COMPUTER MUSIC ANALYSIS VIA A MULTIDISCIPLINARY APPROACH

analysis which pays attention to the creative process. Regarding the first approach we can mention the major studies by Denis Smalley [2], Simon Emmerson [3], Michel Imberty [4], François Delalande [5], Francesco Giomi and Marco Ligabue [6].


These are all inspired by the pioneer researcher Pierre Schaeffer [7] and discussed in a recent book by Sthéphane Roy [8]. Other studies aim to graphically represent the electro-acoustic music flux in multimedia contexts, starting from the musicologist 9s personal listening. All these works aim to describe the listening in order to understand the musical structure and/or timbre.


They sometimes use acoustic representation tools (time- amplitude representations, spectrograms, sonograms). The poietic analysis is a recent research trend which tries to contrast the inevitable individuality involved in the listening process. It studies the composition process [9] [10], or uses computer data as 8objective 9 material to be analysed; one of the first and rare studies was Lorrain 9s analysis of Inharmonique by Jean-Claude Risset [11].


Nevertheless heterogeneity is the characteristics of these researches. We firmly think that all this variety could in part or completely be clarified by a multidisciplinary approach which combines musicology with computer science and perception. Our proposal is to begin to operate towards this research paradigm.


We want to explore how feature extraction and audio content description studies can be useful to the needs of musicology. Up to now the research done in this area is applied, from one hand, to describe single sound objects, on the other hand to automatically transcribe traditional western music or popular music. It is time to work also in the field of electro-acoustic music.


We think that automatic analysis is a useful tool to study the classification of electro-acoustic sounds, their description, the structure derived from their polyphonic overlapping, the style of this music. Automatic analysis can help generating automatic segmentation and/or description of the sounds. All this study must be supported by the knowledge of the digital synthesis used by the author during the compositional process.


This musicological competence aims to give further evidence of the musicologist 9s personal listening. Our research focused on the analysis of a computer music piece: Winter leaves (EDIPAN-PRC-S20-16, 1984), for tape, created in 1980 by Mauro Graziani at the CSC (http://www.dei.unipd.it/ricerca/ csc/) in Padova, using Music360 software (the piece was created with an IBM S7 connected to an IBM 370/158, duration: 8 926 d). This work derives from a precedent analysis by Laura Zattra based on listening (description and graphic score) and analysis of the digital score [12].


The automatic analysis, starting from a reflection on the sound objects, helps the listening and the identification of sound objects 9 flow, and above all, it is an important means to study the problems of the computer music analysis. Winter leaves is therefore a case study. The validity of its research 9s approach needs to be verified with other musical pieces in order to establish an analytical method for the analysis of electro-acoustic music.


We are going to show the results focused on a fragment of Winter leaves , so that we illustrate the method which was tested on the whole piece. The 2 nd section shows the results obtained by Savino Porcelluzzi; the 3 rd section follows and describes the research made by Francesca Nucibella; the 4 th section, by Laura Zattra, compares these results with the musicological analysis. 2.


ANALYSIS OF COMPUTER MUSIC WITH MORPHOLOGICAL DESCRIPTOR TITLE 2.1. The Morphological Descriptor The tool that we first experimented to automatically analyze Winter leaves is called cMorphological Descriptor d (from now MD), designed at Music Technology Group in the UPF of Barcelona by Julien Ricard and Perfecto Herrera [13][14]. It is based on the morphological theory of sounds objects by Pierre Schaeffer [7].


The choice of the MD is due to its performance in the analysis of singles sounds objects. The Schaeffer 9s typo-morphology and his computational implementation seems to describe quite well a sound object, so we decide to use an automatic approach to analyze a piece of electroacustic-music. Nevertheless, in order to adapt the MD to the analysis of an entire piece of music, to pass from the theory to the practice, we need to introduce some simplification and changes.


In fact the MD could analyze only the following criteria: " Dynamic profile: describes the shape of the temporal envelope. " Pitchness profile: discriminates sounds with one predominant pitch (called Pitched), sounds with several pitches (called Complex) and sounds with no pitch (called Noisy). " Pitchness profile: describes whether the pitchness is constant or varies in function of time (in that case, pitchness is the mean value).


" Pitch profile: describe the variation of the pitch, only specified for pitched sounds. For sounds with unvarying pitch, the pitch is given. Pitch-varying sounds are classified according to the type of variation (continuous or stepped) as well as the global envelope of the pitch (e.g.


ascending, descending...). " Harmonic timbre criteria, specified by a numerical value of brightness. " Roughness, described by a numerical value.


According to these criteria, a piano phrase of several low-frequency ascending notes, for instance, would be described as follows: dynamic profile = 'iterative', pitchness profile = 'pitched', pitchness profile = 'unvarying', pitch variation type = 'varyingstepped', pitch envelope = 'ascending', a low brightness value and a low roughness value. 2.2. Procedure of Analysis with MD Our procedure of analysis needs 3 steps procedure: 1) Execution of the program (give the file.wav in input and receive a file.txt in output) 2) analysis of the results given by MD on the file.txt and confrontation with a listening analysis made with an audio editor.


3) if needed, re-compile the program varying threshold values and restart the procedure of analysis. We list here the problems we encountered and the solutions we found: " The Morphological Descriptors MD needs a heavy execution time. It needs more or less 3 hours for analyzing a fragment of 2 minutes, so we had to divide the piece in several tracks to reach a reasonable amount of execution time.


" The MD cannot analyze stereophonic tracks, so we had to mix-down every single track from stereo to mono with a balance of 50% between the right and left channel. " It cannot point out sounds with a frequency over 5 Khz, but luckily in this piece there is any sounds over this frequency. " The MD was designed to analyze singles sounds objects, but in this piece there is a lot of polyphony.


So the only thing that we could do, was to put a higher threshold value. In this way the MD thought to observe less objects but with a good dynamic value. This allowed to take a description more efficient for objects which could not merge in the sound stream.


2.3. Results We show here the results of the analysis of a fragment of Winter Leaves in table 1 (it begins at 3 902 9 9 of the piece and ends at 4 902 9 9). This is a list of the output description given by the MD.


The first 2 columns express the time of beginning and ending of the sound objects detected. The other columns indicate: pitchness, pitch profile, pitchness profile. In the last one a double slash indicates objects with a very short duration (< 1 sec.).


This means that in this case we have not verified if the description was correct, because it is not possible for to ear to hear and manually describe a sound so short. We point out some particular these detections: " 0,0,9755(s) 8,7592(s) Pitched Varying_Other Varying_Stepped Varying Decreasing = this is a good detection, in fact it is the beginning of a fragment with a high dynamic and not so much polyphony. " 37,4261 38,8384 Pitched Varying_Delta Unvarying Unvarying Undefined = the analysis detects a pitched sound but any percussive, iterative sound.


Table1. Description of sound objects detected PITCHNESS PROFILE T. B .


T. E . PITCHNESS P=Pitched C= Complex N= Noisy DYNAMIC PROFILE PITCH PROFILE U= Unvarying V=Varying I=Increasing D=Decreasing O=Other Und=Undefined 0 0,3714 P Varyng_Decrescendo Varying_Stepped U I -1 0,10221 0,3714 0,9755 P Varying_Other Varying_Stepped U O -1 0,23769 0,9755 8,7592 P Varying_Other Varying_Stepped V D -1 0,37963 8,7592 8,9510 C Varying_Delta Unvarying U Und -1 0,31655 8,9510 27,2735 P Varying_Other Varying_Stepped V I -1 0,26064 27,2735 27,4245 C Varying_Delta Unvarying V Und -1 0,53926 27,4245 29,7800 C Varying_Delta Unvarying V Und -1 0,36321 29,78 30,0616 N Varying_Crescendo Unvarying V Und -1 0,69028 // 30,0616 30,2167 C Varying_Delta Unvarying V Und -1 0,42656 // 30,2167 30,6167 P Varying_Delta Unvarying U Und 1225 0,23060 // 30,6167 32,0738 C Varying_Decrescendo Unvarying V Und -1 0,36687 32,0738 33,1024 P Varying_Delta Unvarying U Und 1188 0,39163 33,1024 33,5228 C Varying_Other Unvarying U Und -1 0,20126 // 33,5228 34,0208 P Varying_Other Unvarying U Und 393 0,27205 // 34,0208 35,0004 P Varying_Decrescendo Varying_Continuous U O -1 0,17695 35,0004 35,5024 P Varying_Decrescendo Unvarying U Und 1157 0,09849 // 35,5024 35,5881 P Varying_Crescendo Unvarying U Und 291 0,70269 // 35,5881 35,9881 C Varying_Delta Unvarying U Und -1 0,13035 // 35,9881 36,1187 C Varying_Delta Unvarying U Und -1 0,43168 // 36,1187 36,2412 C Varying_Delta Unvarying V Und -1 0,49766 // 36,2412 36,3555 C Varying_Delta Unvarying U Und -1 0,34555 // 36,3555 36,4861 C Varying_Delta Unvarying U Und -1 0,44900 // 36,4861 36,6126 C Varying_Delta Unvarying V Und -1 0,46434 // 36,6126 36,7391 C Varying_Delta Unvarying U Und -1 0,49512 // 36,7391 36,8575 C Unvarying Unvarying V Und -1 0,41123 // 36,8575 37,1065 C Varying_Other Unvarying U Und -1 0,21969 // 37,1065 37,6004 C Varying_Delta Unvarying U Und -1 0,13153 37,6004 37,8371 C Varying_Delta Unvarying V Und -1 0,19252 // 37,8371 38,0942 C Varying_Delta Unvarying U Und -1 0,27311 // 38,0942 38,3391 C Varying_Delta Unvarying U Und -1 0,16201 // 38,3391 38,7351 C Varying_Decrescendo Varying U Und -1 0,15142 // 38,7351 40,4861 P Varying_Other Unvarying U Und 1188 0,15177 40,4861 41,8984 P Varying_Delta Unvarying U Und 1181 0,20247 41,8984 42,8780 P Varying_Delta Unvarying V Und 192 0,20836 42,8780 42,9922 C Varying_Crescendo Unvarying U Und -1 0,57216 // 42,9922 43,0861 C Unvarying Unvarying U Und -1 0,46519 // 43,0861 43,2576 C Varying_Delta Unvarying U Und -1 0,38774 // 43,2576 43,8290 C Varying_Delta Varying V Und -1 0,43035 // 43,8290 44,8616 P Varying_Other Unvarying U Und 1208 0,23120 44,8616 47,6657 P Varying_Other Varying_Stepped U I -1 0,33272 47,6657 48,4820 P Varying_Other Unvarying U Und 1181 0,48413 48,4820 48,8208 P Varying_Other Unvarying U Und 1204 0,55355 // 48,8208 49,3024 P Varying_Other Unvarying V Und 1168 0,48615 // 49,3024 49,7106 P Varying_Other Unvarying V Und 1166 0,38821 // 49,7106 50,1229 C Varying_Delta Unvarying U Und -1 0,43948 // 50,1229 50,5351 P Varying_Decrescendo Unvarying U Und 1177 0,41058 // 50,5351 51,7555 P Varying_Other Unvarying U Und 1168 0,37407 51,7555 55,2127 P Varying_Other Unvarying V Und 1366 0,43397 55,2127 56,7678 P Varying_Other Unvarying U Und 1158 0,76217 56,7678 59,1229 N Varying_Other Unvarying V Und -1 0,43566 59,1229 59,5229 N Varying_Crescendo Unvarying U Und -1 0,51220 // 59,5229 59,6371 N Varying_Crescendo Unvarying U Und -1 0,58347 // 59,6371 59,7514 N Varying_Other Unvarying U Und -1 0,53471 // 59,7514 59,8739 N Varying_Delta Unvarying U Und -1 0,57636 // 59,8739 60 N Varying_Impulsive Unvarying U Und -1 0,33112 //end 2.3.


Discussion When the so