Last week I went over my potential choices for my project. And the winner is…. drum roll please…

Top-n Music Genre Classification Neural Network! Now this might not be surprising since it was my number one pick, but I am excited nonetheless!

For any of you reading this that don’t fully understand what this project involves I am going to explain it much clearer here. My team will be creating a neural network capable of guessing the genre of songs. Now you might be asking, “Noah, what on Earth is a neural network?” Well you’re in luck because I’m going to elaborate using as little computer science jargon as possible.

Basically, a neural network is an attempt to teach computers how to think like a human, hence the name being based on our brain cells, neurons. Obviously, the computer isn’t actually thinking (at least I hope not!) but we can teach it how to recognize patterns. This involves giving it a ton of examples of what we want it to recognize and then checking if our results are what we expect.

Let’s be more specific. If I want a computer to understand handwritten numbers then I won’t just show it all the numbers once. There can be so much variation between how different people write numbers and even the same person writes their numbers slightly differently each time. If I want it to be able to recognize the number 9, for example, then I have to break it down to the smallest components. It has to recognize that there is a loop on top and a line underneath connected to the right side of the loop. This is obvious to the human brain, but the computer doesn’t know anything to begin with so we have to show it an unfathomable amount of nines until it gains an understanding of the variation that can occur. Eventually it will be able to consistently make the correct guess.

Through repetition, we can teach a computer how to recognize numbers, create images from text (check out DALL·E), and understand song genres!

My team will be forcing our system to listen to music nonstop and then have it make a guess about the genre. If it’s correct, GREAT. If it’s wrong (which it likely will be until it has undergone a TON of training), then no problem, we just keep tweaking the results until it gains an artificial understanding of different genres.

I hope this clears the topic up, but feel free to leave a reply if you have any questions!