Convolutional Neural Network that classifies different genres of music
Based on the AlexNet architecture, the network consists of 3 convolutional layers, and 2 fully-connected layers, classifying music into 10 different genres (Blues, Classical, Country, Disco, Hip-Hop, Jazz, Metal, Pop, Rock, and Reggae). With an accuracy of 63.7% it is not state-of-the-art(6 genres:accuracy of 84%,250 genres: 93.6%).
UPDATE: I tried a few more different architectures and I hope to try out a few more soon.
| Architecture | Accuracy |
|---|---|
| AlexNet | 63.7 |
| SqueezeNet | 64 |
| ResNet34 | 84.5 |
| ResNet50 | 85 |
- Adding an LSTM (or another derivative recurrent neural net) layer that helps capture the sequential nature of music.
- Train and test on more data.