- Audio track (encoded as mp3) of each of the 106,574 tracks. It is on average 10 millions samples per track.
- Nine audio features (consisting of 518 attributes) for each of the 106,574 tracks.
- Given the metadata, multiple problems can be explored: recommendation, genre recognition, artist identification, year prediction, music annotation, unsupervized categorization.
- Please see the paper and the GitHub repository for more information ([Web Link])
Nine audio features computed across time and summarized with seven statistics (mean, standard deviation, skew, kurtosis, median, minimum, maximum):
- Chroma, 84 attributes
- Tonnetz, 42 attributes
- Mel Frequency Cepstral Coefficient (MFCC), 140 attributes
- Spectral centroid, 7 attributes
- Spectral bandwidth, 7 attributes
- Spectral contrast, 49 attributes
- Spectral rolloff, 7 attributes
- Root Mean Square energy, 7 attributes
- Zero-crossing rate, 7 attributes
- features.csv : can be used in classification (use a subset)
- echonest.csv : temporal features (?)
The useful .ipynb according to Salvatore are:
- usage
- analysis
- baselines (starting classification models)