Data Side of the Moon - artwork generated by DALL·E 3
This project is currently a work in progress. Contributions, suggestions, and feedback are greatly appreciated.
How can computer "precieve" music? How we, human, understand music? From a higher perspective, the process of precieve music is consist of 3 main steps: we hear(encode), we understand, and (if we want) we sing(decode) and then we hear (again). Base on this idea, we implement a neural network based on convolutional autoencoder architecture, and then explore the learning representation of music.
The key idea of this project is using unsupervised learning to explore music, which should be distinguished from music genres classification tasks which we consider has its own limitation due to the labels(genres) are limited by human ourselves. We believe that generes always falls behind music.
"Echoes" model architecture
"Echoes" model training
Latent vectors of GTZAN testset
