Page  270 ï~~SoundModels: A language for description of sound and sonic structures based on a taxonomy. Peter Lunden Royal Institute of Technology (KTH) Dept. of Speech Communication and Music Acoustics Box 700 14 S-100 44 Stockholm Sweden email: ludde@kacor.kth.se Abstract. A language for description of sound and sonic structures is described. A typology of sounds is the foundation of this language. The typology is designed by the author based on ideas from Schaeffer and Smalley. The language is implemented in a prototype based multiple inheritance object oriented system which facilitates the partitioning of different notions of sounds into there own distinct inheritance paths. Constraints are applied because they are an elegant solution to the problem of describing dependencies between parameters and the dynamic behaviour of sonic structures. The language does not depend on any particular synthesis method and can be interfaced to many different methods as long as the parameters of the method are closely related to the acoustical result. The language can be used to describe a large universe of sonic-structures with complex dynamic behaviour and complicated interdependencies which makes it useful in a computer music environment. 1. Introduction The purpose of the SoundModel language is to describe sound and sonic structure in an environment for computer assisted composition of electroacoustic music (EAM). The language should cover as much as possible of the audible sound space and not be limited by any stylistic prejudges to for example "musical" sounds. It is essential that the description is in a form that facilitates the synthesis of the describe sound. Composers are generally more interested in the perception of sound rather than there physical constitution. The language must therefor be focused on the perceptual perspective. It should be based on psychoacoustic terminology and not on obscure parameters enforced by the synthesiser technology or by the synthesis method. The language should be designed to facilitate reusability. New models should be defined by recombination of existing models rather than implemented from scratch. 2. Typology The main problem of designing a language to describe sound and sonic structures are the lack of a useful taxonomy of sound. The strategy used in this work is to divide and conquer the sound space, which means that a set of terms has to be defined to divide the space into a set of subspaces. The question is how should the sound space be divided, how many and which di mensions are needed to describe sound from a perceptual perspective? Pitch and loudness are well known features and they are not further discussed. The main problem are the description of timbre. The unfortunate psychoacoustic definition of the term timbre is not very helpful as it is to vague. There has been some research in this area for example [Bismarck 74] and [Bruijn 781. Unfortunly many of the works are limited to static sounds, an approach which limits the usefulness of this results as the temporal features play an important role in the perception of sounds. The most relevant investigations are [Miller & Carterette 75] and [Grey & Gordon 781. The two works points out the spectral envelop as the most salient feature, other important features are the temporal behaviour of the partials and the overall envelop of the sound. A question that remain unsolved is how the focusing of features works. It is well known that the importance of the perceptual features depends on the type of stimuli and on the context. Another limitation of the knowledge in this area is the selection of stimuli used in the investigations. They are very often restricted to "musical" sounds, which far from exhausts the audible sound space. This is a problem because there is not such thing as "musical" sounds for a EAM composer, every sound is potentially useful in a EAM piece. This means that the major part of the sound space is still unexplored and many important features of the perception of sound are left to Music Representation, Data Structures 270 ICMC Proceedings 1994

Page  271 ï~~be revealed. The only works that tries to cope with the whole sound space are the works by Schaeffer [Schaeffer 66] and Smalley [Smalley 86]. SoundModels are greatly influenced by there works. 3. The Structure of SoundModels The sound space is subdivided into a set of domains which can be further divided into subdomains and so on. The domains are the loudness-, pitch-, spectral-, temporal- and spatial domains. Each domain or subdomain covers a set of related features where each feature is described by a set of parameters. A more detailed description is found in [Lunden 93]. 3.1. Spectral domain The spectral domain can conceptually be thought of as a source and filter model, where the spectral profile describes the filter and the discrete or continuous spectrum describes the source. The "Spectral Profile" describes the frequency dependent amplitude of the sound. There are strong evidence that spectral envelop, besides pitch and loudness, is the most salient perceptual feature of sound [Grey & Gordon 78] [Miller and Carterette 75]. The Spectral subdomain describes the overtone structure (or lack of). It is further divided. into Discrete Spectrum which describes sounds with a clear overtone structure., including sounds with distinct pitch and also inharmonic sounds. The other subdivision are Continuous spectrum handles noise sound and sound without overtone structure. 3.2. Temporal domain This domain describes the overall energy distribution in time It is not concerned with the inner details of a sound only the general shape in time. This is very close to Schaeffer definition of the energy profile [Schaeffer 661. The domain divides the sound space in impulsive, iterative, and persistent subdomains and the iterative is further divided into regular and irregular. 3.3. Spatial domain The spatial domain includes features like room acoustics, directional and distance cues. This domain is not implemented in the current version. 4. Temporal behaviour and parameter interdependencies It must be possible to organise time in a hierarchical manner as sonic structures can have a complex structure of time in several layers. Consider the following problem. We have designed a sound-model that can iterate other models. Now we have at least two different layers of time, the local time in the iterated models called A and the local time in the iterating model called B. Imagine that we would like the pitch of A to raise one octave during the duration of B. Time dependent function to handle this can not refer to the local time of A since the pitch will raise an octave for each single iteration and not over whole duration of B as we would like. To accomplish this must the function refer to the local time of the B. This can be achieved with the constraint mechanism of the language, we can constraint the time reference of the pitch envelop to refer to the local time of B. To further demonstrate the power of this mechanism we can do the same thing in another way. We can add a new parameter to B and constraint the pitch of A to depend on the new parameter and we can then attach our pitch envelop to this parameter referring to the local time of B (sorry there is no room in this paper to show any examples). References [Bismarck 74]. Bismarck, G. von. "Sharpness as an attribute of the timbre of steady sounds." Acustica 30 p. 159 -172. 1974. [Bruijn 78]. Bruijn, A. de. "Timbre-classification of complex tones." Acustica 40 p. 108-114. 1978. [Grey & Gordon 78]. Grey, J. M. and Gordon, J. "Perceptual Effects of Spectral Modifications on Musical Timbres." Journal of the Acoustical Society of America 63(5). 1978. [Lunden 93]. Lunden, P. "Knowledge Representation of Sounds and SonicStructures Based on Constraints and Multiple Inheritance." Proceedings of ICMC, Tokyo 1993. [Miller & Carterette 75]. Miller, J. R. and Carterette, E. C. "Perceptual Space for Musical Structures." Journal of the Acoustical Society of America 58(3). 1975 [Schaeffer 66]. Schaeffer, P. "Trait6 des Object Musicaux." Seuil, Paris 1966. [Smalley 19861. Smalley, D. "Spectral Morphology and Structuring Processes." in The Language of Electroacoustic Music, Emmerson, S. (ed). MacMillan Press 1986. ICMC Proceedings 1994 271 Music Representation, Data Structures