Silence as Music: Controllable and Interpretable AI for Strategic Silence Placement

Gokul Srinath Seetha Ram

Keywords: controllability, interpretability, symbolic music, silence, rests, pedagogy, performance systems, co-creative AI, music generation, human-AI collaboration

TL;DR: A controllable, interpretable system that treats silence as a first-class compositional element, enabling co-creative music generation through transparent analysis, cultural presets, and steerable parameters.

Abstract: AI music systems increasingly emphasize controllability and interpretable design. We propose a system that treats silence as a first-class compositional element and enables interactive shaping of silence placement through transparent analysis, cultural presets, and steerable controls. Our method constructs multiple candidate rest patterns from phrase boundaries, melodic tension, rhythmic heuristics, and cultural weights, then selects a mask via a quality function balancing rhythmic entropy, groove preservation, and structural coherence. We present baselines (random 10/25%, phrase-only, tension-only, weak-beats), a proxy for language model without silence prompting, and our hybrid predictor. Across four canonical melodies and three cultural presets, our approach increases rhythmic variety while preserving groove and phrase alignment relative to baselines, offering an interpretable framework for co-creative composition. We release an API, offline demos, audio examples (WAV), and a comprehensive experiment suite to support interactive composition, pedagogy, and performance.

Submission Number: 33