Problem Statement: In the field of music information retrieval, convolutional neural networks (CNNs) have been successful at identifying different music genres by utilizing audio samples that are preprocessed into spectrogram images. I wanted to investigate how accurately a CNN could distinguish between more closely related styles of dance music, and more specifically between different styles of techno.
This project is split into two parts:
-
Part I uses a CNN for a binary classification problem and explores how well a network can identify between two styles of dance music: techno and EDM. Part I provides a more in-depth walk-through and explanation of all the data collection, preprocessing, and network design.
-
Part II uses a similar CNN for a multiclassification problem and explores how well the network can identify between five styles, or “genres” of techno. Part II is more concise and utilizes all of the same processes from Part I.
All code is in python and the tasks performed in this project include:
- Acquiring data from Spotify’s API with the help of Spotipy library
- Preprocessing of mp3 files into spectrogram images using librosa library
- Binary Classification with a 1D and 2D CNN
- Multiclassification with a 1D and 2D CNN