Playhead Beyond The Waveform: Cross-Field Descriptors for Sound Design
Year: 2026
Type: preprint
Abstract: A waveform shows level and little else. We present Playhead, an open, browser-based system that draws calibrated feature lines over the waveform (loudness, activity, tension, harshness, beats, sections, and sound-event types) and adds nine cross-field descriptors borrowed from machine diagnostics, seismology, ecoacoustics, voice science, optimal transport, information dynamics, recurrence analysis, and impact acoustics. Each descriptor has a value fixed by mathematics for a reference signal, and the whole system is validated without listener ratings: closed-form tests, label-free metamorphic self-checks on unlabelled recordings, threshold calibration with held-out and repeated random splits, and a labelled-data test with negative controls. On ESC-50 the descriptors raise random-forest accuracy from 60.6% to 66.8% over a 130-feature hand-crafted baseline (McNemar p = 6 · 10⁻¹³), whereas noise, shuffled, or redundant columns lower it. The same analysis drives processing tools; for seven of them we measure how much of each tool’s change a static EQ, an ideal dynamic EQ, and ideal EQ automation could reproduce, and show that a psychoacoustic trigger acts on different moments than a level trigger.
Author: Sandner, Daniel
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225383935