Many datasets describing musical resources are published today as Linked Data, mainly focusing on metadata, notation, and audio features. However, the availability of musical data as Linked Data is often not enough for musicians, who need additional layers of software and queries to accomplish their workflows. Mashups are compositions created by blending two or more pre-recorded songs, and it has been shown that they can be generated by using SPARQL on top of MIDI music represented as Linked Data. In this demonstration, we showcase SPARQL-DJ, a web-based application implementing those layers for the workflow of generating MIDI mashups using Linked Data and SPARQL. The demonstration focuses on three parts: (1) coverage of the mashup composition workflow; (2) scoping of input retrieval for beatmaching by broadening text search and leveraging track annotations; and (3) management of off-beat and dissonance.
With well-known linked datasets like MusicBrainz, BBC Music and MySpace,
music is an important player in the Linked Open Data cloud. Most of these
datasets represent the metadata of music, like bands, musicians, song titles,
albums, firms, etc. Some works in the Semantic Web have investigated formal
ways of dealing with music itself, like chord representation [
More recently, proposals have been made to represent music notation as
Linked Data, such as the MIDI Linked Data cloud, a dataset that connects
308,443 MIDI songs by representing their tracks, events, and notes as 10,215,557,355
RDF triples1 [
This wealth of information could be a great aid to musicians in supporting
many of their creative workflows. Concretely, the activity of remixing existing
music to make new, genuine compositions, also known as mashups, is not
currently supported by any semantically rich knowledge base. AutoMashUpper, for
example, is an “interactive system for creating music mashups by automatically
selecting and mixing multiple songs together” based on their harmonic
similarity [
Mashups with MIDI Linked Data
Although the creation of mashups involves a more intricate process [
Workflow coverage. Here, the demonstration will cover the steps of Figure 1 by completing the forms in the interface of SPARQL-DJ (Figure 2). First, the user introduces a search string for the first MIDI, getting a number of results that can be inspected by metadata (hovering their URIs), or by playing their MIDI conversions in the browser. After this, users select the first MIDI they wish to mashup. Selecting the second MIDI is analogous, but this time users can search only metric-matching candidates (same tempo values ), or supply their own metric values. Finally, the user selects which tracks from both MIDIs she wants to have in the mashup. Clicking play will create the MIDI mashup and send it to the browser, while clicking download will retrieve the MIDI or RDF/Turtle serialization. This will be demoed with Michele (Beatles) + Smells Like Teen Spirit (Nirvana) and Eine Kleine Nacthmusik (Mozart) + War Pigs (Black Sabbath).
Scoping of mashable candidates. We will showcase how the search for beatmatching songs can be broadened in two different ways. First, we will use the additional metadata available in the MIDIs to use the bands’ names, their genre, or the collection they come from. Second, we will narrow the number of tracks in the final mashup by filtering them based on their track name annotations.
Managing off-beat and dissonance. In this part, we will show cases in which off-beat and harmonic dissonance might occur, even if the mashable songs are metric compatible; and will discuss about how to manage these.
2See queries at https://github.com/rickmeerwaldt/SPARQLmashup and https://github. com/rickmeerwaldt/SPARQLmashup/raw/master/example/example.pdf
3http://virtuoso-midi.amp.ops.labs.vu.nl/sparql (a) MIDI lookup, play, and metric filtering.
(b) Selection of tracks for the mahsup.
In the future, we will extend SPARQL-DJ’s coverage of the mashup creation workflows, we will leverage external links to other datasets for scoping the search of mashable candidates, and we will use harmonic features to manage dissonance. Acknowledgements. We would like to express our gratitude to Ingrid Arcas for her dedication at testing SPARQL-DJ, her contributed mashups, and her valuable design advice.