125 points by music_nerd 1 year ago flag hide 9 comments
deeplearningmusician 1 year ago next
Just published my new project exploring generative AI for music composition! Would love to hear your thoughts on the approach and results. Link to GitHub repo and demo in the comments!
pythoncoder45 1 year ago next
Great work! I'm particularly interested in AI and music. What ML libraries did you use for this project? Would love to try it out.
deeplearningmusician 1 year ago next
Pythoncoder45, I mostly used TensorFlow and the Magenta library. If you're interested, I'll write a blog post on my experience using them. Just subscribe to my newsletter!
musiclover123 1 year ago prev next
Incredible! It sounds like the AI did a great job composing. Have you considered publishing your dataset to encourage further research?
deeplearningmusician 1 year ago next
Musiclover123, I'm so glad you liked it! My dataset includes MIDI files and related metadata, you'll find it in the Github repo. Enjoy using it and make something amazing!
ai_enthusiast 1 year ago prev next
Very cool project! I've worked on music generation as well, though without the AI element. Best of luck with future projects!
deeplearningmusician 1 year ago next
AI_enthusiast, thank you! I've always appreciated the importance of human touch even if we lean on AI. Let's connect and share our experiences!
code_master 1 year ago prev next
I've tried using generative AI previously, but not so successful. Can you share the specific techniques and approaches that you used for this music composition?
deeplearningmusician 1 year ago next
Sure, code_master. My main focus was training a Variational Autoencoder (VAE) with the help of the Magenta library. I used Sargeant's mixtape dataset, which consists of a wide range of music and curated the dataset that included MIDI files and related metadata to make it work with Magenta. Check out this tutorial for the basis of the code: <https://magenta.tensorflow.org/tutorials/midi_seq_generation_overview>