TL;DR
This paper introduces the Music Part Ontology (MPO), a novel semantic framework for formalizing and analyzing the structural attributes of symbolic music. By representing music tracks as graphs of parts and chord progressions, the MPO enables more sophisticated music information retrieval tasks, such as genre classification, by leveraging rich structural and sequential musical data.
What Problem Does This Paper Address?
The vast amount of user-generated symbolic music data available online (e.g., guitar tabs) is often unstructured and difficult for machines to understand. While existing ontologies address aspects like music production or high-level theoretical concepts, there's a gap in formalizing the sequential structure of music tracks, specifically how different parts (like verses and choruses) and chord progressions are arranged. This lack of formalization hinders advanced Music Information Retrieval (MIR) tasks that could benefit from a deep understanding of a song's architecture.
Methods
The authors developed the Music Part Ontology (MPO) using Description Logics and Semantic Web technologies. The MPO defines core classes like Track, Part, ChordProgression, and Chord, connected by properties such as hasPart, hasChordProgression, and hasNextPart to capture sequential relationships. They also introduced transitive properties like isFollowedbyX for inferring relationships. The MPO is designed to be interlinked with existing ontologies like the Functional Harmony Ontology (FHO) and the Music Theory Ontology (MTO). For demonstration, they converted user-generated symbolic music data into RDF graphs, loaded them into a semantic repository, and queried them using SPARQL.
Datasets
The authors gathered user-generated symbolic music data from websites like ultimate-guitar.com. They curated this data by standardizing chord names and part labels. Additionally, they retrieved music genre information using the Spotify Web API. The final dataset used for demonstration comprised 250 music tracks across five genres: Rock, Pop, Jazz, Blues, and Reggae.
Key Insights and Ideas
- Structure is Key: The paper highlights that the sequential and structural attributes of a music track are crucial for enhancing MIR tasks like genre classification.
- Bridging the Gap: The MPO effectively bridges the gap between raw, user-generated symbolic music data and machine-readable knowledge by formalizing song structure.
- Interoperability: The MPO's design allows for seamless integration with other music-related ontologies, creating a richer knowledge graph.
- Beyond Simple Features: By capturing structural elements, the MPO enables queries that go beyond simple audio features or basic metadata, allowing for deeper musical analysis.
Key Results
The paper demonstrates the practical application of the MPO through SPARQL queries. For instance, they successfully retrieved all songs containing a "ii-V-I" progression (common in Jazz) and analyzed its distribution across different genres. Table 1 shows that Jazz has the highest percentage (2.3%) of ii-V-I progressions among the analyzed genres. Table 2 shows the distribution of Chorus sections across genres, with Reggae having the highest percentage (33.3%). These results suggest that structural and sequential attributes can provide valuable insights for genre classification and other MIR tasks.
Limitations and Future Work
The authors acknowledge that defining musical parts can be challenging due to variations across genres and cultures. While they defined eight common parts as subclasses of Part, this categorization might not cover all nuances. Future work includes extending the MPO with more concepts related to music structure and exploring advanced reasoning techniques to identify and define common chord progressions.
Summary updated