Deliverable 3.1 Data Resource Preparation

Author
Abstract
This deliverable D3.1 Data Resource Preparation is submitted in the scope of Work Package 3 and it
aims to provide information about the target repertoires that will be used in TROMPA use cases, as
well as how these repertoires will be imported in the TROMPA ecosystem. The elements of the
selected repertoires are connected to existing online repositories, and will be exploited and enriched
in the TROMPA use cases.
At first, we present an overview of the most common Public Domain music repositories
considered in the TROMPA project for the different use cases: what volumes of data are contained,
the corresponding data licence, an overview of the represented musical styles as well as potential
uses for these repositories. Next we provide an overview of the five TROMPA use cases, including
the technical and repertoire requirements for each of them, and detail the contributions expected to
be made to the repositories through our use cases. Finally, we summarize all target repertoires and
we discuss technical details required for further developments in the project.
In this deliverable, we consider the following existing music repositories: IMSLP covers around
124,000 unique works, providing scanned scored images (~405,000) that can be exploited in our use
cases (e.g. for conversion of scanned images to symbolic format); symbolic scores (~37,000) that can
be directly used for the use cases or for training methods on WP3; and audio recordings (~ 47,000) of
performances. It can also serve as a place to deposit TROMPA contributions to the Public Domain.
Similar to the IMSLP, the Choral Public Domain Library (CPDL) is a sheet music archive, in this
case focusing on vocal and choral music in the Public Domain. It contains around 10,000 unique
works, most of which are available as scanned score images in PDF format.
Europeana.eu is the official EU digital platform for cultural heritage, featuring contributions from
more than 3,000 institutions across Europe. It contains ~320,000 music audio recordings, scanned
scores, and music videos.
MuseScore is an online platform that allows the users of the MuseScore software to publish and
1
share their music scores online. MuseScore mostly contains user-provided transcriptions of music
pieces that are not limited to classical.
ECOLM is an electronic corpus of lute music, predominantly notated in tablature. The total size of
the ECOLM repertory available to TROMPA is about 2,000 encoded pieces, which is supplemented by
a further 6,000 whose encodings have been translated from different formats.
Early Music Online (EMO) is a collection of 300 books of printed music from before 1600, which
have been digitised and made available as images by the British Library. They cover almost all genres
and styles of music from the 16th century, including works by all leading composers of the age. The
music is almost exclusively vocal, for one to twelve voices, mostly printed in part-books. Also in the
early music field are two relevant public collections of Spanish music suitable for choral singing: BDH, containing facsimiles of c100 books of early printed vocal music, and TLdV, which comprises about
2,000 encoded scores of similar music.
MusicBrainz is community-based repository that stores information about artists, their recorded
works and the relationships between them. It describes a large variety of commercial music
(including western classical music), consisting of around 1 million artists and 18 million tracks.
AcousticBrainz provides audio descriptors (rhythm, key, genre tags, mood tags) for almost 4
million tracks identified using MusicBrainz identifiers.
Kunst der Fuge consists of around 19,000 MIDI files of western classical music, some of them
published under a Creative Commons License.
The CDR Muziekweb catalogue contains music in all music styles, including popular, jazz, world
music and classical. CDR strives to collect all music released in the Netherlands; this is the only
criterion to be included in the collection. As of 2018, the collection holds over 600,000 CD’s, 300,000
LP’s and 25,000 music-DVD’s. In total, there are 579 music styles in use. For classical music, the
collection has albums in all styles and genres, from all style periods, labels and countries.
The Vienna 4x22 Piano Corpus consists of score-aligned performance and audio recordings of 4
short piece excerpts by Mozart, Schubert, and Chopin, each of which performed by 22 professional
pianists in 1999.
The humdrum data repository is a collection of musical scores in the Humdrum (kern) file format,
containing large parts of the standard repertoire for piano solo, string quartet, and choirs, including
composers such as J. S. Bach, Domenico Scarlatti, Haydn, Mozart, Beethoven, Hummel, Chopin.
Over the past years, the National Library of the Netherlands has worked on digitizing the
newspapers that were published in the Netherlands over the past centuries. The Delpher platform
provides access to dutch newspapers from the years 1618–1995, amounting to over 12 million
scanned newspaper pages. The newspapers from 1618–1876 are considered to not have any
copyright-protected material, and a full download of all full texts in OCR, ALTO and XML format is
available. This information will be useful to increase and improve contextual understanding of how
musical works and persons historically were perceived.
Biblioteca Digital Hispanica contains around 5,000 high resolution pages of Spanish music of all
periods. In particular, it has a major component of 16th-century printed vocal music from a period
when Spanish composers were highly esteemed. It contains much parallel repertory with 2.17
Tomás Luis de Victoria, which gives an opportunity for exploring possibilities for practical TROMPA
linkage between various manifestations of a given work. This collection, privately compiled by
Nancho Álvarez, contains about 2,000 choral works by leading Spanish composers of the 16th
century, Victoria, Morales, Guerrero, Vásquez and others. The music is highly suitable for singing by
amateur choirs, as it is not too difficult; furthermore, its essential simplicity (relative to later music)
makes it ideal for testing TROMPA’s methods for OMR, score annotation and audio alignment
components as well as our interfaces for music scholars and choral singers.
This deliverable also connects existing public domain archives with the different pilots, defining
core repertoire for the pilots to be enriched during the project. Each pilot corresponds to one of five
TROMPA use cases, respectively targeting music scholars, orchestras, instrument players, choir
singers, and music enthusiasts.
In the Music scholars use case, scholars will be able to find connections between music works on
multiple levels: from co-occurrences of melodies, harmonic and rhythmic progressions, to the
large-scale structural similarities of musical works. The score is the main starting point and anchor
for such research. Our aim for this pilot is to provide an interface for the selection and display of musical scores (sheet music) from the TROMPA collections, for annotating them, and for searching
them (by text or by example). The repertoire for the initial version of the music scholars’ pilot will
mainly comprise early music from the 16th century from different resources. As the project
progresses, we shall provide facilities for digital enquiries by making the search API publicly available
and publishing the bulk of data collected in TROMPA as a public linked open data dataset.
In the Orchestras Pilot will aim to digitize all symphonies by Gustav Mahler, a core repertoire of
most orchestras, making them available free of charge. TROMPA will develop crowdsourcing
technology to engage music lovers in encoding (out of copyright) scores available as digitized score
images from IMSLP. Using RCO orchestra members’ and librarians’ expertise, we will develop a tool
to extract good quality instrumental parts from these scores. The technological and technical
requirements of the orchestras use case is twofold. Firstly, scanned score image analysis software in
conjunction with crowd-sourced annotation mechanisms will be deployed in order to encode
Mahler's music scores in digital format (MEI, or MusicXML). The second technical requirement is the
capability of annotating music scores and share annotations. All the digitized score encodings of the
Mahler Symphonies that will be derived from the use case will be deposited in public domain musical
archives. Moreover RCO will offer its most recent annotations of one of the Mahler symphonies for
digitization, as well as other archives such as annotated orchestral scores of Willem Mengelberg and
a big part of the available Mengelberg Concertgebouw recordings.
The Instrument Players Pilot provides musicians engaging in rehearsal or performance with a
“Performance Companion” system capable of characterising performative aspects of their playing.
By alignment of performance recordings and metadata with musical score encodings, the
characteristics can be assessed and compared against those derived from other performances or
reference recordings. Initially, the pilot will focus on pianists performing Beethoven’s piano works
(primarily his Sonatas, Variation works, and Concertos). Performance recordings and
characterizations produced by the Performance Companion will be captured and published,
contributing to the available repertoire of Beethoven recordings. We will work towards producing a
complete set of Beethoven piano work encodings over the course of the project. The technological
components required for this system can be split into two groups: score alignment and performance
characterisation. Score encodings created for the pilot will be made publicly available. Score
segmentations, created manually and automatically as part of the score alignment task, will be
associated with these encodings and published under open licenses. Finally, performance metadata
and recordings (with performer permission) will be made publicly available.
The goal of the Choir Singers Pilot is to assist amateur choir singers during individual
performance and to provide functionality for the choir conductor to create repertoires and to listen
to performances by choir members. Users of the pilot will be able to synthesize existing scores,
sing-along with the synthesized voices, and receive feedback on their performance. The
accompanying voices will be available for music in Spanish, Catalan, Latin, English and German, and
the pilot will then focus on pieces in these languages from the repertoires considered below. A set of
selected pieces is provided for the first iteration of the pilot, and several composers are selected as
representative of the target languages: Tomás Luis de Victoria, Anton Bruckner and Josquin des Prez.
This pilot will be mostly based on the audio processing techniques to be developed in Task 3.3,
where we will research and develop techniques for audio synthesis of choir singing. The pilot will
collect data from users to improve the voice synthesis algorithms. Users should be able to provide
general ratings for the synthesis (e.g. by rating the overall quality of the synthesis) and to make
timestamped annotations for the generated material, i.e., allowing the user to input free text
comments to inform about specific problems (e.g. “this phoneme sounds weird at this point in time in the soprano voice”). Through the use of the pilot, synthesized versions of the scores will be
generated and stored, accessible through public URLs. These synthesized versions will be associated
to the scores. The same applies for recordings of performances by users of the pilot (needed for
providing automatic feedback), although in this case, their addition to the repertoire will depend on
getting appropriate permission from the user.
In the Music Enthusiasts Pilot we will provide interaction mechanisms with musical cultural
heritage content targeted at people that, although lacking formal musical knowledge, are interest in
learning more about music. The main study of this use case is to build a music recommendation
system, focused on classical music, with the possibility of integrating user feedback and annotations
of the content The music enthusiasts use case will be focused on the existing music collection of CDR
Muziekweb. We will focus on classical music repertoire of this library, but we will not limit ourselves
to that. Techniques on higher level semantics extraction, such as emotion classification,will be
adopted and evaluated during the use case (Deliverable 3.2). This use case will contribute new
evaluation and ground truth data on the CDR repository, e.g. opinions/ratings about the outcome of
a recommendation system, emotion tags and other annotations of music pieces.
The target repertoires for these user pilots come from many separate repositories and may have
diverse representation schemas. We will store metadata from each target repository in the WP5
TROMPA Contributor Environment (CE) and we will link representations the same item in different
repositories to each other so that information about the these items can be shared regardless of
where that information comes from. Additionally, we will import metadata from MusicBrainz for all
musical content in the CE, linking this data where possible to the metadata from each target
repository. Where possible, if the metadata does not yet exist in MusicBrainz, we will add it, allowing
us to contribute feely available metadata. The internal data model of the CE is a subtree of the
schema.org ontology schema and all data and metadata items that will be stored in the CE will be
mapped on this schema. The consortium is currently developing guidelines that describe how
metadata from external repositories can be imported into the CE, and how this data can be
interlinked within the CE. These guidelines are in development and will remain in step with project
advances and requirements. The CE is designed to store only the metadata for the repertoires that
are to be used in TROMPA plus any content produced by TROMPA, new or derived from these
repertoires. Any content that the metadata describes (scores, music recordings, the results of
computational algorithms) will remain stored in external locations. As part of this stored metadata,
publically available URL will refer to the location where this content can be obtained from. We
provide a tool to automatically import metadata from MusicBrainz to the CE. This tool is under
development, but it is anticipated to include release, recording, work (movement and overall work),
composer, and performer information if it is present in MusicBrainz.
Year of Publication
2019
Report Number
TR-D3.1-Data Resource Preparation v1
URL
https://trompamusic.eu/deliverables/TR-D3.1-Data_Resource_Preparation_v1.pdf
Short Title
D3.1