SCIENCE@home

Bach Genre Music Generation with WaveNet—A Steerable CNN-based Method with Different Temperature Parameters

Shang-Bao Luo

Published 2022 in Proceedings of the 4th International Conference on Intelligent Science and Technology

ABSTRACT

Bach's music has been considered as the bible of polyphonic music study and a major corpus of AI research on polyphonic patterns. The application of recurrent neural networks to music generation has been studied for decades, but it suffers from costly training and fails to process long-time samples. Moreover, the recent popularity of GAN-based models and Transformers are also less able to capture local information on a specific task than RNN or CNN due to their large-scale pre-training. This paper proposes a CNN-based WaveNet for generating Bach-style music that demonstrates it can achieve high performance with more concise architecture and faster training process. In the Turing test, this method produces impressive results, which are capable of producing high-quality imitated Bach music on the Bach chorales dataset. The author also introduces a temperature parameter to control the creativity of the music generated by the model. When increasing the value of temperature parameters, the author observed that generated music would gradually deviate from the original style and appear creative short musical melodies and motives.

PUBLICATION RECORD

  • Publication year

    2022

  • Venue

    Proceedings of the 4th International Conference on Intelligent Science and Technology

  • Publication date

    2022-08-10

  • Fields of study

    Not labeled

  • Identifiers
    DOI 10.1145/3568923.3568930
  • External record

    Open on Semantic Scholar

  • Source metadata

    Semantic Scholar

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-20 of 20 references · Page 1 of 1

CITED BY