Working with AI is always exciting! This time we decided to make an application that would assist a user in making music based on some theme that they have in mind. Gotta generate some last minute background track for your school project? Fire this puppy up and you have a brand new, copyright free music track.
To make this, I, personally, had to go through the process of understanding how an LSTM is implemented. I have previously worked with a few traditional Neural Network architectures (e.g. Convolutional Networks) and I am so glad to add LSTMs to my list. It wouldn't have been possible without the help of all the articles that I found online.
To make the input/output work, my teammate had to figure out a way to work with midi files. Having no previous experience with such files, she did an extremely good job at handling the IO. We originally planned on hosting our server on Heroku but we are leaving that to a future date.
What is the future of this app, you ask? This application can be modified to include more than just one instrument and can definitely be extended to able to get trained on audio file formats such as mp3 instead of just midi. We can possibly combine LSTMs with GANs (Generative Adversarial Networks) to create a clearer, more accurate output. We could also incorporate direct input from a midi instrument in our website.
Log in or sign up for Devpost to join the conversation.