the way to interact with it may create a family of soundb sharing properties. A beautifiul example is the rectangular plate, where slight modfications of its length, widh or thickness may result in timbral charges regardng the degree of inhanmonicity. Changing the type of interaction will result in even licher genres of sound (ii) Viitual lutheie We dstinguish two dffeent approahes to viltual instrument onstuction: (iia) 'Real'". This approach encompasses instruments which can be built in Modilys and a the same time, physically plausible. One could considr this as "computer assisted lutheie". A concete exanple in recent work with Modilys is a simulation of the hannonisation of a cicular membrame with a dsk of masses, to achieve an indan tabla [5]. Charging the mass dstfibution furnction aroundthe drsk will leadto dffeent kinds of tabla (iib) "Unreal". This approach encompasses instruments which can be imagined but cannot be physically buildble. Suppose, for instance, a string piercing thlough a membrame t a contact point which varies in time [6], or, more interesting, a ',ecursive bowedstfing" in which the incoming pressure is function of the cuirent string vibration. Imaginative users following this diection can fall in the temptation of implementing exotic topological and physical configurations, which are not really reommended Playing these instmuments can be a very dfficult task, and the user could easily foiget tha the purpose of using Mocdlys in composition should be to stimulate creativity, not to find solutions to artificial problems. It is impoitant to have in mind that a simple string with a bow is already a lich and expressive instrument which takes time and effoit to master. (iii) Simulation of instrumental gesture. The fobmer dscussion and the bowed string exanple serve to introcace another type of utilisation which can be fruitful and efficient only when used with a dear compositional puipose. It is very tempting to try to ieprocbce exactly a Stradvarius violin with a Pelman-like playing. If we compae the time spent by generlions of luthiers to achieve good instriment qiality and the years of professional training for perfoumers, with the number of weeks tha the realization of the delectonic part of a piece takes, we can conclude tha this approach can be as metaphorical as oneous. After all, reprodiction of instruments andplayes is an invaluabletool to get insight into the structure of soundm ad musical gesture. But, when compositional issues are concened, care must be taken not to spendtoo much time on simultion. (iv) Complement to natural instruments. As we cb not want to compete with perfoimers, a logical way to use Mocilys is thus to extend acoustical instauments with souncb which could be dfficult or impossible to produce otherwise. The success of the result dpendc on how dosethe acoustical and the virtual instrument are, the ideal being tha our sense of perception cbes not feel ay ruptiure between the synthetic and the acoustic sound. As a exanple, we have obtained satisfactory results with multiphonics, as even skilled peifomaers have dfficulty in producing this Idnd of sound, and the Modlys simple-ieedinteraction mcoupled with tubes can produce a lich range of multiphonics.

Page  00000003 (v) Convergence of signal and physical mocdlling synthesis techniques. As we explain in the next section, new categories of sound can be created by mixing the causality and expresivity of physical modls with the precise spectral ontrol of signal synthesis. From thesefive possibilities we have experienced tha Modilys can be a powerful tool for creating sound specially when the viitual instruments fill three conditions: originality of the timble, flexibility of the control, and life-like qcality of the sound We have dscoveiedthat 3-structure instruments, such as bow-string-membrane or reed-tube-plae, work particularly well because the qcalities of the two iesonaing structures can finely melt to produe an ambiguous timbie that shares the perceptual properties of both strucures, but tha has, a the same time, an indviduality of its own The same observation can be mad& about physical strutures whose modes have been altered with the dcta of another stucture. Flexible instumental control is indspensable to allow the user to dvelop a '"eeling" for the instrument; this is one of the motivations for a real-time implementation of Modlys. I is woithy to mention a repeated experience we have had with professional musicians: when they hear an interesting Modblys soundwithout knowing that the sound is synthetic, the ieaction is always to associate it with some "contemporay" perfoim technique on this or tha acoustic instrument, without initially thinking that the source of the sound is a completely viitual instrument. This kind of experience would be a good test to validte the &dgree of musical interest tha the Modblys instrument may have.These conclusions arise mainly from the atempt to use Modblys in musical production Today we count a least four pieces using Modblys as the principal synthesizer [7]. 3 "Signalic" physics and "physical" signals Recents extensions of Modilys make static or dynamic changes of frequency, loss andmodishape scaling possible As it is very dfficult to predict how modfictions of a modbl parameter will be reflected by changes in the physical data of the struture and vice-veisa, it might seem contradictory to instantiate a structure from its physical properties for later changing its modil propeties. However, practical mompositional reasons exist to justify our approach [1]. Smooth changes often keep the mcmplexity ofontrol within tolerable limits (bow pressure and velocity may needto be adapted penmanently to a string whose mocdl data rapidy varies). Another synthesis control problem is raised when d&aling with the "modeshape anplitu'", which may iefereither to the global anplitudb, or the anplitud& a the contact or excitaion points. This is the motivaion for the introdiction of the "single-point" object in Modlys. 'S ingle-point" objects are physical strutures dscretized with one point dscribed solely by its modil characteristics. Thus, for one point, the "modeshape anplitud" coiresponcbd to the anplitud& of the spectral cmponent dfined by the mo&d freuency. For the case of one mo&d, the associated physical structure is the dassical mass-spring system, and for this simple case, the mochl properties couldbe ieversed to find the physical properties. For the case of several mocds, we couldsee the single-point as a "modal compilation" of some physical system whose physical propeties' information may be impossible to extract from the mocdl dcta. The fundamental dffeence between signal methocbds and a physical modelling approach to sound is tha the latter takes interactions andspaial properties into acomunt. From the signal point of view, sound is a mere function of time (or a set of functions if several channels are takeninto acomunt) representing a measuiedairpressure, whereas for physical mocdlling, sound represented by the instantaneous vibration of one or several points of a physical structure, depends on a least three variables: time, the excitaion foice (a function of time), and the strucure's impednýe (which can also be a function of time). The nceptual complication of physical modclling, comes from the fact tha if the source of excitaion is an interaction, the injeced foice will be function of the instantaneous vibration. The inherent causal propeties of physical modls are a consequenice of this feedback relationship between the excitaion and the vibration. The musical interest of single-points comes from the fact tha a time signal couldbe represented with spectral information and latter "converted' into a Mochlys object via a single-point. Thus, we can indferently see signals as physical points vibrating in space and also, vibrating objects as signals with spectral content. Conceptually, this couldbe seen as mtertra/zdk&g a sound with the relation of causality (suppose for instance a hammer striking a vowel, or a bowedtrumpet) New dasses of souncds tha share the properties of both signal modls (precise spectral content) andphysical modls (causality, expresivity) are thus obtained The price to pay for this benefit is a higher mcomplexity of cntrol for the instrument. 4 Other axes of interest

Page  00000004 A subset of Mochlys has ben poited to the FTS leal-time environment [8,9] and uns on a Sillicon Graphics with an R4000 processor. Real-time oontirl is impoitant in Mochlys to prototype soundM in a simple way and to make died and live instrumental oontiol possible Also, 'Modalysef, a graphical interface suitable for non-piogrammers has been written in Macintosh Common Lisp by Richaid Polfieman [10]. We are cuirently testing the musical validty of thesetools. [10] Polfieman, R. [/so-A/efaah Dsia nfor Soffwae Basead SoudaSy/es Systes. Ph. D. Thesis, Univesity of Heitforcbhire, United Kingdm, 1997. References [1] Eckel, G., ovino, F., Causse, R 'Sound Synthesis by Physical Modlling with Mochlys", in Proceedings of the International Symposium on Musical Acoustics, Dourdbn, France, pp. 478-482, 1995. [2] Adrien, J.M. Eazied St/uatues Complere VYratei, App/liiw o d /aSy irhseapr Afotdles P/hysioes. Ph. D. Thesis, Universitedc Paris VI, Paris, 1988. [3] Morrison, J., Adrien, J.M., 'MOSAIC: A Framewoik for Mochl Synthesis ", Computer Music Journal, Vol. 17, No 1, pp 4546, 1993. [4] Adrien, J.M. 'The Missing Link: Mocil Synthesis ", in G. De Poll, A Picalli, and C. Roads, eds., Represewt/ains of MAfuial Sgzds. MIT Press, Cambridge, Massahussets, 1991. [5] Castdain, L 'Tiis exemples d& constructions dinstrument'. Inteinal Repoit, IRCAM, 19%. [6] Bonnet, M.D. Codes Frott&s etiAfomwciYqe Ph. D. Thesis, Ecole d& Hautes Etudes en Sciences Sociales, Paris, 1997. [7] Duchs, R. 'Sonata", for viola and decttonics, to be premiered in 1998. Naon, L Piece in progiess for accoldbon and delectnics, to be premiered in 1998. Stubbe, HP. 'Masks", for violin and delectnics, premieredin Copenhagen, May 1997. Watkins, R. 'The Juniper Tree", Operafor 5 singers, chamber orchestra and decttonics, premieredin Munich, April 1997. [8] D6chdle F., De Ceoco M., Puckette M., Zicaidli D., 'The Ircam Real-Time Platfrm: Evolution and Peispectives", proceedings of the 1CMC, Aaihus (Danemmak), 1994 [9] Iovino, F., Schnell, N. 'Pieliminaiy notes for the Mochlys-FTS implementation". Intemal Repoit, IRCAM, 19%.