Getting started with generative AI with AWS DeepComposer
summary
Console demo shows AWS DeepComposer converts MIDI into piano roll images and uses GANs to add accompaniment via a GAN generator vs discriminator loop, AR‑CNN to perform sequential AR-CNN edit events (add/remove single notes) that remove many but not all out‑of‑tune notes, and Transformers that extend melodies into Transformer sequences >1000 tokens. Live example turns Twinkle Twinkle into a Bach‑style output then layers rock instruments; models run in the cloud and produce audible, iterative improvements.
AI-written summary · about
chapters (19)
- 0:00Introduction to session
- 1:05What is generative AI
- 2:32Machine learning basics
- 5:21Discriminative vs generative models
- 6:46Generative AI techniques overview
- 7:32AWS DeepComposer features
- 8:14How GANs work
- 11:56GANs real-world uses
- 13:43Autoregressive models explained
- 16:31AR-CNN demo samples
- 18:27Transformers introduction
- 21:21Transformers in DeepComposer
- 22:54Transformers applications
- 23:57Getting started with DeepComposer
- 26:28Creating music in Music Studio
- 28:41Enhancing melody with AR-CNN
- 29:01Adding instrument tracks with GANs
- 30:07Chartbusters challenges
- 31:14Conclusion and resources
related talks
01h 04mKeynote 3. Interactive generative models for music
43mA Composer’s Guide to Creating with Generative Neural Networks
24mCreating an AI Musician with JavaScript
44mWith AI, Does Music Have the Blues?
45mGenerative AI for Image and Video Production
53mGenerative AI: Overview and use cases
54mIntegral scherzo: how machines compose and sing
35mGenerative AI in the real world