[Paper Review] A Computational Analysis of Real-World DJ Mixes using Mix-To-Track Subsequence Alignment

Tae‐Jun Kim, Minsuk Choi+3 more|arXiv (Cornell University)|Aug 24, 2020

Music and Audio Processing11 references4 citations

TL;DR

This paper proposes a subsequence dynamic time warping (DTW) method to align real-world DJ mixes with their original tracks, enabling precise extraction of cue points, transition lengths, and musical changes. Using 1,557 mixes from 1001Tracklists, the study reveals that DJs rarely transpose keys (only 2.5% of tracks transposed, mostly by one semitone) and tend to align cue points within 8 measures (73.6%) or 16 measures (86.2%) across different DJs, indicating shared creative conventions.

ABSTRACT

A DJ mix is a sequence of music tracks concatenated seamlessly, typically rendered for audiences in a live setting by a DJ on stage. As a DJ mix is produced in a studio or the live version is recorded for music streaming services, computational methods to analyze DJ mixes, for example, extracting track information or understanding DJ techniques, have drawn research interests. Many of previous works are, however, limited to identifying individual tracks in a mix or segmenting it, and the sizes of the datasets are usually small. In this paper, we provide an in-depth analysis of DJ music by aligning a mix to its original music tracks. We set up the subsequence alignment such that the audio features are less sensitive to the tempo or key change of the original track in a mix. This approach provides temporally tight mix-to-track matching from which we can obtain cue-points, transition length, mix segmentation, and musical changes in DJ performance. Using 1,557 mixes from 1001Tracklists including 13,728 tracks and 20,765 transitions, we conduct the proposed analysis and show a wide range of statistics, which may elucidate the creative process of DJ music making.

Motivation & Objective

• To enable large-scale computational analysis of real-world DJ mixes by overcoming limitations of prior small-scale, isolated studies on track identification, segmentation, and alignment.
• To develop a robust mix-to-track subsequence DTW method that is invariant to tempo and key changes, ensuring accurate alignment between mixes and original tracks.
• To extract and analyze cue points, transition lengths, and musical changes in DJ performances using a large dataset of 1,557 mixes and 13,728 tracks.
• To investigate whether DJs follow shared musical conventions in cue point selection, transition design, and pitch/tempo manipulation.
• To provide statistical evidence for DJ mixing practices, supporting future work in automatic mix generation and music information retrieval (MIR).

Proposed method

• Propose a subsequence dynamic time warping (DTW) framework that aligns audio segments from a DJ mix to original tracks, with audio features made invariant to tempo and key changes.
• Use MFCCs and key-invariant chroma features to improve robustness against pitch transposition and tempo variations during alignment.
• Extract cue-in and cue-out points from the warping paths to define track boundaries and transition regions in the mix.
• Evaluate mix segmentation performance by comparing automatically extracted cue points against human-annotated boundaries, analyzing differences in human annotation policies.
• Compute transition lengths in beats and analyze cue point deviations across DJs to assess agreement and identify common musical structures.
• Conduct statistical analysis on tempo adjustments and key transpositions using aligned audio segments, focusing on distribution and magnitude of changes.

Experimental results

Research questions

• RQ1 To what extent do DJs modify the tempo and key of original tracks in real-world mixes?
• RQ2 How do DJs align transitions between tracks in terms of musical structure, particularly in relation to phrase lengths?
• RQ3 How consistent are cue points selected by different DJs for the same track, and what is the distribution of deviations between their choices?
• RQ4 Which type of cue point (e.g., cue-in, cue-out, or both) most closely aligns with human-annotated boundaries in mix segmentation?
• RQ5 What patterns emerge in musical changes such as tempo and key adjustments that reflect DJ mixing conventions?

Key findings

• Only 2.5% of the 24,202 matched tracks were transposed, and 94.3% of those transpositions were by just one semitone, indicating minimal key changes in DJ mixes.
• The majority of tempo adjustments were small, with most tracks retaining their original tempo, suggesting DJs prioritize preserving the original track's integrity.
• Transition lengths in DJ mixes peak every 32 beats—commonly a musical phrase—indicating DJs align transitions with structural repetitions in dance music.
• 73.6% of cue point pairs across different DJs deviated by within 8 measures (32 beats), and 86.2% by within 16 measures, showing strong agreement on cue point placement.
• 23.6% of cue point pairs had zero deviation, and 40.4% were within one measure (4 beats), indicating shared conventions in selecting cue points.
• Human-annotated boundaries in mix segmentation most closely align with cue-in points of the next track, suggesting that DJs and listeners often perceive the start of a new track as the boundary.