Paper Session Chair: Craig Vear
11:00h
Tokenization of MIDI Sequences for Transcription
Authors: Florent Jacquemard, Masahiko Sakai, Yosuke Amagasu
There generally exists no simple one-to-one relationship between the events of a MIDI sequence, such as note-on and note-off messages, and the corresponding music notation elements, such as notes, rests, chords, and ornaments. We propose a method for building a formal correspondence between them through a notion of tokens in an input MIDI event sequence and an effective tokenization approach based on a hierarchical representation of music scores. Our tokenization procedure is integrated with an algorithm for music transcription based on parsing wrt a weighted tree grammar. Its effectiveness is shown in examples.
11:20h
Engraving Oriented Joint Estimation of Pitch Spelling and Local and Global Keys
Authors: Augustin Bouquillard, Florent Jacquemard
We revisit the problems of pitch spelling and tonality guessing with a new algorithm for their joint estimation from a MIDI file including information about the measure boundaries. Our algorithm does not only identify a global key but also local ones all along the analyzed piece. It uses Dynamic Programming techniques to search for an optimal spelling in terms, roughly, of the number of accidental symbols that would be displayed in the engraved score. The evaluation of this number is coupled with an estimation of the global key and some local keys, one for each measure. Each of the three types of information is used for the estimation of the other, in a multi-step procedure. An evaluation conducted on a monophonic and a piano dataset, comprising 216 464 notes in total, shows a high degree of accuracy, both for pitch spelling (99.5% on average on the Bach corpus and 98.2% on the whole dataset) and global key signature estimation (93.0% on average, 95.58% on the piano dataset). Designed originally as a backend tool in a music transcription framework, this method should also be useful in other tasks related to music notation processing.
11:40h
Morton Feldman's โProjections One to Fiveโ โ Exploring a Classical Avant-Garde Notation by Mathematical Remodelling
Authors: Markus Lepper, Baltasar Trancรณn y Widemann
The compositions Projection 1 to Projection 5 by Morton Feldman are an important milestone in the application of graphical notation. The meta-language tscore allows easy construction of a computer model of the original scores. On this model, automated performance, graphical rendering, and different analyses can be applied. The practical implementation work brings up the peculiarities of the original notational meta-model and scores, which, without this effort, are easily overlooked.
12:00h
DJster Revisited โ A Probabilistic Music Generator in the Age of Machine Learning
Author: Georg Hajdu
DJster is a probabilistic generator for musical textures based on Clarence Barlowโs legacy program Autobusk, further developed by Hajdu since 2008. The 2023 revision for Max and Ableton Live includes new features improving the versatility of the application and enabling data exchange between the synchronous and asynchronous incarnations. The synchronous incarnation of DJster can be used to preview a texture to be further developed as a sketch in the asynchronous one. DJster allows the real-time addition and modification of tonal and metric profiles departing from Barlowโs original fixed-input paradigm. This motivated an exploration of metric interpolations by means of self-organizing maps and an extension of Jean-Claude Rissetโs illusion of an ever-accelerating rhythm. Furthermore, the implementation of a novel melodic cohesion parameter allows transitions from a sequence of events to a probabilistic process, the latter being the original modus operandi. Finally, DJster as a style-agnostic music generator can be embedded in machine-learning contexts to make user interaction a more rich and intuitive experience.