TUTORIALS

Parallel Tutorial Session 1 (Morning):

Tutorial 1: Million Song Dataset [PDF]
Thierry Bertin-Mahieux, Matthew D. Hoffman, and Daniel P.W. Ellis (Columbia University).
Tutorial 2: Audio Content-based Music Retrieval [PDF]
Meinard Muller (Saarland University and MPI Informatik) and Joan Serra (Universitat Pompeu Fabra).

Parallel Tutorial Session 2 (Afternoon):

Tutorial 3: Practical Linked Data for MIR Researchers [PDF]
David De Roure and Kevin Page (University of Oxford).
Tutorial 4: Musicology [PDF]
Anja Volk and Frans Wiering (Utrecht University).

Abstracts

Tutorial 1: Million Song Dataset

This tutorial introduces the Million Song Dataset, a freely available collection of audio features and metadata for a million contemporary popular music tracks. The dataset was recently released, and the main goals of this tutorial are: 1) explain the content of the dataset, including the additional datasets on cover songs and lyrics, and 2) demonstrate that working with such an amount of data is easier than it looks in terms of code, computational resources, etc. The tutorial will be done by two of the creators of the dataset.

This tutorial is of particular appeal to the researchers working on: music recommendation, music similarity, automatic tagging, cover song recognition, song segmentation, artist identification, lyrics analysis, web data mining, and library science/music management.

Tutorial 2: Audio Content-based Music Retrieval

Even though there is a rapidly growing corpus of available music recordings, there is still a lack of audio contentbased retrieval systems allowing to explore large music collections without manually generated annotations. In this context, the query-by-example paradigm is commonplace: given an audio recording or a fragment of it (used as query or example), the task is to automatically retrieve all documents from a given music collection containing parts or aspects that are similar to it. Here, the notion of similarity used to compare different audio recordings (or fragments) is of crucial importance, and largely depends on the application in mind as well as the user requirements.

In this tutorial, we present and discuss various contentbased retrieval tasks based on the query-by-example paradigm. More specifically, we consider audio identification, audio matching, version (or cover song) identification and category-based retrieval. A first goal of this tutorial is to give an overview of the state-of-the-art techniques used for the various tasks. Furthermore, a second goal is to introduce a taxonomy that allows for a better understanding of the similarities, and the sometimes subtle differences, between such different retrieval scenarios. In particular, we elaborate on the differences between fragment-level and document-level retrieval, as well as on various specificity levels found in the music search and matching process.

Tutorial 3: Practical Linked Data for MIR Researchers

Linked Data is an approach that combines structured semantics with the large-scale distributed architecture proven through the World Wide Web, and is proving to be an approach with great potential that has generated signicantinterest in the MIR and the wider music community (borneout by several papers and demonstrations at previous ISMIR conferences, and music-related submissions at conferences such as ISWC). It is a means by which we can use, publish, enhance, and most importantly link between the growing number of de-centralised information sources in the web of data, so as to develop new MIR systems that are improved by their access to these datasets, and which increase their utility by making results more readily consumable and linkable to the rest of the Semantic Web.

This tutorial takes a whole-system approach with coverage of information representation through all stages of the MIR research cycle, and as such is appropriate for all music information researchers who have an interest in how the SemanticWeb and Linked Data can support and enhance their work. Code examples will be provided in a largely complete form, with exercises focussed on creation and manipulation of the data models rather than any specic programmingor scripting language. This is a new tutorial that will be presented for the rst time at ISMIR 2011; the presenters are aware of the wide-ranging “Music and the Web of Linked Data” tutorial presented at ISMIR 2009, and hope to provide a more focussed and practical application of Semantic Web technologies to the domain. Attendees will leave the tutorial with a clear motivation for introducing Linked Data to their research and the basic skills and familiarity to enable them to do so.

Tutorial 4: Musicology

MIR researchers’ fondest enemies are the musicologists. Apparently sharing a common lust for music, we have to, and at times even want to interact with them, but their strange behaviour never ceases to baffle us. Why do they react so negatively to our cool technology, giving us only ill-defined concepts and crappy ground truths in return? This tutorial proposes an anthropological excursion into the strange territory of musicology, where we will meet the natives and explore their habits and value systems. We will discuss how musicological domain knowledge can be turned into a valuable resource for MIR researchers. We will draw up a number of guidelines that will help MIR researchers to interact successfully with musicologists when you meet them on your own.

This tutorial is aimed at all MIR researchers who are curious about musicology. Researchers who are motivated by a love for music but possess little formal training in music and therefore need to depend on ‘domain knowledge’ in their research will especially benefit from this tutorial. It is helpful but not necessary for participants to have a basic knowledge of elementary music theory. No advanced knowledge is needed and the presenters will aim at maximum accessibility and understandability of the content.