Digital Music Lab - Analysing Big Music Data

Music research, particularly in fields like systematic musicology, ethnomusicology, or music psychology, has developed as "data oriented empirical research", which benefits from computing methods. In ethnomusicology particularly, there has been a recent growing interest in computational musicology and its application to audio data collections. Similarly, the empirical study of performance of Western music, such as timing, dynamics and timbre and their relation to musical structure has a long tradition. However, this music research has so far been limited to relatively small datasets, because of technological and legal limitations.

On the other hand, researchers in Music Information Retrieval (MIR) have started to explore large datasets, particularly in commercial recommendation and playlisting systems (e.g. The Echo Nest, Spotify), but there are differences in the terminologies, methods, and goals between MIR and musicology as well as technological and legal barriers. The proposed Digital Music Lab will support music research by bridging the gap to MIR and enabling access to large music collections and powerful analysis and visualization tools.

The Digital Music Lab project will develop research methods and software infrastructure for exploring and analysing large-scale music collections. A major output of the project will be a service infrastructure with two prototype installations. One installation will enable researchers, musicians and general users to explore, analyse and extract information from music recordings stored in the British Library (BL). Another installation will be hosted by the Centre for Digital Music at Queen Mary University of London and provide facilities to analyse audio collections such as the I Like Music, CHARM and the Isopohnics datasets, creating a data collection of significant size (over 1m pieces).

We will provide researchers with the tools to analyse music audio, scores and metadata. The combination of state-of-the-art music analysis on the audio and the symbolic level with intelligent collection-level analysis methods will allow for exploration and quantitative research on music that has not been not possible at this scale so far.

The results of these analyses will be made available in the form of highly interactive visual interfaces. Musical questions we will explore include: how does performance style change change over time in relation to a particular genre or style, in classical, world, jazz, or popular music; how might performances of a given genre vary by geographical location; how does a performer's individual performance aesthetic develop over their lifetime; how might we identify the influence of one performer on another. Starting points for the analysis will be questions of musical timing and structure in piano music as well as in folk songs. We will also explore more generic musicological questions, such as the role of specific instruments in different cultures using data mining on the collection level, e.g. for relating similarities on the signal and metadata level. The resulting derived data that can be aggregated for research use, and the annotation of audio files with metadata, using all open standards such as the Music Ontology.

The use of the proposed framework will be demonstrated in musicological research on classical music (building on the AHRC-funded CHARM and CMPCP research centres), as well as in folk, world and popular music. All results will be made available as open data/open source software. We feel that this project has the potential to bring together communities from musicology and MIR to mutual benefit.

The beneficiaries of this project include those directly involved, and those involved through our partner the British Library, and from any likely commercialisation of project outcomes. Those directly involved are:
(1) Users of the British Library, in particular of the Digital Music Collections
(2) The British Library, in particular the British Library Labs
(3) Existing and future users of Sonic Visualiser/Annotator (estimated as at least 10,000 currently) who will benefit from the extended capabilities that will be developed by this project.
Those indirectly involved are:
(4) Potential licensees and adopters of the music analysis tools showcased by this project
(5) Potential licensees and adopters of the big data analysis infrastructure and interfaces developed for this project
(6) Customers of licensees and adopters, specifically musicologists and the music listening public

These different constituents benefit in differing ways. Users of the BL will be able to access large data sources, to analyse large datasets, to download analysis results and to have access to interactive visualisations displaying analysis results. The BL will be able to improve their service and infrastructure, and will be able to exploit the large amounts of data, which already exist in the BL Sound Archive. Users of Sonic Visualiser/Annotator will benefit from tools that offer batch analysis of large corpora and provide interactive visualisations of analysis results. Potential licensees of music analysis tools, primarily industrial and public bodies in the music technology and digital libraries sectors, will be in a position to analyse efficiently and robustly large amounts of recordings, while their customers will enjoy a more informative and compelling experience of music appreciation. Finally, licensees of big data analysis technologies will benefit from an existing infrastructure, tested on one of the largest libraries worldwide, while the provided technology will be transferable to other types of data, for example books, images, videos, or metadata.

We expect beneficiaries (1) to (2) to gain significant benefit during the lifetime of this project, with this increasing as the main outcomes of the project are disseminated. We foresee benefit to beneficiaries (3) during and after the project. Finally, beneficiaries (4), (5) and (6) should see benefit after the end of the project.

The project includes regular interaction between researchers, partners, and users, with particular emphasis on a training workshop for the developed tools, which will take place towards the end of the project. We will also develop documentation for the project, with the objective of making it available for use by the greater public, and we will create and manage a project website and repository. In order to inform and engage users, press activities will be organised, including presence in university Open Days, blogs and social media. Mechanisms to present this project to the public will be sought in conjunction with Press and Publications Office of City, Queen Mary, and UCL. We expect also to present the work at the London-based Music Tech Festival, which attracts potential beneficiaries from the creative industry.