- Year: 2015–present
- Type: Software project
Overview
Isomer is AI software I developed to help composers discover hidden patterns in historical music and use those patterns as creative starting points for new compositions.
Unlike generative AI systems designed to replace human creativity, Isomer is designed to augment human creativity — to help composers see possibilities they might not have noticed on their own.
The Core Question
After years of building systems that teach computers to understand music (see Clio), I became interested in a different question:
Could machines help humans create better music?
Not by generating music automatically, but by revealing patterns and relationships that human composers could then use as creative material.
Example Use Case
A composer working on a new piece might ask Isomer: “What mood similarities do you find in these musical clips?” Isomer could analyze the complete batch of sound recordings/scores and return:
- Every instance of similar melodic/timbral patterns
- The harmonic context around each instance
- Variations and transformations of the pattern
- Related patterns that share distant (or opposing) characteristics
The composer can then study these examples, extract principles, and apply them in their own work. In the future, the user will be able to ask Isomer to generate new material based on the analysis.
Philosophy
Isomer is based on the idea that creativity is pattern recognition and transformation.
Great composers don’t invent entirely new musical languages from scratch. They study historical music, internalize its patterns, and then transform those patterns in novel ways.
Isomer accelerates this process by making pattern discovery faster and more systematic — but the creative decisions remain entirely in the hands of the human composer.
How It Works
Isomer analyzes large collections of music (MIDI files, scores, recordings) to discover:
- Harmonic patterns: tension created by pitch/timbral combinations (momentary/trends)
- Melodic motifs: recurring intervallic patterns, contour shapes
- Rhythmic structures: perceptual pulse patterns, agogic accents
- Formal relationships: sectional proportions, developmental trends
The system then presents these patterns to the composer as creative prompts — not as finished music, but as raw material for further exploration.
Technical Approach
- Machine learning for pattern extraction from MIDI and audio
- Graph-based analysis to map relationships between musical ideas
- Constraint satisfaction to generate variations that preserve core characteristics
- Interactive visualization to help composers explore discovered patterns
Built with Java, C++, and open-source music analysis libraries.
Current Status
Isomer is in active development. I use it in my own compositional work and am exploring partnerships with music schools and conservatories for educational applications.
Related
- Clio Music Analysis — teaching computers to understand musical structure
- What Was the Question? — album created using Isomer