Introduction to Symbolic Music Processing with Partitura (Tutorial at ISMIR 2022)

This tutorial is an introduction to symbolic music processing using Partitura by Carlos Cancino-Chacón, Emmanouil Karystinaios, Silvan David Peter and Francesco Foscarin.

This tutorial was recorded during ISMIR 2022 in Bengaluru, India on Dec. 4th, 2022.

Chapters

1. Introduction to Symbolic Music Processing (Francesco Foscarin) 00:00
2. Introduction to Partitura (Silvan David Peter) 29:45
3. Symbolic Music Alignment (Carlos Cancino-Chacón) 01:34:30
4. Pitch Spelling (Emmanouil Karystinaios): 02:15:48
5. Drum Generation with Transformers (Emmanouil Karystinaios) 02:41:00

Abstract

Symbolic music formats (e.g., MIDI, MusicXML/MEI) can provide a variety of high-level musical information like note pitch and duration, key/time signature, beat/downbeat position, etc. Such data can be used as both input/training data and as ground truth for MIR systems. Here you can find an introduction to symbolic music, with a presentation of the typical symbolic data types.

This tutorial aims to provide an introduction to symbolic music processing for a broad MIR audience, with a particular focus on showing how to extract relevant MIR features from symbolic musical formats in a fast, intuitive, and scalable way. We do this with the aid of the Python package Partitura. To target different kinds of symbolic data, we use an extended version of the ASAP Dataset, a multi-modal dataset that contains MusicXML scores, MIDI performances, audio performances, and score-to-performance alignments.

This tutorial is an ensemble of notebooks that will guide you through the basics of the Partitura library. The tutorial is divided into four parts:

1. An introduction to the Partitura library with all basic I/O functionality and the basic data structures.

2. A tutorial on how to use the Partitura library to perform automatic alignment between performances and their respecitve scores.

3. How to implement a Pitch spelling model using Partitura.

4. How to create a Transformer Based Beat Generator using Partitura.

The motivation behind this tutorial is to promote research on symbolic music processing in the MIR community. Therefore, we target a broad audience of researchers without requiring prior knowledge of this particular area. For the hands-on parts of the tutorial, we presuppose some practical experience with the Python language, but we will provide well-documented step-by-step access to the code in the form of Google Colab notebooks, which will be made publicly available after the tutorial. Furthermore, some familiarity with the basic concepts of statistics and machine learning is useful.

This work receives funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme, grant agreement No 101019375 (Whither Music?).