Tutorial Schedule (October 27, Monday)

The morning tutorials are:

Tutorial 1: "Why is Greek music interesting? Towards an ethics of MIR" by Andre Holzapfel and George Tzanetakis

Location: Chang Chin Room (10F)
Slides: Part1, Part2

The initial goal of this tutorial is to provide an overview of musical styles in Greek culture, and to indicate various features of these musics that make them challenging and interesting for research in Music Information Retrieval (MIR). This tutorial is addressed to everybody interested in extending the diversity of her/his evaluation data, this way targeting generality of MIR approaches. On the other hand, the tutorial is aimed to provide a lively overview over a range of styles, that we hope will be informative and inspiring for any music listener. The tutorial will initially provide an overview of various styles of rural and urban music styles in the various areas of Greece. Then, we will focus on some styles we are particularly familiar with, and point out a variety of research tasks that is apparently quite challenging for those musics, such as beat tracking, mood estimation, transcription and chord estimation. In conclusion, inspired by the diversity of Greek music and the problems such diversity poses for our research, we will reflect on the possibility of universal approaches to music processing, and discuss ethical implications for our work on recommendation systems for the musics of the world.

Tutorial 2:"Musical structure analysis" by Meinard Mueller and Jordan Smith

Location: Song Bo Room (10F)
Slides: Part0, Part1, Part2
Handouts: Part0, Part1, Part2
Speaker information: link1, link2

One of the attributes distinguishing music from other sound sources is the hierarchical structure in which music is organized. On the lowest level, one may have sound events such as individual notes, which are characterized by the way they sound, their timbre, pitch and duration. Such sound events combine to form larger structures such as motives, phrases, and chords, and these elements again form larger constructs that determine the overall layout of the composition. This higher structural level is specified in terms of musical parts and their mutual relations. For example, in popular music such parts can be the intro, chorus, and verse sections of the song. Or, in classical music, it can be the exposition, development, and recapitulation of a sonata movement. The goal of music structure analysis is to divide a given music representation into temporal segments that correspond to musical parts and to group these segments into musically meaningful categories.

In this tutorial, we review the most important segmentation and structure analysis principles and then discuss state-of-the-art techniques—many published in just the last few years—that exploit specific characteristics of music. The goals of this tutorial are: first, to explicitly discuss the simplifying model assumptions that each computational procedure is based on; second, to present recent research directions within music structure analysis and to show how the various principles can be applied and combined; and third, to discuss problems involving the evaluation of automated procedures and the use of so-called "ground-truth" reference annotations.

The afternoon tutorials are:

Tutorial 3: "Jingju music: concepts and computational tools for its analysis" by Rafael Caro Repetto, Ajay Srinivasamurthy, Sankalp Gulati and Xavier Serra

Location: Chang Chin Room (10F)
Slides: http://compmusic.upf.edu/node/251

Jingju (also known as Peking or Beijing opera) is one of the most representative genres of Chinese traditional music. From an MIR perspective, jingju music offers interesting research topics that challenge current MIR tools. The singing/acting characters in jingju are classified into predefined role-type categories with characteristic singing styles. Their singing is accompanied by a small instrumental ensemble, within which a high pitched fiddle, the jinghu, is the most prominent instrument within the characteristic heterophonic texture. The melodic conventions that form jingju modal systems, known as shengqiang, and the percussion patterns that signal important structural points in the performance offer interesting research questions. Also the overall rhythmic organization into pre-defined metrical patterns known as banshi makes tempo tracking and rhythmic analysis a challenging problem. Being Chinese a tonal language, the intelligibility of the text would require the expression of tonal categories in the melody, what offers an appealing scenario for the research of lyrics-melody relationship. The role of the performer as a core agent of the music creativity gives jingju music a notable space for improvisation. The lyrics and scores cannot be taken as authoritative sources, but as transcriptions of particular performances.

In this tutorial we will give an overview of Jingju music, of the relevant problems that can be studied from an MIR perspective and of the use of specific computational tools for its analysis. The tutorial will be organized in three parts. The first will be an introduction to Jingju from a musicological perspective, the second will cover diverse audio analysis tools of relevance to the study of Jingju (using http://essentia.upf.edu), and finally in the last part we will present and discuss specific examples of analyzing Jingju arias using those tools (work done in the context of http://compmusic.upf.edu).

Tutorial 4: "MiningSuite, a comprehensive framework for music analysis, articulating audio (MIRtoolbox 2.0) and symbolic approaches" by Olivier Lartillot

Location: Song Bo Room (10F)
Slides: https://miningsuite.googlecode.com/svn/trunk/ismir2014tutorial.pdf

This tutorial presents an in-depth introduction to MiningSuite, a continuation of MIRtoolbox, an innovative environment featuring a large range of audio and music analysis tools. Thanks to an adaptive syntactic layer on top of Matlab, complex design of audio or music analysis operations can be written in a very concise way through a simple assemblage of operators featuring a large set of options. The integration of expertise developed in separate areas of study into common modules encourages further reuse of these individual methods and their intermingling into a common framework. The MiningSuite features an innovative and integrative set of symbolic-based musicological tools related to, among others, segmentation in the form of hierarchical grouping, melodic reduction and modal analysis. An innovative method for exhaustive pattern mining allows detailed motivic and metrical analyses. Audio and symbolic representations (in MIDI and score-like formats) and processes are tightly interconnected: Operators dedicated to high-level musical features extraction (tonal, metrical, structural analyses) integrate signal processing, statistical and symbolic-based methods, and accept both symbolic and audio input.

The tutorial, suitable for both novices and experts, will give an overview of these different audio and symbolic approaches available in the framework, and will explain how to take benefit of the capabilities of the environment via the user-friendly syntax. At the last part of the tutorial, we will dwell a little into the description of the architecture of the MiningSuite (significantly different from the previous MIRtoolbox project) and of the core classes that govern the general capabilities of the framework. Will be described for instance the rich format of the output results, or a syntactic layer within the operators’ Matlab code that simplifies and clarifies the code while taking care of the matrix optimisations in the background. We will explain how you can write new modules, and will present the open-source collaborative platform hosting the MiningSuite project, with versioning control, integrated source code browsing and code review, issue tracker and user’s manual available in a wiki environment.