Information Management and Preservation
Information and Software Engineering Group (IFS)Institute of Software Technology and Interactive Systems (ISIS)
Vienna University of Technology
Million Song Dataset Benchmarks - Downloads
Complete Feature Files (in WEKA ARFF Format)
| Feature Set | Version | Dimensions | Download | md5 | Size | |
|---|---|---|---|---|---|---|
| Rhythm Patterns Features (config) | ||||||
| Rhythm Patterns | 1.0 | 1440 | msd-rp-v1.0.arff.gz | md5 | 4,13GB | |
| Statistical Spectrum Descriptors | 1.0 | 168 | msd-ssd-v1.0.arff.gz | md5 | 641MB | |
| Rhythm Histograms | 1.0 | 60 | msd-rh-v1.0.arff.gz | md5 | 240MB | |
| Temporal Statistical Spectrum Descriptors | 1.0 | 1176 | msd-tssd-v1.0.arff.gz | md5 | 3,97GB | |
| Temporal Rhythm Histograms | 1.0 | 420 | msd-trh-v1.0.arff.gz | md5 | 1,48GB | |
| MVD | 1.0 | 420 | msd-mvd-v1.0.arff.gz | md5 | 1,32GB | |
| Marsyas (config) | ||||||
| MARSYAS timbral features | 1.0 | 124 | msd-marsyas-timbral-v1.0.arff.gz | md5 | 412MB | |
| jMir (details) | ||||||
| Low-level features (Spectral Centroid, Spectral Rolloff Point, Spectral Flux, Compactness, Spectral Variability, Root Mean Square, Zero Crossings, and Fraction of Low Energy Windows) |
1.0 | 16 | msd-jmir-spectral-all-all-v1.0.arff.gz | md5 | 52MB | |
| Low-level features derivatives | 1.0 | 96 | msd-jmir-spectral-all-all-v1.0.arff.gz | md5 | 287MB | |
| Method of Moments | 1.0 | 10 | msd-jmir-methods-of-moments-all-v1.0.arff.gz | md5 | 36MB | |
| Area of Moments | 1.0 | 20 | msd-jmir-area_of_moments-all-v1.0.arff.gz | md5 | 67MB | |
| Linear Predictive Coding (LPC) | 1.0 | 20 | msd-jmir-lpc-all-v1.0.arff.gz | md5 | 54MB | |
| MFCC features | 1.0 | 26 | msd-jmir-mfcc-all-v1.0.arff.gz | md5 | 72MB | |
Ground Truth assignments
All Music Guide genres (http://allmusic.com)
| Ground Truth Name | Number of Labels | Descritpion |
|---|---|---|
| MSD Allmusic Genre Dataset (MAGD) | 422,714 | details |
| MSD Allmusic Top Genre Dataset (Top-MAGD) | 406,427 | details |
| MSD Allmusic Style Dataset(MASD) | 273,936 | details |
Classification Tasks
| Description | |||||
|---|---|---|---|---|---|
| Non-stratified splits | |||||
| 90% training data | MAGD | Top-MAGD | MASD | ||
| 80% training data | MAGD | Top-MAGD | MASD | ||
| 66% training data | MAGD | Top-MAGD | MASD | ||
| 55% training data | MAGD | Top-MAGD | MASD | ||
| Stratified splits | |||||
| 90% training data | MAGD | Top-MAGD | MASD | ||
| 80% training data | MAGD | Top-MAGD | MASD | ||
| 66% training data | MAGD | Top-MAGD | MASD | ||
| 55% training data | MAGD | Top-MAGD | MASD | ||
| Splits with fixed size per genre | |||||
| 1,000 samples training data / genre set | MAGD | Top-MAGD | MASD | ||
| 2,000 samples training data / genre set | MAGD | Top-MAGD | MASD | ||
Statistic Files
| Description | Download |
|---|---|
| Missing tracks | missing_tracks.txt |
| All sample properties | sample_properties.csv.gz |
| Mp3 bitrate statistics | bitRateStats.csv |
| Mp3 channel statistics | channelStats.csv |
| Mp3 samplerate statistics | sampleRateStats.csv |
| Mp3 sample length statistics | songLengthStats.csv |
Scripts
| Description | Version | Link |
|---|---|---|
| Assign ground truth labels to feature file | 1.0 | download |
| Merge feature file with ground truth | 1.0 | download |
| Merge feature file with multi label ground truth | 1.0 | download |
| Filter specific features from feature file | 1.0 | download |