Music generation using deep learning
Transformers are being used to create music because of their capacity to record long-term dependencies and produce music of a high caliber. Yet, the absence of chord progression tokens is a drawback of employing transformers for music creation. Especially in genres like jazz and pop, chord progressi...
Saved in:
| 主要作者: | |
|---|---|
| 其他作者: | |
| 格式: | Final Year Project |
| 語言: | English |
| 出版: |
Nanyang Technological University
2023
|
| 主題: | |
| 在線閱讀: | https://hdl.handle.net/10356/167694 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Nanyang Technological University |
| 語言: | English |
| 總結: | Transformers are being used to create music because of their capacity to record long-term dependencies and produce music of a high caliber. Yet, the absence of chord progression tokens is a drawback of employing transformers for music creation. Especially in genres like jazz and pop, chord progressions play a significant role in music and are frequently used to infuse a piece with harmony and tension.
The transformer model's lack of chord progression tokens makes it difficult to create music that adheres to a particular harmonic structure. To overcome this drawback, various strategies have been put forth, including the use of external knowledge sources and the modification of the transformer model to incorporate chord progression tokens.
The goal of this study is to investigate whether adding chord progression labels and a new parallel attention decoder will result in more musically appealing and realistic pieces. There were strong evidence in the pieces the model produced that it was able to pick up on the idea of chords and chord progressions, as well as produce better music metrics because it understood these hidden structures of music. |
|---|