Page  550 ï~~Musical Aanalysis of t Msic with Sczagrum Martha Brech Technische Uiversitat Berlin H 51 Str. des 17. Juni 135 D - 1000 Berlin 12 Abstract: Campter generated sonagrams may serve as a base for analysis for all kinds of music, especially cmpter music, which cannot be notated otherwise. Considerations about the handling of sonagrams are discussed in this paper as well as an analysis system for sonagram notation and first results are presented In this paper some ideas will be presented which arose during a musicological project about experimental canputer and electronic music. It deals with understanding and perception of this kind of music, asks for the leading parameters of it and the necessary musical analysis. For musicologists analysis of experimental electronic and computer music is a problem for musicologists due to the lack of formal notation. Usually, some sort of listening transcription are made to suit the interest only of the individual researcher. Thorensen (1985) and &nalley (1986) gave listening and analysing systems with fixed definitions. These works help a lot, but the matter is still very complicated and contains the problem of all manual transcription: the lack of objectivity. Sa r as Notation Sonagrams may be helpful here. They show: - on x axis: the passing time - on y axis: frequencies, the amplitude of each frequency is shown by the printing density That means that sounds are to be seen as visual patterns which even can be seperated from each other in case they overlap. So they can be used as a kind of (postscriptive) notation. The idea of using sonagrams as notation is as old as the analogue sonograph. R. Cogan even gave an analysing system based on photographed sonagrams at the ozsilloscope, but only ccmputer generated sonagrams are able to offer sufficient solution with printouts. At the Electronic Studio at the Technical University, Berlin, an FFT-based sonagram program, inspired by Tams i gvary and his project ideas (see IOKC 1990), was written on Microvax by Holger Becker. It is able to analyse frequencies up to 16 khz and to print them linearly or logarithmicly (at any logarithm base). There are 16 dynamic steps in a variable dynamic range up to 120 dB. The representation of time can vary up to a meter or more per second. (The actual length is due to the variation of fIT-Size and overlap.) It turned out that the most effective printout of a usual sonagram had the following parameter settings: spectrum: in octaves, dynamic range: 60 dB (=3.75 dB per step), time: 13 an per seond. Raing anxi thierstandingSoag Eye and ear are different perception organs. Although reading of sonagrams usually is easy, it must be learnt. The printouts show all frequencies in a given sound equally, the fundamental tone as well as the overtones, so that one souind may be seen as a pattern with multiple ranks. This will not only be aproblem, if many different sounds are involved -in that case the printout will turn into a nonseperable grey (e.g. Bohor II by Xenakis looks like this) - it also means that all partial tones are printed with their absolute amplitude. Therefore ICMC 550

Page  551 ï~~a sound with relativly fewer overtones is printed darker than a sound. with a lot of overtones even if the ear percieve them as equally loud. 4 Additionall in hearing perception frequencies are filtered. Some:.:_,. frequencies may occur louder to the eye (because they are printed: =F- -,..- -. darker) than they are heard. There may be even cases where a pattern appears to be overwhelming by looking at, but is not heared at all. (The s am shows such a sound, a second sound overlap, but the N glissando upwards cannot be heard, it is masked by lower frequencies.) _: Filtering the sonagram through the listening perception curve is no real solution because soundpatterns, while overlapping, are sometimes easier to be identified in upper frequencies. Since a sonagram is a quantitative notation, listening to the music while reading and N analysing seems to be the best approach.!Lsi1 Analysis with Sngu As postcriptive notation, sonagrams print out the sounding result of a piece of music. For computer music this implies that the sonagrams will be able to seperate a lot of different sounds but tell nothing about their creation. On first sight, this might be difficult because the main emphasis in cxnputer music lies in the creation of new sounds. In fact, in most books about canputer music there is not much more to learn. But putting together these sounds makes the music. The sound itself is only part of a piece. Therefore, from a musicological point of view, the question of the relationship between sound and its construction is of interest too. In a carmuication system between canposer and audience s are placed on the side of the audience. Hence, the following questiond arise: What is reaching the audience, what can be perceived. Later it can be asked, if some of the sonagrams' findings correlate with the anposer's intention. T7e Phrceiver's View Perceiving is an individual process which is based both on experience and human capacity in information processing. Fcperience here means neither cultural nor personal experience of life or other aspects of apperception, it just implies that in this field of music there are either expert or untrained listeners. Therefore, the evidence of such an individual analyzing result is questionaible. Speed of learning must be taken into account. Stoffer (1985) e.g. investigated the perception of structure (in tonal music) by a click test. He cpared two groups, one with high musical ccqetence and trainin and another with low musical canpetence. Much to his own surprise the second group seemed to have learned during the test: In contrast to the first session's results Stoffer dicovered only little difference between the two groups after four sessions. As analyzing involves frequent listening to a piece, learning can even take place there and hence will be less individual as assumed. Thus, the perception of the senses - ear and brain, the simple biological facts, which can be explored experimentally, are of interest. In most cases, investigations are made with tonal music. However, some are valid for the perception of capter music too, like the fusion of sinus tones to musical tones or duration and rembein principles etc. Fraisse (1982) writes of a madman of five perceiving channels dependin on the nature of the events. They are limited to 2 or 3 when oly duration is concerned. Althuh rhythm is usually of lesser interest in capter music, duration of son layers may be important, especially in perceiving structures. 1982 Erickson wrote of a maximmn of three melodic streams which can be maintained while listenin. In my experience this corresponds to analyzing capter music: Up to now it was impossiLble to distingih nxore than three layers of sound. If new sound qualities ICMC 551

Page  552 ï~~occur, they mix with the ones already heard into a sound layer with the nearest similarity to the new sound. Thus, even three different layers of sound can melt into one. Perceiving structures is another important feature. In most cases a structure may be fully understood only after listening to the piece several times, like Pollard - Gott researched (1983), if it can be understood fully at all. Here, research results which deal with remembering are helpful. While analyzing the music with sornams it will be senseless to take all features in account. With the exception of contradictory and unclear parts, a piece can be analyzied as fundamentally as possible and later one can ask, if all parameters which have been found could be perceived or not. Hierarchical structuring of the musical parameters have to be taken into account, although it is still unknwn kw they are structured. Analymxu sysm An analysing system had to be found which is able to seperate all different musical parameters for ctmptermusic. These are: 1. sections and subsections 2. streams of sound / layers (sounds which are seperable from each other, develop differently etc.) 3. single sound parameters 4. direction of sound in space (right, left, moving etc.) 5. tension / movement in time 6. dynamic 7. length (of sections and sounds) As a preliminary version the list can be extended with every new piece analysed. If parameters do not seem to be important or are unchanged throughout the piece (e.g. dynamic), they simply can be left out. Important are the single sound parameters (2). A sound will be described by some of the attributes from the list of parameters. There are subdivisions like: 2.1. Spectral range bottom, bottom - middle, middle, middle - top, top, centered. 2.2. Relation of tone to overtones: harnic, nonrammnic, sparse (few overtones, low amplitude), rich (many overtones, high amplitude) 2.3. Evelope: 2.3.1 beinning: attack, no attack (fade in, overtones building up, noises at start 2.3.2 continuation: steady, beating 2.3.3 ending: abrupt, decay (c in overtones, lower amplitude), noises at end, transfonnation into new sound 2.4. Space Impression: reverberation, impression of distance etc. if it belongs to the sound. If further discription is needed (e.g. if there are formants), it should be added. All musical parameters exept the single sound parameters are represented by graphical signs. Ha qthe AnalynxM System As has been said earlier, analysing will be done by reading the sonagram alongside listening. While doing so several times, all parameter signs can be written into the sonagram. Itturned out to be most practicable to write all down later on an extra sheet. All parameters should be listed for every section as a kind of abstract notation. Here, the ccsprio of section lengths, sound qualities etc. will take place as well as the interpretation of the results acoording to musicological questions. Results About 8 pieces of copter music have been analised with this system until now. Therefore, some2 results can be given, althog mo~re analysing must be done to be able to speak about copter music in general. Analysis with soagams helps to clear problems which cannot be solved just by hearing analysis. That is especially the case, when sections overlap or when similarity of sounds can be seen and better decribed literally. The lack of place ICMC 552

Page  553 ï~~allows just two short examples. The first is taken from Jonathan Harvey's "Ritual Melodies". The begirnin of the piece is shown. The two sections are similar to each other in some points, e.g. length (11 sec.) and the kind of transformation of sound in the long layer. As analysis shows later, the first part of the piece is made up of eleven sections. With one exception, their duration is 11 sec. or it's multiple. Sometimes the sections overlap, than the duration includes the overlapping. The longest section of 55 sec. consists partly of the beginning sections in original or shortcuts (8sec.). The sec -f - t i t i i t I I i I 1 I i i i 16 K --.:.- 8 K --"- ----...............,, 4 2K ga 500 1_250 11'j1 162 Harvey, Ritual Melodies Hz tions' overall duration will surely be not entirely percieved due to the capacity of short time memory. Most of the sounds seem to be in some relation to the ones of the beginning section. All in all,i the construction principle of this piece's first part is quite evident. As time movement and dynamic built longer sections, this part (and the entire piece) seems to be carefully woven and held together, as students said after listening to the piece for the first time. Hence, some of the points may be perceived unconsciously. The second example is taken from Trevor 16 K ".Wishart's "V x 5". This piece is narraK'-_" _- ot _ tive, the sections and parts follow each M. other and there are only few similarities of sounds. It seems to follow the nature 2 K.of a sounds, which in turn play with a L" perceiving time, over the largest part and shows less construction in a mathema500 - tical sense. The part shown here is a 250 constructed one, it is made up of three 125 k hierarchically interchanged repetitive ( ' ~ patterns. For me it is interesting that this part is situated in themost diffiHz Wishart, Vox 5 cult part of a piece: Around the end of the second third (in duration). Refem 1.Cogan, Robert: New Images of Musical Sound, Cambridge, Mass. (Harvard tkUiv. Press)1984 2.Erickson, Robert: New Music and Psychology, in: Deutsch, Diana: The Psychology of Music New York u.a. Academic Press)1982 3.Fraisse,Paul: Rhythm and Tempo,in:Deutsch, Diana,1982 4.Pollard - Gott, Lucy:Eiergence of Thematic Concepts in Repeated Listening to Music in: Cognitive Psychology, Vol.15 5.Smalley, Denis:Spectro-rorphology and Structuring Processes in:E uerson, Simon: The Language of Electroacoustic Music,ILondon (Macmillan Press) 1986 / Reprint 1990, 6.Stoffer, Thomas H.: Representation of Phrase Structure in the Perception of Music,in: Music Perception, Vol.3, 1985 7.Thorensen, Lasse:Auditive Analysis of Musical Structures,A stzmary of analytical terms, graphical signs and definitions in:Royal Swedish Academy of Music:IC Conference on Electro-acoustic Music, Stockholm, Sweden 1985, Royal Swedish Academy of Music o.O. 1988, S.65 ICMC 553