GD-Studio/embeat_45m_spotify_tracks

Embeat 45M Spotify Tracks

A large-scale music metadata dataset containing 45 million Spotify tracks, combined with the Spotify metadata from Anna's Archive and artist genres from Every Noise at Once.

GitHub project: https://github.com/gdstudio-org/Embeat

Dataset Preview

>>> from datasets import load_from_disk
>>> ds = load_from_disk("GD-Studio/embeat_45m_spotify_tracks")
>>> ds
Dataset({
    features: ['track_id', 'track_name', 'isrc', 'popularity', 'explicit', 'artist_idx', 'artist_id', 'artist_name', 'artist_popularity', 'artist_genres', 'artist_genre_idx', 'album_id', 'album_name', 'release_year', 'duration', 'time_signature', 'tempo', 'key', 'mode', 'danceability', 'energy', 'loudness', 'speechiness', 'acousticness', 'instrumentalness', 'liveness', 'valence'],
    num_rows: 45059660
})
>>> ds[0]
{'track_id': '2plbrEY59IikOBgBGLjaoe', 'track_name': 'Die With A Smile', 'isrc': 'USUM72409273',
 'popularity': 1.0, 'explicit': 0, 'artist_idx': 7, 'artist_id': '0du5cEVh5yTK9QJze8zA0C',
 'artist_name': 'Bruno Mars', 'artist_popularity': 0.95, 'artist_genres': 'dance pop, pop',
 'artist_genre_idx': 8, 'album_id': '10FLjwfpbxLmW8c25Xyc2N', 'album_name': 'Die With A Smile',
 'release_year': 2024, 'duration': 251, 'time_signature': 3, 'tempo': 158, 'key': 6, 'mode': 0,
 'danceability': 0.521, 'energy': 0.592, 'loudness': -7.777, 'speechiness': 0.0304,
 'acousticness': 0.308, 'instrumentalness': 0.0, 'liveness': 0.122, 'valence': 0.535}

Dataset Overview

Each row represents ONE track (Spotify track_id), enriched with:

• Track metadata (name / ISRC / popularity / explicit)
• Primary artist (id / name / popularity / genres)
• Album metadata (id / name / release year)
• Spotify audio features (danceability, energy, etc.)

Dataset size: 45,059,660 rows

Sorted by: popularity (desc), track_id (asc)

Filtering Rules

• Track selection: tracks with popularity >= 1 (Spotify 0..100 scale)
• Audio features selection: keep only records with null_response == 0

Indexes

We create two categorical indices for compact downstream modeling:

artist_idx

• Range: 0..N
• 0 is reserved for missing/empty/"<UNK>" artist_name
• 1..N are assigned by descending artist popularity (higher popularity => smaller index)

artist_genre_idx

• Range: 0..M
• 0 is reserved for missing/empty/"<UNK>"
• We obtained data from everynoise.com, and use specialized formulas to calculate the optimal genre for each artist (Docs -> v0.4 chapter)
• If an artist has a valid artist_genre_idx but the artist_genres is empty, it means artist_genre_idx is inferred based on related artists
• 1..M are assigned by everynoise.com (https://everynoise.com/everynoise1d.html)

Field Definitions

Track Info

Field	Type	Description
track_id	string	Spotify base62 track id
track_name	string	Track name
isrc	string	International Standard Recording Code (ISRC). May be empty/missing
popularity	float32	Track popularity, normalized to 0.0..1.0 (original Spotify scale 0..100)
explicit	int8	0 = non-explicit, 1 = explicit

Artist Info (Primary Artist)

Field	Type	Description
artist_idx	int32	Popularity-based index (see "Indexes")
artist_id	string	Spotify base62 artist id (redirect-applied)
artist_name	string	May be empty for unknown artists
artist_popularity	float32	Normalized to 0.0..1.0, rounded to 2 decimals
artist_genres	string	Comma-separated genres (delimiter is ", "), may be empty
artist_genre_idx	int16	Frequency-based index (see "Indexes")

Album Info

Field	Type	Description
album_id	string	Spotify base62 album id
album_name	string	Album name
release_year	int16	Album release year. 0 means missing/unknown/invalid

Audio Features

Field	Type	Description
duration	int32	Duration in seconds, computed as floor (duration_ms / 1000)
time_signature	int8	Estimated time signature. Typically 3..7
tempo	int16	Estimated tempo in BPM, rounded to nearest integer
key	int8	Estimated key as pitch class. 0=C, 1=C#/Db, 2=D, ..., 11=B. -1 means unknown
mode	int8	Modality. 1=major, 0=minor
danceability	float32	0.0..1.0
energy	float32	0.0..1.0
loudness	float32	Overall loudness in dB. Typically -60.0..0.0
speechiness	float32	0.0..1.0
acousticness	float32	0.0..1.0
instrumentalness	float32	0.0..1.0
liveness	float32	0.0..1.0
valence	float32	0.0..1.0

Mapping Files

• artist_genre_map.json: artist_genre_idx -> artist_genre name

Acknowledgements

• Anna's Archive
• Every Noise at Once

License

CC BY-NC 4.0

FOR STUDY PURPOSES ONLY. DO NOT USE IT COMMERCIALLY!

Provider

GD Studio

2026-06-26