The Million Song Dataset (MSD) contains a variety of metadata and audio features for 1 million songs. The dataset is structured into multiple files, with the main dataset stored in HDF5 format. Here’s a breakdown of the important columns:
1. Metadata
| Column Name | Description | Example |
|---|---|---|
song_id | Unique identifier for a song | SOQHXMF12A6D4FD69A |
title | Song title | Bohemian Rhapsody |
artist_id | Unique identifier for the artist | ARJ7KF01187FB3F5F1 |
artist_name | Name of the artist | Queen |
release | Album or single name | A Night at the Opera |
year | Release year | 1975 |
duration | Length of the song in seconds | 355.47 |
track_id | MusicBrainz Track ID | TRMMMYQ128F932D901 |
2. Audio Features
| Column Name | Description | Example |
|---|---|---|
tempo | BPM (beats per minute) | 120.5 |
key | Musical key (0 = C, 1 = C#, …, 11 = B) | 5 (F Major) |
mode | 1 = Major, 0 = Minor | 1 |
time_signature | Time signature (3, 4, or 5) | 4 (4/4 time) |
loudness | Overall loudness in dB | -5.5 |
danceability | A measure of how danceable a song is (0-1) | 0.8 |
energy | Energy level of the track (0-1) | 0.9 |
3. Genre & Similarity
| Column Name | Description | Example |
|---|---|---|
artist_terms | List of genre tags | ['rock', 'classic rock'] |
artist_familiarity | Popularity score (0-1) | 0.85 |
artist_hotttnesss | Hotness score (0-1) | 0.9 |
similar_artists | List of similar artists’ IDs | ['AR1', 'AR2', 'AR3'] |
4. Additional Features (Segment-Level Analysis)
These are detailed audio features analyzed per second for deeper audio processing:
- MFCCs (Mel-Frequency Cepstral Coefficients) – Used in audio signal processing
- Pitch & Timbre Features – For sound quality analysis
Would you like help fetching, analyzing, or visualizing this data for probability and statistics? 🚀