Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tune It Up: Music Genre Transfer and Prediction

Fidan Samet, Oguz Bakir, Adnan Fidan

TL;DR

This work adapts CycleGAN for music genre transfer between Jazz and Classical using symbolic MIDI data, and enhances it with auxiliary discriminators and triplet loss to improve content preservation while applying target-genre style. Genre prediction via multiple classifiers identifies an MLP as the most reliable evaluator, achieving up to 87.72% accuracy, which guides model selection alongside subjective testing. The best performing setup—combining auxiliary discriminators with triplet loss—achieves Jazz→Classic transfer accuracy of 69.4% and Classic→Jazz of 39.3%, with subject tests indicating melody retention and plausible perceptual genre shifts. The study demonstrates a feasible pipeline for music genre transfer in the symbolic domain and outlines directions for stronger evaluation and broader genre coverage.

Abstract

Deep generative models have been used in style transfer tasks for images. In this study, we adapt and improve CycleGAN model to perform music style transfer on Jazz and Classic genres. By doing so, we aim to easily generate new songs, cover music to different music genres and reduce the arrangements needed in those processes. We train and use music genre classifier to assess the performance of the transfer models. To that end, we obtain 87.7% accuracy with Multi-layer Perceptron algorithm. To improve our style transfer baseline, we add auxiliary discriminators and triplet loss to our model. According to our experiments, we obtain the best accuracies as 69.4% in Jazz to Classic task and 39.3% in Classic to Jazz task with our developed genre classifier. We also run a subjective experiment and results of it show that the overall performance of our transfer model is good and it manages to conserve melody of inputs on the transferred outputs. Our code is available at https://github.com/ fidansamet/tune-it-up

Tune It Up: Music Genre Transfer and Prediction

TL;DR

This work adapts CycleGAN for music genre transfer between Jazz and Classical using symbolic MIDI data, and enhances it with auxiliary discriminators and triplet loss to improve content preservation while applying target-genre style. Genre prediction via multiple classifiers identifies an MLP as the most reliable evaluator, achieving up to 87.72% accuracy, which guides model selection alongside subjective testing. The best performing setup—combining auxiliary discriminators with triplet loss—achieves Jazz→Classic transfer accuracy of 69.4% and Classic→Jazz of 39.3%, with subject tests indicating melody retention and plausible perceptual genre shifts. The study demonstrates a feasible pipeline for music genre transfer in the symbolic domain and outlines directions for stronger evaluation and broader genre coverage.

Abstract

Deep generative models have been used in style transfer tasks for images. In this study, we adapt and improve CycleGAN model to perform music style transfer on Jazz and Classic genres. By doing so, we aim to easily generate new songs, cover music to different music genres and reduce the arrangements needed in those processes. We train and use music genre classifier to assess the performance of the transfer models. To that end, we obtain 87.7% accuracy with Multi-layer Perceptron algorithm. To improve our style transfer baseline, we add auxiliary discriminators and triplet loss to our model. According to our experiments, we obtain the best accuracies as 69.4% in Jazz to Classic task and 39.3% in Classic to Jazz task with our developed genre classifier. We also run a subjective experiment and results of it show that the overall performance of our transfer model is good and it manages to conserve melody of inputs on the transferred outputs. Our code is available at https://github.com/ fidansamet/tune-it-up

Paper Structure

This paper contains 18 sections, 10 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. The Approach
  4. Experimental Settings
  5. Dataset
  6. Music Genre Prediction
  7. Naive Bayes Algorithm
  8. K-Nearest Neighbors Algorithm
  9. Random Forest Algorithm
  10. Multi-layer Perceptron Algorithm
  11. Experimental Results
  12. Generator Networks
  13. Auxiliary Discriminators
  14. Triplet Loss
  15. Auxiliary Discriminators & Triplet Loss
  16. ...and 3 more sections

Figures (7)

  • Figure 1: The model of CycleGAN. The sub-figure a shows cycles, sub-figures b and c show the detailed views of cycles.
  • Figure 2: The architecture of CycleGAN
  • Figure 3: Our proposed auxiliary discriminator added model architecture. $D_{A,aux}$ represents the auxiliary discriminator for domain A to domain B transfer, which distinguishes transferred sample to domain B and a random sample from domain A and domain B. $D_{B,aux}$ follows the same steps for domain B to domain A transfer.
  • Figure 4: Number of samples for each music genre of the dataset proposed in brunner2018symbolic
  • Figure 5: Test accuracies according to different k values in k-NN algorithm
  • ...and 2 more figures