# Melody Generation using an Interactive Evolutionary Algorithm

**Authors:** Majid Farzaneh, Rahil Mahdian Toroghi  
**arXiv:** [1907.04258](https://arxiv.org/abs/1907.04258)  
**Date:** 2019-07-07  
**Summary page:** https://www.summarizepaper.com/en/paper/1907.04258v1/

## Key points

- Growing interest in using computers to generate music in the field of artificial intelligence
- Deep learning techniques are effective for generating sequences of music
- Evaluating the quality of machine-generated music is a major challenge
- Proposed methodology uses interactive evolutionary optimization with human expertise in scoring generated melodies
- Bi-LSTM recurrent neural network used to model music quality scoring
- Innovative approach to melody generation using a Genetic algorithm introduced
- Melodies generated through the Genetic algorithm are evaluated using the Bi-LSTM network
- Results demonstrate capability of creating pleasurable melodies with desired styles and pieces
- Methodology is efficient compared to existing data-oriented evolutionary systems, producing high-quality melodies at a faster pace
- Research contributes significantly by combining interactive evolutionary optimization with deep learning techniques and incorporating human expertise into the evaluation process

## Summary

In the field of artificial intelligence, there has been a growing interest in using computers to generate music. Deep learning techniques have emerged as effective methods for generating sequences of music. However, one major challenge is how to evaluate the quality of music generated by machines. In this research paper titled "Melody Generation using an Interactive Evolutionary Algorithm," authors Majid Farzaneh and Rahil Mahdian Toroghi propose a methodology that addresses this problem. The proposed methodology is based on an interactive evolutionary optimization method, where human expertise plays a crucial role in scoring the generated melodies during the training process. To model the music quality scoring, the authors utilize a Bi-LSTM recurrent neural network. This network is trained to evaluate the generated melodies based on their musical qualities. Additionally, the authors introduce an innovative approach to melody generation using a Genetic algorithm. The melodies generated through this algorithm are then evaluated using the Bi-LSTM network developed earlier. The results of this evaluation demonstrate that the proposed method is capable of creating pleasurable melodies with desired styles and pieces. One notable advantage of this methodology is its efficiency compared to existing data-oriented evolutionary systems as it not only produces high-quality melodies but also does so at a faster pace than state-of-the-art systems. Overall, this research contributes significantly to the field of music generation by providing a novel approach that combines interactive evolutionary optimization with deep learning techniques and incorporates human expertise into the evaluation process while leveraging advanced neural networks which enables creation of enjoyable and stylistically accurate melodies.