Hi everyone!
Before I talk about my favorite technology from our capstone project, I thought I should give a little bit of context for how its being used. Our project is titled ‘Top-N Music Genre Classifcation Neural Network’ and is centered around creating a program where a user can submit an audio clip and then receive some statistics detailing which genres the music clip most likely fits into. This is done using a neural network trained in Tensorflow on various music clips from different genres. The music data is gathered using a Python library called Librosa, which contains a multitude of functions for loading, analyzing, and saving audio files. Although Tensorflow plays a large role in our project, since it is largely responsible for developing the neural network, Librosa has felt more interesting to learn about and delve into thus far.
In my experience coding with Python, I have never had the opportunity to work with audio-related data. While I have built other models using packages like Tensorflow, they have always been focused on other types of data. Learning about the process of analyzing audio data has been extremely interesting, and I have found some unexpected similarities in how audio is handled compared to other data types I have worked with in the past. For example, a project I worked on professionally dealt with EEG data, and part of the data pipeline involved performing an operation called a Fourier Transform. In essence, the Fourier Transform ‘transforms’ data from one domain to another (e.g. time domain to frequency domain). While I was familiar with the process, I was surprised when it came up that we needed to transform our audio data using the same process, but in Librosa! It was exciting that I could apply some past knowledge I had in an unexpected way, and I think that is one of the main reasons I have found Librosa and this project so interesting!
Leave a Reply