TLDR: This research introduces a novel rule-based method for generating musical variations by parsing existing tunes into a ‘Pathway Assembly’ grammar. By applying 19 types of mutations directly to this grammar, rather than the notes, the system creates new tunes that maintain structural coherence. The study, focused on Irish traditional music, analyzes how tunes evolve over multiple mutations using edit distance and discusses the musical and ethical implications of this generative approach.
Researchers have developed a new rule-based method for generating music by creating variations of existing tunes. This innovative approach focuses on understanding and manipulating the underlying structure of a piece of music, rather than making random changes to the notes themselves. The goal is to produce new tunes that are related to the original but offer meaningful and coherent variations.
The core of this method involves parsing an existing tune to identify its ‘Pathway Assembly’ (PA). Think of PA as a blueprint that reveals all the repetitions and structural elements within a tune. To achieve this, the system utilizes the Sequitur algorithm, which helps in discovering hierarchical structures in sequences. The output of this parsing process is a ‘grammar’ – essentially a set of rules that define the tune’s structure.
Instead of directly altering the musical notes, the system applies ‘mutations’ to this grammar. This is a crucial distinction, as mutating the grammar ensures that changes are coherent across the tune, preserving its overall musical patterns. If changes were made directly to the notes without considering structure, the result would likely be random and unmusical. The research paper details 19 different types of mutations that can be applied to these grammars, including adding, removing, swapping, or reversing parts of the grammar. These mutations are applied randomly to automatically manipulate the tune’s underlying structure.
After a mutation is applied to the grammar, the system expands the modified grammar to generate a new tune. This new tune is a variation of the original, and the process can be repeated multiple times to observe how tunes evolve over successive mutations. The study uses metrics like ‘edit distance’ (also known as Levenshtein distance) to quantify how much a tune changes on the surface after mutations. This helps in understanding the impact of different mutation types.
The research specifically focused on Irish traditional tunes, representing them as lists of pitch values. Interestingly, using absolute pitch values for representation yielded shorter and more effective grammars compared to using intervals between notes. The study found that certain mutations, particularly one that defines a new rule based on existing rules and numbers, had a more significant impact on the resulting tune’s edit distance.
While the system can create fascinating variations, it currently focuses solely on pitch sequences and does not account for rhythm, time signature, or harmony. This means that the generated tunes might have altered lengths or unaligned notes compared to the original. Future work aims to incorporate these musical features to allow for more controlled and musically intuitive variations, such as ensuring proper cadences at the end of phrases or tunes.
Also Read:
- MIDI-VALLE: Advancing Expressive Piano Performance Synthesis with Neural Codec Language Models
- Unmasking AI Music: How Audio Augmentations Challenge Deepfake Detection Models
The authors also addressed ethical considerations, particularly in the context of traditional music. Unlike some AI music tools that synthesize entirely new tunes from a corpus, this system generates variations of a single, specific existing tune. This approach ensures that the original tune and its potential author are acknowledged, respecting the provenance important in oral traditions. The goal is not to create tunes that ‘fool’ listeners into believing they are human-composed, but rather to explore novel artistic effects by intentionally going beyond the normal parameters of the tradition. You can read the full research paper here.
