Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Cluster 9 ☁Satvik N, Elena A, Conan L
Cluster 9 ☁Satvik N, Elena A, Conan L
The Objective
Generate music using machine learning.
HypothesisWe can utilize the GAN model to generate music, by training the algorithm on chroma and piano roll pairs.
GenerativeAdversarialNetwork
What is a GAN?
EXAMPLE
The Concept● As the generator gets better at coloring the
image, the discriminator gets better at detecting which one is the fake/original.○ Positive feedback loop
● The end product is a generator that can make images that appear real
● We theorize that we can use the same model to generate music.
PHASE 1TRAINING THE GAN - Successfully generate
music using chromas
Chromas encapsulate the chords and general prevalence of notes in a song. Chroma puts every note on a 12 value spectrum to provide a visual representation of the notes in a musical piece.
Training Data122 songs, all transposed to the key of C/Am, with their respective chromas
First Iteration
357,000 Epochs Later...
PHASE 2USE THE GENERATOR - Convert music in other
genres to pop
ExtensionConvert music in one style (i.e. classical) into another genre.
Because the GAN has only been exposed to pop music, the chroma that the user inputs will be made into a piano score in the style of pop.
Conclusion● Hypothesis somewhat supported by success
in converting Canon in D● A variety of music genres with several pieces
must be tested to truly test our hypothesis● Future development: train the computer
extensively and manually adjust the GAN algorithm to match our needs
● Learned a lot about machine learning, music theory, and working with Python
AcknowledgementsWe’d like to thank all of the people who have taught us in these last four weeks and helped us with our final project:
● Shlomo Dubnov, professor of Computer Music at UCSD● Jacob Sundstrom, Ph.D student of computer music● Mauricio de Oliveira, professor of Mechanical & Aerospace Engineering at UCSD● Aren Akian, Ph.D student of computer science● Gualter Moura, teacher fellow● Everyone from Cluster 9 ☁
● and viewers like you :)
Questions?