Open AI have developed an open source scripts that can write songs, I want to use the concept of it to build music genre translation tool.
What it does
This project will translate one style of music to another style.
How we built it
I want to use an autoencoder to build the encoder and decoder. By modify an 1D autoencoder of music and design the training process.
Challenges we ran into
It's pretty challenge than I thought, at the end I haven't have a very good demo for the music translation. I got a initial demo for the music translation.
Accomplishments that we're proud of
I was able to understand some music processing concept and a preliminary demo of the result.
What we learned
Learned the 1D conv network and audio preprocessing stuffs.
What's next for MusicTranslation
I will build a network to reconstruct the music based on latent feature. Current encoder and decode has lot of noise. I would like to improve the performance of the network.