Z-ENG: Music Generation with Deep Learning Algorithms

2023-2024 tavasz

Nincs megadva

Téma leírása

A breakthrough happened in content generation with the help of deep learning techniques recently. It leads also to societal changes, as machine learning generated content can replace human created content in some use-cases. The most famous technique can be found in the field of natural language processing (NLP) and is related to the GPT models (e.g. ChatGPT, GPT-J). However, visual content generation has also prominent text-to-image generation techniques (e.g. https://arxiv.org/pdf/2301.00704.pdf).

The goal of this project is to investigate music generation algorithms, whether AI algorithms can replace human artists. The question is whether the style of certain composers can be replaced by AI. In order to answer this question, music generation algorithms should be trained to mimic the style of a certain composer. The generated content is then evaluated by experts.

A starter for music generation can be found at: https://arxiv.org/pdf/2011.06801.pdf

The student will work on audio content generation as follows:

  • Understand the current music generation algorithms
  • Hopefully introduce a novel technique
  • Acquire a dataset and train candidate algorithms
  • Generate audio content with the candidate algorithms

Maximális létszám: 10 fő