Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Conserving Human Creativity with Evolutionary Generative Algorithms: A Case Study in Music Generation

Justin Kilb, Caroline Ellis

TL;DR

The paper tackles preserving human creativity in the era of scalable generative models, arguing that data-driven generation risks eroding originality. It proposes a human-in-the-loop approach using Differential Evolution (DE) on real-valued melody vectors, guided by single-user feedback, and demonstrates this with six songs that earned signing opportunities. A mutational rule $v_i^{G+1} = x_{r1}^G + F \cdot (x_{r2}^G - x_{r3}^G)$ illustrates the DE mechanism, highlighting how directional search can capture a creator's preferences beyond random exploration. The work suggests a practical, human-centric path for computer-assisted music production, with potential integration into DAWs and cross-domain applicability, providing a foundation for culturally relevant, personalized artistic expression as generative techniques scale.

Abstract

This study explores the application of evolutionary generative algorithms in music production to preserve and enhance human creativity. By integrating human feedback into Differential Evolution algorithms, we produced six songs that were submitted to international record labels, all of which received contract offers. In addition to testing the commercial viability of these methods, this paper examines the long-term implications of content generation using traditional machine learning methods compared with evolutionary algorithms. Specifically, as current generative techniques continue to scale, the potential for computer-generated content to outpace human creation becomes likely. This trend poses a risk of exhausting the pool of human-created training data, potentially forcing generative machine learning models to increasingly depend on their random input functions for generating novel content. In contrast to a future of content generation guided by aimless random functions, our approach allows for individualized creative exploration, ensuring that computer-assisted content generation methods are human-centric and culturally relevant through time.

Conserving Human Creativity with Evolutionary Generative Algorithms: A Case Study in Music Generation

TL;DR

The paper tackles preserving human creativity in the era of scalable generative models, arguing that data-driven generation risks eroding originality. It proposes a human-in-the-loop approach using Differential Evolution (DE) on real-valued melody vectors, guided by single-user feedback, and demonstrates this with six songs that earned signing opportunities. A mutational rule illustrates the DE mechanism, highlighting how directional search can capture a creator's preferences beyond random exploration. The work suggests a practical, human-centric path for computer-assisted music production, with potential integration into DAWs and cross-domain applicability, providing a foundation for culturally relevant, personalized artistic expression as generative techniques scale.

Abstract

This study explores the application of evolutionary generative algorithms in music production to preserve and enhance human creativity. By integrating human feedback into Differential Evolution algorithms, we produced six songs that were submitted to international record labels, all of which received contract offers. In addition to testing the commercial viability of these methods, this paper examines the long-term implications of content generation using traditional machine learning methods compared with evolutionary algorithms. Specifically, as current generative techniques continue to scale, the potential for computer-generated content to outpace human creation becomes likely. This trend poses a risk of exhausting the pool of human-created training data, potentially forcing generative machine learning models to increasingly depend on their random input functions for generating novel content. In contrast to a future of content generation guided by aimless random functions, our approach allows for individualized creative exploration, ensuring that computer-assisted content generation methods are human-centric and culturally relevant through time.
Paper Structure (14 sections, 4 equations, 3 figures)

This paper contains 14 sections, 4 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Creativity fueled by mutation
  3. Human-guided mutation
  4. Methods
  5. Representation of melody with numbers
  6. Differential Evolution Algorithm
  7. Initialization
  8. Mutation
  9. Crossover
  10. Selection
  11. Potential advantages of Differential Evolution
  12. Single user training
  13. Results
  14. Discussion

Figures (3)

  • Figure 1: Left: A noise vector creates a distribution of generated content that deviates away from training dataset replication. Right: Increasing the maximum allowable deviation by increasing the bounds of the random distribution in the noise vector.
  • Figure 2: Left: Mutation of new content is bounded by the limits of the random function in a noise vector. Right: Mutation is guided by a human's score.
  • Figure 3: Efficient mutation, guided by human feedback, explores outside the bounds of a random distribution.