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. We present LoopLens, a research probe for loop-based music composition that visualizes audio search results to support creative foraging and assembling. Our evaluation reveals a behavioral split: experts leverage multimodal cues to exploit narrow sets of loops, while novices rely on broad auditory exploration due to limited musical vocabulary.
Introduction: Search as a Creative Act
In traditional Digital Audio Workstations (DAWs) like Ableton Live or Logic Pro, search is often treated as an auxiliary task, tucked away in side panels. However, in loop-based production, the act of searching, selecting, and juxtaposing pre-existing sounds is the composition process. LoopLens bridges the gap between discovery and creation, reframing search not as a utility for finding files, but as the primary creative act.
The LoopLens System
LoopLens is a web-based research probe that integrates a sample browser directly with a simplified DAW timeline. Key technical features include:
- Multimodal Search: Users can search via keywords, filters (BPM, Key, Instrument), or an AI-driven "Find Similar" feature.
- AI-Driven Similarity: The system uses CLAP (Contrastive Language-Audio Pretraining) embeddings to calculate distances between audio files, surfacing the most similar and dissimilar loops.
- Radial Waveforms: In addition to standard linear plots, LoopLens offers "polar plots" (circular waveforms). These are designed to highlight the cyclic, repetitive nature of musical loops.
User Study and Key Insights
A within-subject user study with 16 participants (ranging from novices to DAW experts) performed both convergent (goal-directed) and divergent (open-ended) tasks. The results highlighted how domain expertise shapes search behavior.
The Expertise Gap
- Constraint Satisfaction: For experts, search is about narrowing. Once they chose an initial track, they used filters to turn the search space into a logic puzzle where only "compatible" pieces remained. Experts were significantly faster, inserting their first loop in a median of 39.2s vs. 243.1s for novices.
- The Vocabulary Bottleneck: Novices struggled with query formulation. They often knew what they wanted to hear but lacked the technical vocabulary (e.g., "chiptune" or "high-pass filter") to find it. AI-driven "Find Similar" features acted as a vital "warm start" for these users.
Perception of AI and Visualization
- The "Curse of Knowledge": Experts were often skeptical of the AI's similarity scores. Because they perceive music across multiple dimensions (timbre, rhythm, harmony), a single scalar similarity value felt opaque and untrustworthy.
- Visualizing Rhythm: Radial waveforms were polarizing but functional. Users found them superior for quickly identifying "sparse" vs. "busy" rhythmic patterns, as the circular layout makes the "four-on-the-floor" structure of EDM visually obvious.
Conclusion and Future Work
LoopLens demonstrates that search interfaces can serve as powerful creative mediums. Future work should move away from single similarity scores toward multidimensional transparency (e.g., rhythm vs. timbre) and explore how search interfaces can serve as "training wheels" for music theory, helping users build their musical vocabulary over time.
Summary updated