Page  00000303 A Computer-Assisted Music Analysis System: Daphne Jian-Li Liu Doctoral Program in Engineering, University of Tsukuba Rumi Hiraga Tsukuba College of Technology Shigeru Igarashi Department of Computer Science, University of Tsukuba Abstract A computer-assisted musical analysis system Daphne is described in its analysis functions and user interface design. Daphne is an analysis system in order to get information of musical structure and others which is usable and even necessary for systems of performance synthesis and other systems related to performance. Users can analyze in WYSIWYG way where score images and performance are provided. As well as the automated analysis or specification of analysis is directly done on the score image, the feedback of analysis is promptly shown on the score. For convincing the analysis, performance is visualized based on musical structure with analyzed information. 1 Introduction Daphne is a computer-assisted musical analysis system in order to get analysis information of musical structure and others in WYSIWYG way. Analyzed information is used for computer music systems related to performance synthesis. Daphne belongs to a computer music project Psyche [5) which pursues generating expressive musical rendition of classical music composed in the eighteenth and the nineteenth centuries. There are many systems in the project, all related to performance rendition, which are categorized to automated accompaniment and performance visualization besides performance rendition. The analysis information is used by these systems since all the systems are based on musical structure. Not many but there are some computerassisted musical analysis systems. Humdrum [6] is a powerful analysis system whose purpose is mainly score analysis which does not intend music creation. Thus it has a strong set of functions, specified by a line command style, for analyzing musical scores with statistics such as to finding the similar patterns in rhythm or melody (analyzing Leitmotiv of the whole piece of Die Walkiire [9] is an example of using Humdrum), but obtained information is not intended for musical performance rendering nor composition. Poco [3], on the other hand, is a tool for handling musical expression. Some part of Poco is used for analyzing musical performance of MIDI data. We expect Daphne to be used not only by computer scientists interested in music but also by professionals of music. In order to make the system sophisticated in the sense of both music and computer system, Daphne's user interface should be user friendly especially for professionals of music. Musical analysis with score and performance is available on Daphne as if people analyze with scores (music sheets), even with those of different editions, and performance (by CDs or their own play) on their desks or pianos in the real world. Not only analyzed information is promptly shown on the score image, but also performance is visualized using the information. 2 System Design Users can obtain the following analysis information automatically with computer-assistance or specify them by themselves. Analysis items were selected on the basis whether they are useful for performance rendering. Since expected users do not necessarily get used to computer systems, we attached importance to Daphne's user interface; let professionals in music who are not familiar with computers feel few obstacles in learning and using Daphne. 2.1 Analysis items 1. musical structure Performance is regulated based on musical structure of several levels. Based on Riemann's phrasing analysis, which takes eight measures as a norm where two and four di ICMC Proceedings 1999 -303 -

Page  00000304 visions also have musical meanings; we call the standard divisions (motif for two measures, phrase for four, and sentence for eight) structure units. Occurrences of these units are automatically analyzed to some extent. Other analytical items from 2 to 4 can be regarded as attributes of an occurrence of musical structure. Since there are exceptions in the size and the meaning of structure units, users can also specify their own structure unit and its occurrence besides the standard units. 2. relationships among occurrences of musical structure Relationships, especially the resemblance, between occurrences of musical structure regulate performance expression. Occurrences in resemblance are played resembled [7], though dynamics and agogics are changed concerning the role of each occurrence in the whole musical piece. 3. structural functions Notes in an occurrence of musical structure could have different meanings and play their own role. Initiative (the most tensional), anacrusis (a range where music is getting tenser prior to the initiative), and desinence (a range where tension is diminished following the initiative), are structural functions found in an occurrence of musical structure [1]. The information of structural functions is useful in automatic performance rendering. 4. harmony and chord progression, tonality Analysis of harmony, chord, and their progression give hints to the analysis of 1 and 2, and even to 3. The automated analysis starts from perceiving chord by its name, derives tonality, then gives functions and cadence (close) which are clues to performance expression. 5. musical forms Musical forms help understanding composer's intention. Analysis of musical forms can be derived from analyzed information from 1 to 4 to some extent. Due to the complexity of musical forms, there can often be the case that a user has to assign a form. 2.2 User Interface 2.2.1 WYSIWYG Score sheets are required for musical analysis in the real world. Thus Daphne provides users score sheets with a mouse and a popup menu, not a set of line commands, unless otherwise users have to specify positions or ranges in a piece for analysis with complicated numbers in an unfamiliar syntax. For example, in spite of the frequent appearance, it is not simple to represent the position of the second note of a triplet with the information of a bar number and the offset from the beginning of the bar. Not only for specification but also for confirmation, using score sheets is the user friendly way for musical analysis. Actually Daphne has a set of line commands which correspond to the WYSIWYG analysis. 2.2.2 Performance Replay During an analysis session, users can listen to performance (by several players as far as there are performance data in MIDI) of the piece. Using a tool of our laboratory for playing MIDI data, users can listen to performance for the required range specified on the displayed score. 2.2.3 Performance Visualization Using the analysis information, performance is visualized for confirming analysis or, conversely, visualization helps analysis. Visualized figures are not the bar graph type but new types of figures as in Hiraga's [4] or Urushibara's [11]. For example, using Urushibara's figure, the suitability of analyzed structural functions can be judged. 3 Example Daphne has been implemented on Windows 98 with Visual Basic and Visual C++. A user specifies a name of a piece to analyze, then a default set of files for the piece are loaded; a coded score using a score description language, a bitmap file for a score, an event-position file, and performance data (Fig. 1). If an analysis session is a contiguous one to a prior session of the same piece using the same score edition, a Daphne-file which holds the analyzed information is also loaded at the beginning of the session. In other words, those files except Daphne-file should be prepared previous to a session of new analysis by the user oneself or socalled Daphne-administrator, if there is. A bitmap file of a score is made from a display copy of the output from ScoreMaker, Kawai's score display software, to which a scanned image of the score is given as input. An event-position file includes x-y positions of musical events (notes) on the bitmap file. The "default set" means that there may be other sets for each edition of a score which can be loaded during the session. A coded score exists -304 - ICMC Proceedings 1999

Page  00000305 for each edition of a score because of the subtle but important differences among editions. For example, a note is ornamented with staccato in a edition, while not in others, or slur and pedals are differently specified among editions. 3. musical analysis 1. user specification: a piece (and an edition) 4. Daphne-file ---(created, updated) - bitmap file for the piece 2. files loaded - event-position file for the bitmap file by Daphne - coded score file for the piece - performance data, user actions -! file handling by Daphne Figure 1: Daphne and its 10 The output of an analysis session is called Daphne-file. A Daphne-file consists of note events each of which has augmented attributes of analysis items described in 2rl besides attributes found in a score (pitch, note value, the position represented with a measure number and metrics). After specifying a piece to analyze, Mazurka Op. 7, No. 3 by Chopin as an example, the first page of the score is displayed. Users can see other pages if necessary on multiple displays which are controlled by Windows 98. The whole pages of the piece is shown in Fig. 2. A range for analysis is notified on the score by a mouse, then followed by the popup menu (Fig. 2) where analysis options are selected: an analysis item, the way of analysis (automation or user specification), and other item specific things. Functions selectable from the top level of the popup menu other than "analysis" are to play performance either for a specified range or the whole piece and to visualize performance based on the musical structure and other information obtained by the analysis. An example scenario of analysis from the ninth to the twenty-fourth measures of the score is as follows. * musical structure This section consists of two sentences, each of which has two phrases consisting of two motifs. These are all analyzed automatically and shown as a horizontal bar. The automatic analysis of the structure is basically with the knowledge of the size of each structural unit. Auftakt for the second motif (the third beat in the tenth measure) is detected automatically. Auftakt for the fifth and the sixth motifs are detected also. * relationships Relationships are analyzed as the degree of resemblance between occurrences of musical structure. The result is shown on the bar for a structure with a referable occurrence and the degree where "1" is the highest (resemble well). "SEED" means the occurrence to be reference from which some other occurrences are developed. * structural functions A note to play the initiative role in the phrase from the ninth measure is specified with a "0" mark on a note. Some rules of performance rendering system of Psyche requires the information of initiative (8]. It is also used by Urushibara's performance visualization [11]. 4 Conclusion A computer-assisted musical analysis system Daphne is described in its design and an example usage. The obtained information through Daphne with the score imitating the real world user interface is used by systems related to performance rendering. As described in 2.1, Daphne adopted Riemann's musical structure currently. Though we think any one theory of musical structure is not enough for the purpose of generating expressive performance, some widely accepted theories of musical structure, such as Implication-Realization model [2] or Generative Theory of Tonal Music [10], should be in concern in the future for broadening musical perspectives. In order to facilitate user interface, we will involve a table to show correspondence between occurrences of musical structure with tonality information. As for computer-assistance of Daphne, besides the current function of providing users candidates of analysis information automatically, it is planned to include functions of proposing analysis items for a range of a musical score by understanding score information and analysis so far. Analysis database for holding analysis information from several editions, proposed by a professional in music, will be a new application of Daphne. The database will be used for understanding the difference between editions and for analyzing performance by several players. ICMC Proceedings 1999 -305 -

Page  00000306 rii"~E~EIL Figure 2: An example of using Daphne References [1] Berry, W. T.: Structural Functions in Music, Dover Publications, 1987. [2] Cooper, G. and Meyer, L. B.: The Rhythmic Structure of Music, University of Chicago Press, 1960. [3] Desain, P. and Honing, H.: POCO: an environment for analysing, modifying, and generating expression in music, Proc. of ICMC, pp. 364-368, 1990. [4] Hiraga, R., Igarashi, S., and Matsuura, Y.: Visualized Music Expression in an ObjectOriented Environment, Froc. of ICMC, pp. 483-486, 1996. [5] Hiraga, R. and Igarashi, S.: Psyche: University of Tsukuba, Computer Music Project, Proc. of ICMC, pp. 297-300, 1997. [6) Huron, D.: The Humdrum Manual, Center for Computer Assisted Research in the Humanities. [7] lyatomi, A., Hiraga, R., and Igarashi, S.: Unfolding Performance Expression with Rules Based on Music Structure, Proc. of the Annual Conference of JSAI, 11, pp. 280-282, 1997. [8] Koike, H., Hiraga, R., Igarashi, S., Mizutani, T., and Shio, M.: Automatic decision system of parameter values for agogic rules, SIGMUS, IPSJ, 31, 1999. [9] Kornstadt, A.: SCORE-to-Humdrum: A Graphical Environment for Musicological Analysis, Computing in Musicology, 10, Center for Computer Assisted Research in the Humanities, pp. 105-122, 1995. [10] Lerdahl, F. and Jackendoff, R.: A Generative Theory of Tonal Music, The MIT Press, 1983. [11] Urushibara, M., Hiraga, R., Igarashi, S., and Koike, H.: Performance Visualization and its Application based on Music Structure, Proc. of the Annual Conference of JSAI, 13, pp. 167-170, 1999. -306 - ICMC Proceedings 1999