The ultimate music genre classifier using both acoustic features and spectrogram
The dataset is a collection of 30 seconds music samples in four different genres and is gathered from deezer.com.
The dataset including raw audio files and extracted features is available at: https://www.kaggle.com/datasets/kianeliasi/genrenet
Mean and variance of the follwing features:
- Root Mean Square
- Chroma Feature
- Spectral Centroid
- Spectral Bandwidth
- Rolloff Frequency
- Zero-Crossing Rate
- Mel-Frequency Cepstrum Coefficients
Additionally:
- Black and white pectrogram images
- Fully-connected deep network for acoustic features
- Convolutional Neural Network for spectrograms
Aggregation method: choose the model with more confident outputs