Music Genre Classification

The goal of the project is to have a way for music creators or listeners to better understand the music world around them by classifying any music input to a genre

Comment

Inspiration

Me and my team were inspired by our passion for music and what applications it could have in the software development world

What it does

Our Neural Network processes 30 second segments of sound data to determine which genre that segment of music is likely to be categorized as. This involves both the training and testing data, which are split into 10 3-second clips of data that can be analyzed sequentially using a Recurrence Neural Network (RNN)

How we built it

We used the numpy python library to implement efficient matrix multiplication and transformations, which we took full advantage of to store and manipulate neuron state over time, as well as their interactions with the weights and biases.

Challenges we ran into

We ran into two main challenges we ran into: 1) Buffer size limitations: When dealing with large matrices, especially as they grow and shrink over the course of thousands of iterations of algorithms, it can be difficult to ensure that the data will stay in the bounds of numbers that are practical for computation. Known as the exploding and vanishing gradient problems, these issues caused us quite the headache to try and solve 2) Implementation difficulties: Because we decided to implement such a complicated Neural Network from scratch, we inevitably ran into some serious challenges to our understanding along the way. From figuring out how to implement back propagation to experiencing firsthand why GRU and LSTM are necessary, despite being difficult to implement, we ran into a lot of challenges, but learned a lot as a result

Accomplishments that we're proud of

We are proud that from little to no knowledge in the field of AI, that we were able to practically bring a neural network model to life. While it sits incomplete at the time of submission, about 90% of our goal was accomplished. It is amazing how much we could get accomplished in such a short amount of time.

What we learned

Together, our team learned the important math fundamentals behind AI. Trying to understand every step behind this new concept to us was no easy feat. We learned how to manage and work on a project in a group setting and just how complicated AI can be. We challenged ourselves to make a neural network from scratch and gained a whole lot of helpful insight.

What's next for Music Genre Classification

We are still planning on finishing the project after the hackathon. For both of us, these were new concepts, but ones which interested us deeply, hence our commitment to sacrifice 24 straight hours of our lives to implementing it. For this reason, we will continue to work after the deadline until we both have a good understanding of the fascinating mathematics and computer science that allow computers to think.

Built With

Share this project:

Updates