LoopLens: Supporting Search as Creation in Loop-Based Music Composition

Sheng Long; Atsuya Kobayashi; Kei Tateno

cs.HCMar 9, 2026

LoopLens: Supporting Search as Creation in Loop-Based Music Composition

Authors: Sheng Long, Atsuya Kobayashi, Kei Tateno

Organizations: Northwestern University Evanston, USA · Sony Group Corporation Tokyo, Japan

Abstract

Creativity support tools (CSTs) typically frame search as information retrieval, yet in practices like electronic dance music production, search serves as a creative medium for collage-style composition. To address this gap, we present LoopLens, a research probe for loop-based music composition that visualizes audio search results to support creative foraging and assembling. We evaluated LoopLens in a within-subject user study with 16 participants of diverse musical domain expertise, performing both open-ended (divergent) and goal-directed (convergent) tasks. Our results reveal a clear behavioral split: participants with domain expertise leveraged multimodal cues to quickly exploit a narrow set of loops, while those without domain knowledge relied primarily on audio impressions, engaging in broad exploration often constrained by limited musical vocabulary for query formulation. This behavioral dichotomy provides a new lens for understanding the balance between exploration and exploitation in creative search and offers clear design implications for supporting vocabulary-independent discovery in future CSTs.

Figures & tables

Evaluating What?	How?	Example(s)
Creative Process	Users’ self-reported answers using questionnaires/survey	NASA-TLX ( Hart and Staveland, 1988 ) , Creativity Support Index (CSI) ( Carroll and Latulipe, 2009 ) , Cognitive Dimensions of Music Notations ( Nash, 2015 )
Creative Process	Users’ interaction-log	Kim et al. ( Kim et al., 2025b )
Creative Process	User interview	Most user studies do this
Creative Outcome	Judgments by humans (experts and/or lay users)	Louie et al. ( Louie et al., 2022 ) , Loui et al. ( Loui et al., 2024 )
Creative Outcome	Homogenization analysis	Anderson et al. ( Anderson et al., 2024 )
Creator Characteristics (general-purpose)	Torrance Test of Creative Thinking (TTCT) ( Torrance, 1968 )	Loui et al. ( Loui et al., 2024 )

Table 1. Common approaches to evaluating creativity support tools (CSTs) in HCI, organized by whether they target creative processes, outcomes, or creator characteristics. The table illustrates both widely used domain-general instruments (e.g., CSI, TTCT) and domain-specific measures (e.g., Gold-MSI, Music Divergent Thinking Test).

Figure 1. Example of existing radial charts and diagrams. Except the Eternal Jukebox (f), all other forms of visualizing music information takes in “nice” input formats that are not strictly audio but symbolic music notation. See Draper et al. ( Draper et al., 2009 ) for a survey on radial methods for visualization and Lima et al. ( Lima et al., 2021 ) for music visualization techniques.

Figure 2. Screenshots from Ableton Live 12 and Apple Logic Pro’s interfaces for searching and browsing music.

Figure 3. Screenshot of LoopLens user interface. Note that the purple shade for DAW is manual annotation and not what the actual interface appears like. Interface

Figure 4. Preview of song during drag and drop.

Figure 5. Screenshot of DAW in LoopLens .

Figure 6. Screenshot of “Find Similar” modal for the song, Piano (Resonance/Room) .wav , by StonedB, in LoopLens .

Figure 7. Overview of the user study procedure.

Participant ID	Prior Musical Training	DAW	Company Role
P1	3/2/0	Advanced	Research
P2	0/6-9/6-9	None	Research
P3	1/3-5/0	Beginner	Engineering
P4	4-6/6-9/10+	Advanced	Research
P5	0.5/0/0	Advanced	Research
P6	0/3-5/10+	None	Research

Table 2. Overview of details related to participant’s backgrounds. For “prior musical training”, the numbers correspond to the number of years received in musical theory training, music instrument training, and total years of practicing music on a daily and regular basis, corresponding to questions (3), (4), and (1) in Section C.1 . A participant could have no training in music theory, yet could have been practicing music on a daily basis for more than 10 years (e.g., P16 ).

Figure 8. Picture prompt for divergent task. A picture with blue background light showing people watching concert during night time.

Figure 9. Scatter plots showing the relationship between participant DAT scores and self-reported years of experience, faceted by three training types: Music Theory, Instrument, and Practice. The vertical dashed line represents the mean DAT score.

Figure 10. Distribution of participants’ ratings on self-perceived experience using LoopLens .

Panel of Figure 11 — Figure 11. Word clouds for both the (a) convergent task and the (b) divergent task. Notice that “Chiptune” appeared in search queries for the convergent task even though it did not appear in the tags provided on the reference song page.

Figure 12. P12 using Slider 140bpm by Clinthammer ( https://freesound.org/s/179542/ ) as an example. (a) and (b) are the radial and linear waveforms of the same song.

Figure 13. Homogenization analysis of participant’s composition with bootstrap confidence interval overlaid on top. Error bars indicate resampled uncertainty around the mean.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 14. Screenshot of Ableton Live 12’s tool, taken from https://www.ableton.com/en/live/ . Note that the browser is on the left.

Figure 15. Screenshot of Apple Logic Pro, taken from https://support.apple.com/guide/logicpro/welcome/mac . Note that the Loop Browser is currently collapsed but can be toggled via a button on the top left corner of the interface.

Computational Creativity

Computational creativity studies how machines can generate, evaluate, and collaborate on novel, valuable ideas and artifacts.

See all 114 papers

Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation
Sep 24, 2026Matt SandlerRecommender Systems Music Information Retrieval
Feed.fm San Francisco, California, USA
Quality-Diversity Search in Sound Generation: Investigating Innovation Engines for Audio Exploration
Jun 8, 2026Björn Þór Jónsson, Çağrı Erdem, Stefano Fasciani +1Evolutionary Optimization Quality-Diversity Optimization
1*Department of Informatics, University of Oslo, Norway. · Department of Musicology, University of Oslo, Norway.
Inducing Sustained Creativity and Diversity in Large Language Models
Mar 19, 2026Queenie Luo, Gary King, Michael Puett +1Language Model Decoding Diverse Text Generation
Ph.D. candidate, Department of East Asian Languages and Civilizations, Harvard University · Albert J. Weatherhead III University Professor, Institute for Quantitative Social Science, Harvard University · Walter C. Klein Professor of Chinese History and Anthropology, Harvard University +1

Sep 24, 2026J/K move · Enter open · S save

Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation

Matt Sandler

Feed.fm San Francisco, California, USA

Recommender Systems Music Information Retrieval Music Recommendation Information Retrieval Audio-Text Retrieval

Open paper AskPDF arXiv

A deployed B2B music-discovery platform serves both query-driven search (text prompts, vibe tags) and seed-driven recommendation (seed-track and artist stations) over one licensed catalog, one LAION-CLAP joint audio-text embedding space, one candidate-generation filter, and one ranking head -- and neither path consumes end-listener behavioral signal. In this content-only regime, curator judgment is the principal feedback signal available, and offline cosine similarity predicts it poorly: 38% of cosine-nearest neighbors are rejected by curators. The rejections reveal a clean partition: a majority (55%) are sound failures the encoder could address (style, tempo, mood mismatch), and a substantial minority (37%) are context failures orthogonal to the waveform (wrong language, holiday content, devotional content, rights and lyric flags). We deploy this sound-vs-context decomposition as feedback infrastructure, routing each failure mode to the layer that can absorb it: context failures to a constraint filter at candidate generation, sound failures to an embedding reweighting head at the representation layer -- both below the search/recommendation split, so a single curator loop maintains both experiences. On 1,200 curator judgments collected over two production rounds one month apart, the combined intervention reduces rejection rate from 38.17% to 28.83% (-24.5% relative, McNemar chi-squared = 22.4, p = 2.2e-6). An accounting decomposition attributes 4.08 pp of the drop to filter-eligible categories and 5.25 pp to the rest; the deployment was unblinded and compound, so this is a production accounting bound, not a causal estimate. We present this as an industrial case study rather than a validated general method, and close with lessons for content-only discovery: the failure partition is orthogonal to the paradigm partition, and corrections land in layers shared by both paradigms.

Abstract

Computational Creativity

Explore similar work

Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation

Quality-Diversity Search in Sound Generation: Investigating Innovation Engines for Audio Exploration

Inducing Sustained Creativity and Diversity in Large Language Models