AI-supported processing of handwritten transcriptions for hungarian folk songs in a digital environment

Размещено на http://allbest.ru

Liszt Ferenc Academy of Music (Budapest) RCH Institute for Musicology

AI-supported processing of handwritten transcriptions for hungarian folk songs in a digital environment

Matyas Bolya Dr. Matyas Bolya is an Associate Professor at the Liszt Ferenc Academy of Mu-sic (Budapest) as Head of the Folk Music Department, and coordinator of the plucked string instrument division. Since 2014 he has been a research fellow at the Institute for Musicology (Budapest)¹ as digital transition coordinator of the Folk Music Archives. He founded the Folk Music Research Group at the Liszt Academy in 2022. His main field of interest is Hungarian folk music, the history of its research (with focus on the instrumental music), describing the musical folklore phenomenona with the help of database tools, the history of Folk Music Archives, folk music classification through artificial intelligence, modern plan-ning processes from archiving to publishing, as well as the influence of the tech-nical background on the research methodology., DLA

Budapest



Abstract

My research focuses on creating an AI-supported Digital Research Environment (DRE) that helps analysing and systematizing folk music tunes with the help of the latest information theory and database management results. The study may be extended to the entire source material accumulated by researchers so far, thus integrating Hungarian ethnomusicology results of the last hundred years. In this way, new dimensions of structural analysis open up and a large amount of information can be processed that already exceeds the limits of human musical memory.

Previous computerized music analysis experiments in Hungary have inadequately defined the role of artificial intelligence. In our case, the AI-supported digital environment that is the subject of the research does not work independently, because the researcher's scientifically abstract thinking, preferences, and the recognition of characteristic melodic elements cannot yet be replaced by computer data processing.

Crucial goal of the research is to precisely define the researcher's role in musical data processing. Thus the attitude of researchers rejecting software support may change in favour of actually using our digital framework.

For the first time in Hungarian folk music research history, a detailed and documented digital research environment can be created, integrating the useful, relevant software tools.

We can map out data entry problems and define the standard format of the musical data suitable for mass input and analysis.

If possible, we will replace the previously widely used optional data with scalable data to have a broader range of parametrization and search options, and their free combination allows us to study new scientific models. With DRE, the validity range of the previous scientific musical classification can be more precisely specified and the processing as well as classification of unreported melodies and the process of type creating can be significantly accelerated.

The most significant debate in the previous research has been the dataset specification of analyses. I am convinced that only similarly processed tune-data-elements can be compared, so one of the most critical tasks is to determine the input data's standard format and information density. As a first step, the digital conversion of the musical manuscript needs to be solved. International research has mainly led to results in the recognition of printed music, some of which can be used in the project, but many new developments are also needed.

Keywords: Al-supported Digital Research Environment (DRE), Optical Music Recognition (OMR), Musical Manuscripts, Hungarian Folk Songs, scientific musical classification, ethnomusicology, digital archives, folklore database.



Introduction This article is a transcript and edited version of a presentation given at the NETWORK- SHOP 2022 national IT conference in Debrecen, 22 April 2022. The presentation is avail-able here (in Hungarian): URL: https://prezi.com/view/4WH6mWduD8DuUHyrwKOp/ (Access: 23.10.2022)

The focus of the three-year research programme is the development of a digital environment supported by artificial intelligence to facilitate the analysis and systematisation of folk music melodies, based on the experience of Hungarian folk music research and the latest results of information theory and database management. In this article, I present the main issues and first results of the programme, which was launched at the end of 2021 [Bolya 2021-2024]. This research support software environment uses digital melody databases and AI as a tool, with musicology as the conceptual framework. By integrating the research results of the last one hundred years, the study can be extended to the entire source material accumulated by our predecessors. In this way, by leaping orders of magnitude, it is possible to process a vast amount of information that already exceeds the limits of human musical memory.

The first ethnomusicologists, working with a truly modern methodology all over Europe, became interested in questions of the systematisation of melodies. This fitted in well with the prevailing intellectual currents of the turn of the 19th and 20th centuries: the thousands of melody data collected by Hungarian researchers were sufficient for valid studies and at the same time satisfied the factual and systematic demands of positivism. This was complemented by the influence of the theory of evolutionism, which can also be traced in the research of music as an integral part of general human culture; it drew successive epochs, gradual changes on the blank pages of Hungarian folk music history.

The primary - and practical - aim of the systematisation endeavours was to register the vast paper collections, as only then could they become effective research aids. Our first ethnomusicologists were, in addition to being researchers, highly qualified musicians, so the need for a synthesising scientific concept soon emerged, alongside the development of the necessary principles of registration. Thereafter, the two activities - registration and systematisation - ran in parallel; the researcher's habitus, focus and work ethic determined the degree of completion and the afterlife of these scientific constructs.

To be able to find a principle for melody sorting, it was first necessary to break away from the methodology of linguistics. At the turn of the century, it seemed obvious that the sorting of melodies could be done on the basis of textual analogies. This thought barrier was due to the fact that, thanks to centuries of library practice, there was much greater experience available in the systematisation of texts and textual information. However, sorting the initial tones of melodies by height, based on the analogy of the alphabetical order, was not successful: experience had shown that the beginning of melodies simply did not give a character pattern and could not be used as a grouping element [Scheurleer 1899-1900, p. 219-220] The need for sorting from a musical point of view is well illustrated by a call for entries in 1900; the translation of the call is: “What is the best method of sorting folk and popular songs by their musical (and not textual) characteristics, in a dictionary-like manner?” The two approaches are perfectly reflected in the two entries received. The Viennese music historian Oswald Koller used the initial notes of the melodies as the basis for sorting, using the dictionary alphabet as a model. The other applicant was Ilmari Krohn, a Finnish composer whose idea was further developed by Bela Bartok and Zoltan Kodaly and applied to the arrangement of Hungarian folk songs..

The solution in short: the vast majority of folk songs are arranged in lines, and the sequence of numbers describing the relative position of the end notes is a unique identifier that is already suitable for forming larger arrays and establishing kinship relations. To this end, the melodies were transposed to a common root note for ease of comparison and, as the methodology evolved, further investigative aspects and technical principles of sorting, not detailed here, were introduced, from syllable number through rhythmic patterns to generalisation of melodic lines.



Background

I published results serving directly as a basis of our research in a separate volume entitled Information Theory and Hungarian Folk Music Research [Bolya 2019]. folk music digital hungarian

In the research phase, closed by compiling the book, I examined the most important stages of sorting of Hungarian folk music with the help of information theory, collected aspects of melodic analysis, grouped them, then examined earlier sorting principles according to the unified system of criteria thus developed, and finally formulated the sorting sequences. The different scientific models thus became comparable, emphasis of creator's intentions became clear and it became evident at which stage of the system designing process the systems under study were at.

Several databases can be considered as precursors of my research programme [Bolya 2014; Bolya - Both - Kukar 2019; Both et al. 2018, Bolya - Both et al. 2021], the most important being the Sound Archive of the Institute of Musicology [Bolya 2019/2020/2020/2021, 4th edition], which has already completed its fourth phase of development in 2021. The Sound Archive is not only a precursor but also the primary source database for the research. The freely accessible, vast database Around 200,000 tunes, representing 12,000 hours of recordings. also contains music descriptive data, with a multidimensional referencing system linking metadata to sound recordings, transcriptions, paper documentation, geocodes and time codes.

International literature of the topic is much richer than Hungarian language one. Digital musicology is seen as part of digital humanities, now considered traditional abroad [IMS Study Group “Digital Musicology”; Collections of articles on Frontiers].

Nevertheless, there is a dichotomy in this field: while many results of theoretical research are presented in projects and articles [Blot et al. 2017; Paulus - Mtiller - Klapuri 2010; West et al. 2010; De Roure et al. 2010; CHARM 2009 etc.], these results are rarely integrated into the search and analysis modules of large databases [Glaubitz 1996-2013; Bergel et al. (n.d.); Toiviainen - Eerola 2004; British Library Sounds 2020].

Objectives of the research

In the field of scientific modelling of the systematization of melodies, we have relied mainly on the musical memory of a few outstanding researchers and have mostly used empirical methods for musical type creation. Bartok, Kodaly, Jardanyi, Dobszay and Bereczky Famous Hungarian Folk Music Researchers: Bela Bartok (1881-1945), Zoltan Ko-daly (1882-1967), Pal Jardanyi (1920-1966), Laszlo Dobszay (1935-2011) and Janos Bereczky (1942-). built up their systems on the basis of patterns they had intuitively perceived by studying thousands of melodies, and so they established the justly world- famous Hungarian analytical folk music systematics. A comprehensive study of the Central Folk Music Collection, containing nearly 250'000 melodies and serving as the archival background of their research, however, is beyond the limits of human memory.

Today, the combined capacity of computerized data processing and the constantly evolving technology can help advance the methodology. This allows us to extend the scope of sorting melodies and to perform set analysis and statistical analysis based solely on descriptive data, independently of the scientific structures that have strongly dominated previous research thinking.

I believe that the research programme described above is able to integrate the results of the last hundred years of Hungarian folk music research by creating a digital research environment, and that the study can thus be extended to the entire source material accumulated by our predecessors. In this way, new dimensions of structural analysis open up and a large amount of information can be processed that goes beyond the limits of human musical memory. Previous attempts at computer-based melodic analysis have not adequately appreciated the role of artificial intelligence. In our case, the AI-assisted digital tool that is the subject of this research does not work independently, since the scientifically abstract reasoning, weighting and identification of the dominant melody-forming elements by the researcher cannot yet be replaced by computerized data processing.

Processing steps

An important objective is to precisely define the place of the researcher in the process of melody data processing, what also may soften the attitude of researchers who basically reject software support. For the first time in the history of Hungarian research, a well-documented digital research support environment can be created, integrating software tools that can be utilized for folk music research.

We will explore problems of data entry, and will define a standard format for mass melody data input and analysis. We will replace the previously used optional descriptive data with scalable data where possible, providing a wider range of parameteri- sation and search possibilities; their free combination will also allow the testing of scientific models not yet tested. The tool will also allow to better define the range of validity of previous scientific melody systematization attempts, and significantly speed up the processing and classification of unsorted melodies, as well the work of type creation.

The most controversial factor in the experiments so far has been the database of analyses. I am convinced that only melody database items with similar processing can be compared, so one of the most important tasks is to define a standard format and information density for the input data. At present, artificial intelligence is not able to recognize the correlation between a simplified melody skeleton and a richly ornamented melody variant with a different syllable count and a scientific level, detailed notation.

Even within the profession, the recognition of such a melodic relationship is often controversial, indicating that researcher subjectivity - well understood and scientifically justified - plays an important role in modelling. In parallel, we will explore the problems of data input and define a standard format for melody data (i.e. time-pitch function) suitable for mass input and analysis.

As a first step, the digital conversion of the manuscript melody notes needs to be solved. International research has mainly led to results in the recognition of sheet music, some of which can be used in the project, but many new developments are also needed. This will be followed by the definition of the melody skeleton - an area where AI may be of help - and the creation of the melody database, and finally the development of a digital environment for comparative analysis. I will report on these phases later on.

Optical score recognition

Recognition of sheet music, i.e. the conversion of digital image formats into digital audio information (Optical Music Recognition) [Bain- bridge - Bell 2001; Rebelo - Ana - Fujinaga - Ichiro - Paszkiewicz - Fili- pe et al. 2021] is a dynamically developing field with a number of software solutions available on the market today.

Music is, in simple terms, a time- pitch function, the visual representation of which is the musical score. It is important to emphasise that the graphical placement of the symbols expressing the sound value does not correspond to the sound value, nor is it proportional to it. The reason for this is readability, as the graphic image would be too disorganized if time units were also respected. Thus, the practical representation of music as a function does not conform to mathematical rules, and this will be an important element during coding.

The process of sheet music recognition is inadvertently compared to text recognition, but there are many important differences. Although the symbol system of general musical notation, not counting instrument- specific symbols, consists of relatively few symbols, their placement and combination determine the interpretation. Music notation therefore has a specific grammar, without which decoding is impossible.

Character recognition, on the other hand, does not go beyond the identification of letters and words, i.e. it is a one-dimensional reading of information adapted to the baseline reference.

Fig. 1. Time Signature Spacing

Fig. 2. Note Spacing

But what about manuscript scores? Decoding musical information is a complex task not only for algorithms but also for a musician. It follows that handwritten sheet music is necessarily more accurate than a handwritten text. An illegible word does not prevent further interpretation of the text; knowledge of the language and context triggers an automatic correction process in the reader (i.e. not necessarily by the decoding program).

Music without semantics, however, is meaningless. If, for example, a musician cannot decide whether a note falls on a line or between lines, the decoding process is interrupted. But there is no doubt, that in the case of manuscript music, simplifications to speed up writing and violations of grammatical rules are often encountered.

Problems arising during the processing of manuscripts

In the past 120 years of Hungarian folk music research, nearly 200 000 manuscript records of varying degrees of detail have been written. Why is it important to digitise these at a time when software-based sound recognition and processing is developing at an almost unimaginable pace? The question is legitimate and it is certain that these two data collection methods will meet at a later stage of research. These records are time capsules in which the scientifically abstract thinking of the researchers of the given time can be traced. By processing these manuscripts, the irreplaceable experience of generations of researchers can be channelled into the data analysis process.

This brings us to the first practical task, which is to establish the right ratio of AI to human resources in the processing. Since the structure and mark-up of manuscripts are not uniform, some tasks cannot be left to machine algorithms. More specifically, the development effort would not be proportional to the expected result. Examples of such elements, which are crucial for analysis, are the recognition of melody articulation, the identification of lines, the decoding of abbreviations and the definition of the analysis area. After studying thousands of manuscript transcriptions, the following typical factors that make decoding difficult were identified: 1) the different types of simplification, which have become widespread due to the need to save paper and to speed up work; 2) several melody data in one image; 3) identification of the area containing useful information See Appendix for examples..



Plans

It can be concluded that previous attempts at computer-based melody analysis have not defined the role of researchers and artificial intelligence properly, and this has led to mistrust towards such initiatives on the part of folk music researchers. Hopefully, our research will resolve this contradiction by maintaining researcher control and allowing everyone to freely try out the tool even at the development stage.

Our research will help to draw a more distinct line between AI and researcher roles, and allow for analyses that go far beyond human musical memory. The results will extend the reach of Hungarian folk music research into the field of non-strophic material and instrumental music as well. The system being developed will be able to process any kind of folk music database, and international cooperation will also make it possible to compare different music cultures. By means of the help of information technology new dimensions of structural analysis will open up and processing of large amounts of information will be possible.

Results may also be useful in the training of researchers, as the interface can be used to learn the well-recognised analytical thinking of Hungarian folk music research. By generalising the experience, at a later stage, research may be extended to the study of larger musical units found in other genres, as well as towards interdisciplinary collaborations. In the longer term, the tool is also likely to attract international interest, as there is currently no comparable complex software support for melody analysis on the market.



References

Bainbridge, D. - Bell, T. (2001). The Challenge of Optical Music Recognition. Computers and the Humanities. Nr 35. P. 95-121. URL: https://doi. org/10.1023/A:1002485918032 (Access: 23.10.2022).

Bergel, G. et al. (n.d.). Broadside Ballads Online from the Bodleian Libraries. URL: http://ballads.bodleian.ox.ac.uk/_(Access: 23.10.2022).

Blot, G. et al. (2016). Melody and Rhythm Through Network Visualization Techniques. 12th International Symposium on Computer Music Multidisciplinary Research (Sao Paulo, Brazil, July 5-8. 2016). Revised Selected Papers. URL: http:// dx.doi.org/10.1007/978-3-319-67738-5_2 (Access: 23.10.2022).

Bolya, M. - Both, M. - Kukar, M. (2019). Ethiofolk. Online Ethiopian Folk Music and Folk Dance Archive. Polyphony, ELKH RCH Institute for Musicology. URL: www.ethiofolk.com/ (Access: 23.10.2022).

Bolya, M. - Both, M. et al. (2021). Folk_ME. Folk Music Education for Future Generations. Polyphony. URL: https://www.folk-me.com/hu_(Access: 23.10.2022).

Bolya, M. (2019). Informacioelmelet es nepzenekutatas. Rendszeralkotas, ny- ilvantartas, digitalis archvvum. [Information Theory and Hungarian Folk Music Research. Musical Systematization, Registration, Digital Archive]. L'Harmattan, HAS RCH Institute of Musicology, Budapest.

Bolya, M. (2021-2024). Nepzenei dallamrendezesMl-vel tamogatott digitalis kornyezetben. [Al-supported Digital Research Environment for the Classification of Folk Tunes]. `OTKA' postdoctoral excellence programme (PD_21), project ID: 137969. URL: https://nkfih.gov.hu/palyazoknak/nkfi-alap/tamogatott-projektek- pd21 (Access: 23.10.2022).

Bolya, M. (2022). Keziratos dallamlejegyzesekfeldolgozasa Ml-vel tamoga- tott digitalis kornyezetben. [Al-supported Processing of Handwritten Transcriptions for Hungarian Folk Songs in a Digital Environment]. Online presentation. URL: https://prezi.com/view/4WH6mWduD8DuUHyiwKOp/ (Access: 23.10.2022).

Bolya, M. (ed.) (2019/2020/2020/2021). Zenetudomanyi Intezet HangarMvum. [Sound Archives of the Institute for Musicology]. ELKH RCH Institute for Musicology, 4th edition. URL: https://zti.hungaricana.hu/en/ (Access: 23.10.2022).

Both, M. et al. (2018). Polyphony Project. Online FolkMusic Archive. Online archiving system for processing and publishing Ukrainian folk music collections. URL: www.polyphonyproject.com. (Access: 23.10.2022).

British Library Sounds. World & traditional music. (2020). URL: https:// sounds.bl.uk/World-and-traditional-music/_(Access: 23.10.2022).

CHARM (AHRC Research Centre for the History and Analysis of Recorded Music). (2009). URL: https://charm.rhul.ac.uk/about/about.html (Access: 23.10.2022).

Collections of articles on Frontiers. URL: https://www.frontiersin.org/jour- nals/digital-humanities/sections/digital-musicology_(Access: 23.10.2022).

De Roure, D. et al. (2010). SALAMI. Structural Analysis of Large Amounts of Music Information. URL: https://www.researchgate.net/publication/44023837_ SALAM[_Stmctural_Analysis_of_Large_Amounts_of_Music_]nfoimation (Access: 23.10.2022).

Glaubitz, R. (1996-2013). The Aria Database. URL: http://www.aria-data- base.com/_(Access: 23.10.2022).

IMS Study Group “Digital Musicology”. URL: https://www.musicology.org/ networks/sg/digital-musicology_(Access: 23.10.2022).

Paulus, J. - Muller, M. - Klapuri, A. (2010). Audio-based music structure analysis. Proceedings of the 11th International Society for Music Information Retrieval Conference, ISMIR 2010. P. 625-636. URL: https://www.researchgate.net/publica- tion/289662964_Audio-based_music_structure_analysis (Access: 23.10.2022).

Rebelo, A. - Fujinaga, I. - Paszkiewicz, F. - R. S. Marcal, A. - Guedes, C. - S. Cardoso, J. (2012). Optical music recognition: state-of-the-art and open issues. International Journal of Multimedia Information Retrieval 1. P. 173-190. URL: https://doi.org/10.1007/s13735-012-0004-6 (Access: 23.10.2022).

Scheurleer, D. F. (1899-1900). Welches ist die beste Methode, um Volk- und volksmaBige Lieder nach ihrer melodischen (nicht textlichen) Beschaffenheit le- xikalisch zu ordnen? [What is the best method of sorting folk and popular songs by their musical (and not textual) characteristics, in a dictionary-like manner?]. Zeitschrift der International MusikgeseUschaft. Nr 1. P. 219-220. https://ar- chive.org/details/ZeitschriftDerInternationalenMusikgesellschaft011899-1900/page /n221/mode/2up (Access: 23.10.2022).

Toiviainen, P. - Eerola, T. (2004). Digital Archive of Finnish Folk Tunes. Database. URL: http://esavelmat.jyu.fi/index_en.html_(Access: 23.10.2022).

West, K. et al. (2010). The Networked Environment for Music Analysis (NEMA). P 314-317. URL: https://doi.org/10.1109/SERVICES.2010.113 (Access: 23.10.2022).



Анотація

Обробка рукописних транскрипцій угорських народних пісень у цифровому середовищі за підтримки штучного інтелекту

Доктор (PhD) Матіас Боля (Matyas Bolya) - доцент Музичної академії імені Ференца Ліста (Будапешт), завідувач кафедри народної музики та координатор відділу щипкових струнних інструментів.

Музична академія Ференца Ліста (Будапешт) Центр гуманітарних досліджень Інституту музикології (Будапешт)

З 2014 року - науковий співробітник Інституту музикознавства (Будапешт) як координатор оцифрування Архіву народної музики. 2022 року заснував дослідницьку групу народної музики в Академії Ліста. Основна сфера його інтересів - угорська народна музика, історія її дослідження (з акцентом на інструментальній традиції), опис феномена музичного фольклору, послуговуючись інструментарієм баз даних, вивчення історії архівів народної музики, класифікація народної музики за допомогою штучного інтелекту, сучасні процеси планування - від архівування до публікації, а також вплив технічної бази на методологію дослідження. Список публікацій: МТМТ. Угорська наукова бібліографія.

Пропонована стаття висвітлює дослідження трирічної наукової програми, що спрямована на створення цифрового дослідницького середовища (Digital Research Environment) за підтримки штучного інтелекту, щоб сприяти аналізові та систематизації народних мелодій на підставі спадкоємного досвіду угорської етномузикології та останніх результатів теорії інформації й управління базами даних. Репрезентовано основні тези та перші результати програми, впровадженої наприкінці 2021 року. Цифрове дослідницьке середовище використовує бази даних оцифрованих мелодій як ресурс, штучний інтелект як інструмент, а музикознавство як концептуальну основу.

Дослідження також ґрунтується на наявних міжнародних розробках, у яких цифрове музикознавство трактують як частину цифрової гуманітарної науки. Водночас у цій галузі існує дихотомія: хоча у проєктах і статтях описано багато результатів теоретичних досліджень, однак їх рідко інтегрують у модулі пошуку та аналізу великих баз даних.

Детально базові результати дослідження опубліковано в монографії під назвою „Теорія інформації та дослідження угорської народної музики” [Bolya 2019].

У книзі проаналізовано найважливіші етапи класифікації угорської народної музики за допомогою теорії інформації, зібрано та згруповано аспекти мелодичного аналізу, перевірено їх відповідність єдиній системі, отже, таким чином розроблено критерії та остаточно сформульовано послідовності сортування мелодій.

У галузі наукового моделювання систематизації мелодій узято до уваги досвід кількох видатних дослідників і здебільшого використано емпіричні методи створення музичних типів. Барток, Кодаї, Ярдані, Добшай і Берецький побудували свої системи інтуїтивно, вивчаючи тисячі мелодій, і відтак створили всесвітньо відому угорську народномузичну аналітичну систематику.

Сьогодні комбінована потужність комп'ютеризованої обробки даних і технології, що постійно розвиваються, можуть допомогти вдосконалити методологію. Це дає змогу розширити сферу сортування мелодій, виконувати аналіз наборів і статистичний аналіз винятково на підставі описових даних, незалежно від наукових концепцій.

Попередні експерименти з комп'ютеризованого аналізу музики в Угорщині неадекватно оцінювали роль штучного інтелекту. У нашому випадкові цифрове середовище з підтримкою штучного інтелекту, яке є предметом дослідження, не працює самостійно, оскільки науково-абстрактне мислення дослідника, уподобання та розпізнавання характерних мелодичних елементів поки що не можуть бути замінені комп'ютерною обробкою даних.

Найважливішим предметом дискусії в попередніх дослідженнях була специфікація набору даних для аналізу. Очевидно, що порівнювати можна лише однаково оброблені музичні елементи, тому серед найважливіших завдань - визначити стандартний формат та інформаційну щільність вхідних даних.

Як перший крок - потрібно вирішити цифрове перетворення нотного рукопису. Міжнародні дослідження здебільшого дали результати в розпізнаванні нотних знаків, деякі з яких можна використати у проєкті, однак ще необхідно чимало нових розробок.

Прикладами вирішальних для аналізу елементів є розпізнавання артикуляції мелодії, ідентифікація рядків, декодування абревіатур і визначення сфери аналізу.

Після вивчення тисяч транскрипцій рукописів було виявлено наступні типові фактори, що ускладнюють декодування: 1) різні види спрощення, які набули поширення через необхідність економії паперу та прискорення роботи; 2) кілька даних мелодії в одному зображенні; 3) визначення сфери, що містить корисну інформацію.

Остаточна мета дослідження полягає в тому, щоби точно визначити роль дослідника в обробці музичних даних. Отже, ставлення дослідників, які відмовляються від підтримки програмного забезпечення, може змінитися на користь фактичного використання пропонованої цифрової системи. Уперше в історії досліджень угорської народної музики можна створити детальне та задокументоване цифрове дослідницьке середовище з інтеграцією корисних відповідних програмних засобів.

Визначаються проблеми введення даних і стандартний формат музичних даних, придатних для масового запровадження та аналізу. Заміна раніше широко використовуваних опціональних даних розширеними даними дасть змогу отримати ширший діапазон параметризації та параметрів пошуку, а їх вільне поєднання оптимізує вивчення нових наукових моделей. За допомогою середовища цифрових досліджень можна більш точно визначити діапазон застосування попередньої наукової музичної класифікації, а обробку, як і класифікацію мелодій, можна значно прискорити.

Ключові слова: цифрове дослідницьке середовище (ЦДС) з підтримкою штучного інтелекту, оптичне розпізнавання музики (ОРМ), музичні рукописи, угорські народні пісні, наукова музична класифікація, етномузикологія, цифрові архіви, база даних фольклору.



Appendix

The article (project no. 137969; AI-supported Digital Research Environment for the Classification of Folk Tunes) has been implemented with the support provided by the Ministry of Innovation and Technology of Hungary from the National Research, Development and Innovation Fund, financed under the PD21 funding scheme.

Explanation of abbreviations necessary for the identification of manuscripts published in the Appendix:

AP - Abbreviation for Akademiai Pyral (Academical Pyral), an identifier of the custom-made records in the Folk Music Archive of the Institute for Musicology.

The abbreviation refers to the rights holder of the collection (Hungarian Academy of Sciences) and the French company Pyral, which provided the materials and technology.

According to archival practice, when collecting folk music or folk dance, technicians copied music originally recorded on tape to special lacquer discs.

Researchers used these discs, known as `torture copies', to note down the recordings and usually cited the AP number as the source in their publications.

BR - prefix to the inventory number of the Bartok System

D - Webster wire (magnetic sound carrier)

Mg - magnetic tape (magnetic sound carrier)

MH - phonograph cylinder owned by the Museum of Ethnography (Budapest)

ZTI - Institute for Musicology (Budapest) https://library.hungaricana.hu/hu/view/ZTI_AP_04301 -04330/?pg=400&layout=s



Fig. 3. Simplifications 1: For a variant, only the deviation is indicated Archive ID: ZTI_AP_01011a; sound recording: ZTI_D_215-016 https://library.hungaricana.hu/hu/view/ZTI_AP_01011-01036/?pg=0&layout=s

Fig. 4. Simplifications 2: Marking of repetitive parts Archive ID: ZTI_AP_08663f; sound recording: ZTI_Mg_02806 https://Hbrary.hungaricana.hu/hu/view/ZTI_AP_08649-08680/?pg=206&layout=s



Fig. 5. More data in one image Archive ID: ZTI_AP_04329e-f; sound recording: ZTI_Mg_01276A

https://library.hungaricana.hu/hu/view/ZTI_AP_04301-04330/?pg=400&layout=s



Fig. 6. An original manuscript before processing Archive ID: BR_03943; sound recording: MH_24l8a http://systems.zti.hu/br/hu/search/3969?cod=24l8a

Fig. 7. Preparatory analysis for machine processing (After the grayscale conversion, the AI algorithm is helped by using differentcolor markings).



Fig. 8. A cautionary example of Bartok's legendary green ink corrections Archive ID: BR_01194; sound recording: MH_1502a http://systems.zti.hu/br/hu/search/1194?cod=1502a (In the original manuscript, the darker-toned corrections are green.)

Размещено на Allbest.ru

...