Hi everyone!
My project for this quarter is the Top-N Music Genre Classification Neural Network. Essentially we are creating a model that can take a user’s song input and return what genre(s) the song may be with a confidence value. So far I am really enjoying working on this project and all of the technologies I have worked with but also the technologies my teammates have worked with including librosa, TensorFlow, and PyTorch. So far my teammates and I have chosen to split the work into individual sections that we can later put together to make everything work.
Right now my teammate Taylor and I have been working to expand our data set and creating Mel spectrograms for our song clips. The original data set that we are using is the GTZAN dataset: https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification
This data set consists of 10 genres with 100 songs each so there are 1000 song clips. This dataset already has Mel spectrograms however another teammate Anthony has noticed there is some inconsistency between the way our spectrograms look and the ones created for the GTZAN dataset. We’ve come to the conclusion that it would be best to use Librosa to recreate the spectrograms so all of our spectrograms can fit the same parameters. This kind of leads me to what my favorite technology is. I have really enjoyed using Librosa and it is the main technology that I have used for the project thus far.
Librosa is a Python package and it is mainly used for both music and audio analysis, which is perfect for what my group is trying to accomplish with our model. As I mentioned we have been using it to create Mel spectrograms and will likely continue to do so. I personally don’t have anything I would change about Librosa I feel like it is working as intended and helping us greatly, especially with expanding our dataset and we might even use it to create new spectrograms for the GTZAN dataset. Overall I think my group and I are making great progress towards our goals and I am really enjoying working with Librosa and hearing about my teammate’s work with other technologies. Even for the technologies that I haven’t worked directly with I feel like I am learning so much about their capabilities.
Here’s a link to more information about Librosa if anyone is interested!
https://librosa.org/doc/latest/index.html
Leave a Reply