Skip to main content
Have a personal or library account? Click to login
Raveform: A Dataset of Metrical and Functional Structure Annotations for EDM Tracks in DJ Mixes Cover

Raveform: A Dataset of Metrical and Functional Structure Annotations for EDM Tracks in DJ Mixes

Open Access
|Apr 2026

Figures & Tables

Figure 1

An illustration of components of the Raveform dataset. A DJ mix includes a track list, provides beats of the mix and track and has alignments between the DJ mix and each track at the beat level. Also, mix points are extracted based on the alignment. A subset of tracks are manually annotated structures: beats, downbeats, and functional segment boundaries with labels.

Table 1

Summary of the dataset components.

FileContent
mixes.jsonlMetadata for DJ mixes, including track lists.
tracks.jsonlMetadata for individual tracks that appear in the mixes.
alignments/
└ *.jsonl
Mix‑to‑track alignments for each mix, including estimated mix points.
beats/
└ mixes/*.json
└ tracks/*.json
Beats estimated for DJ mixes and tracks.
structures/
└ beats/*.csv
└ segments.json
Human‑annotated structure labels for a subset of tracks.
Figure 2

A representative structural pattern commonly observed in the proposed dataset, with darker regions indicating higher energy.

Table 2

A summary of the vocabulary for segment labels.

NameCharacteristic Summary
IntroAppears at the beginning. Primarily consists of percussive sounds, designed to facilitate seamless mixing for DJs.
BuildupTypically precedes a breakdown or a drop. Gradually increases energy by introducing musical elements one by one.
BreakdownAppears before a drop. Characterized by a sudden drop in energy to heighten contrast with the upcoming drop.
DropThe main section of the track, conveying its core musical idea. It is the most energetic and danceable part, usually featuring all instruments together.
CooldownFollows a drop and precedes a breakdown or outro. Gradually reduces energy, functioning as the opposite of a buildup.
BridgeOften occurs between two breakdowns before the final drop. Offers contrast and builds anticipation for the final drop. Rare in this dataset.
OutroOccurs at the end of the track. Functions as the opposite of an intro, yet similarly features percussive sounds to aid in mixing.
Ambient‑introAn intro section without percussive elements. Beats are nearly unrecognizable; the section consists mainly of melodic, harmonic, or ambient textures.
Ambient‑outroAn outro section without percussive elements. Sonically similar to an ambient‑intro.
Table 3

Summary statistics of the Raveform dataset.

The number of mixes4,902
The number of unique tracks56,873
The number of played tracks73,505
The number of tracks with structural annotations1,423
The total length of mixes (in hours)6,522
The number of available transitions53,780
Figure 3

The genre distribution of the DJ mixes.

Figure 4

The genre distribution of the structure‑annotated tracks.

Figure 5

The tempo distribution of the structure‑annotated tracks in beats per minute.

Figure 6

The distribution of the fraction of identified tracks per DJ mix.

Figure 7

The distribution of the fraction of correctly aligned beats.

Figure 8

The distribution of the number of segments in the structure‑annotated tracks.

Figure 9

The number of appearing segments of each segment label.

Figure 10

The distribution of segment length in seconds (top) and measures (bottom).

Figure 11

The common structure of EDM tracks.

Figure 12

Average loudness (dB) values across frequency bands for each segment label.

Table 4

Cross‑dataset evaluation results of metrical and functional structure analysis tested on Harmonix set and Raveform. Boldface indicates the best performance for each metric on each test set. denotes scores achieved with label mappings (see Section 6.1.3 for details).

Training Set (Genre)ModelBeatDownbeatSegmentLabel
F1CMLtAMLtF1CMLtAMLtHR.5FPWFSf
Tested on Harmonix set (Popular Music)
Harmonix (Pop)All‑In‑One0.9580.9130.9640.9150.8730.9320.6600.7380.769
Raveform (EDM)All‑In‑One0.8100.6220.8560.7270.5910.7920.5090.533*0.543*
Both (Pop+EDM)All‑In‑One0.9530.8950.9630.9210.8760.9390.6590.7200.751
10 Datasets (Various)Madmom0.9410.8590.9550.8050.7560.882
Tested on Raveform (EDM)
Raveform (EDM)All‑In‑One0.9910.9850.9910.9650.9640.9710.8350.8470.890
Harmonix (Pop)All‑In‑One0.9300.8900.9180.7530.7460.8120.6350.542*0.643*
Both (Pop+EDM)All‑In‑One0.9900.9850.9890.9670.9650.9710.8350.8420.890
10 Datasets (Various)Madmom0.9470.9300.9380.6690.6780.792
DOI: https://doi.org/10.5334/tismir.288 | Journal eISSN: 2514-3298
Language: English
Submitted on: Jun 10, 2025
Accepted on: Jan 9, 2026
Published on: Apr 20, 2026
Published by: Ubiquity Press
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year

© 2026 Taejun Kim, Jongsoo Kim, Hyungyu Kim, Juhan Nam, published by Ubiquity Press
This work is licensed under the Creative Commons Attribution 4.0 License.