Learn from the first place winner of the first AWS DeepComposer Chartbusters Bach to the Future Challenge
AWS is excited to announce the winner of the first AWS DeepComposer Chartbusters Challenge, Catherine Chui. AWS DeepComposer gives developers a creative way to get started with machine learning (ML). To make the learning more fun, in June 2020 we launched the Chartbusters Challenge, a competition where developers use AWS DeepComposer to create original compositions and compete to showcase their ML and generative AI skills. The first challenge, Bach to the Future, required developers to use a new generative AI algorithm provided on the AWS DeepComposer console to create compositions in the style of the classical composer Bach.
When Catherine Chui first learned about AWS DeepComposer, she had no idea that one day she would be the winner for the Chartbusters Challenge and be the star of a blog post on the AWS Machine Learning Blog. We interviewed Catherine about her experience with the challenge and how she created her winning composition.
Getting started with machine learning
Before entering the Chartbusters Challenge, Catherine had no prior experience with ML, and described herself as a beginner with AI. She was first introduced to AWS DeepComposer when her husband, Kitson, attended AWS re:Invent in 2019. Kitson is a teacher of the first AWS Educate Cloud Degree course, Higher Diploma in Cloud and Data Centre Administration, at the Hong Kong Institute of Vocational Education (IVE).
“After Kitson accompanied an IVE student to AWS re:Invent 2019 to join the AWS DeepRacer Championship Cup, he attended an AWS DeepComposer workshop and brought one unit back to Hong Kong,” recalled Catherine. “That was the first time I had a look at the actual product and got some understanding of it.”
Catherine was inspired to compete in the Chartbusters Challenge when she saw her husband playing with the AWS DeepComposer keyboard with his students. Catherine loves classical piano music, having achieved the Associated Board of Royal Schools of Music (ABRSM) level 7.
“At first, I was surprised why he was playing a tiny piano at home with his students, as I knew he is an IT teacher, not a music teacher. I do feel his piano skills are weak, so I started to help him to teach his students piano skills.”
Her curiosity with the AWS DeepComposer keyboard led her to work with Kitson’s students, who were also competing in the Chartbusters Challenge. After helping the students with their piano skills, she was inspired to compete in Bach to the Future.
“Within an hour of learning, I completed my first song with AI, which is fun and exciting!”
Building in AWS DeepComposer
To get started, she connected her AWS DeepComposer keyboard to the console and recorded an input melody. Catherine chose the Autoregressive Generative AI technique and Autoregressive CNN (AR-CNN) Bach model. The AR-CNN algorithm allows you to collaborate iteratively with the ML algorithm by experimenting with the hyperparameters to create an original composition. When deciding how to manipulate her model, she took the following into account:
“We can use the output from one iteration of the AR-CNN algorithm as an input to the next iteration to help make it better and smoother. I kept the creative risk as low as possible, and didn’t remove too much original notes as I would like to keep my original melody.”
The following screenshot shows the Music studio page on the AWS DeepComposer console.
Catherine spent a couple of hours recording her melody, then spent time adjusting the compositions to enhance her composition. She created around 10 compositions and evaluated each composition until she was satisfied with her final melody. Catherine found the arpeggio and chord functions within the AWS DeepComposer keyboard to be helpful for auto-generating notes for her composition.
She learned more about the AR-CNN algorithm in the AWS DeepComposer learning capsules. Learning capsules provide easy-to-consume, bite-size content to help you learn the concepts of generative AI algorithms.
“I learned the concept of AR-CNN, two neural networks that have a sophisticated design to help in adding and removing notes. I wouldn’t get to experience it outside this setting. Although at the moment I’m still not familiar with Amazon SageMaker, I think the learning capsules will help me in the future.”
The following screenshot shows the available learning capsules on the AWS DeepComposer console.
You can listen to Catherine’s winning composition, “Garden Partying in Bach,” on the AWS DeepComposer Soundcloud page.
The AWS DeepComposer Chartbusters Challenge Bach to the Future helped Catherine, who had no background in ML, develop her understanding of generative AI in just a few hours and then win the first AWS DeepComposer Chartbusters Challenge.
“The challenge inspired me that machine learning can bring up some ideas on composing new and creative music. I will keep joining the next challenges to learn something new and advanced about machine learning. Also, I will further help my husband’s students at IVE by giving them feedback on music.”
Congratulations to Catherine Chui for her well-deserved win!
We hope Catherine’s story has inspired you to learn more about ML and AWS DeepComposer. Check out the next AWS DeepComposer Chartbusters Challenge, The Sounds of Science, that will run from 9/1 to 9/23.
About the Author
Paloma Pineda is a Product Marketing Manager for AWS Artificial Intelligence Devices. She is passionate about the intersection of technology, art, and human centered design. Out of the office, Paloma enjoys photography, watching foreign films, and cooking French cuisine.