Imprint: Held Spectral Afterimages from Per-Bin Adaptive Hysteresis

IMPRINT is a spectral effect, designed for real-time use, in which loud spectral events leave a held afterimage. The effect then transforms that image with rules taken from the auditory-memory literature. The core decision, whether an STFT bin holds an image, is a per-bin adaptive-hysteresis state machine with local propagation (B5). B5 comes from the frozen Spectral Latch comparison against a coupled bistable field. On six drum test files B5 agreed with the field at an audible-weighted IoU of 0.937 (imprint setting) and 0.744 (freeze setting), at about 24 instead of 456 operations per bin and frame. The simpler B4 agreed as well, and no baseline met the pre-registered 0.95 criterion. B5 was chosen for its release control, not for a measured advantage. The production engine reproduces the reference B5 maps exactly: 3 of the 6 test files × 2 settings, 0 differing frame-bins. On this core sit three renderings of auditory memory: AGE (pitch blur growing with age, levels forgotten first, a colour path into the complement), GIST (the summary statistics of the last two bars) and MOTION (representational momentum or the spectral motion aftereffect). DISINTEGRATE adds an artistic decaying loop. A TEXTURE switch, six forms and JOURNEY are reported as conventional sound-design additions. On the author’s Intel Core i9-12900H at 48 kHz stereo with 512-sample blocks, the engine used 2.8 % of one core with every layer off and 14.8–17.5 % with every layer on. No deadline was missed and nothing was allocated in the audio callback. It misses deadlines at 64-sample buffers, and at 96 kHz in stereo with every layer on. Seven removed or demoted layers are documented. No listening test was run; four experiments are proposed. Files: the technical note (PDF); a reproducibility archive (ZIP) with the C++17 engine (v1.7.1), its tests (98 checks with an optional data file, 97 without), the benchmark and the author’s results, the identity verifier with its reference data, the frozen Spectral Latch records, and a SHA-256 manifest. Licences: the note, data, results and records are CC BY 4.0; the source code is under the PolyForm Noncommercial License 1.0.0 (see LICENSE-CODE.md in the archive).

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-07
DOI
https://doi.org/10.5281/zenodo.24041327
Primary Topic
Music Technology and Sound Studies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Imprint: Held Spectral Afterimages from Per-Bin Adaptive Hysteresis

Vasily Psykovsky
Zenodo (CERN European Organization for Nuclear Research)
Music Technology and Sound Studies
article

Imprint: Held Spectral Afterimages from Per-Bin Adaptive Hysteresis

Vasily Psykovsky
article en

Abstract

IMPRINT is a spectral effect, designed for real-time use, in which loud spectral events leave a held afterimage. The effect then transforms that image with rules taken from the auditory-memory literature. The core decision, whether an STFT bin holds an image, is a per-bin adaptive-hysteresis state machine with local propagation (B5). B5 comes from the frozen Spectral Latch comparison against a coupled bistable field. On six drum test files B5 agreed with the field at an audible-weighted IoU of 0.937 (imprint setting) and 0.744 (freeze setting), at about 24 instead of 456 operations per bin and frame. The simpler B4 agreed as well, and no baseline met the pre-registered 0.95 criterion. B5 was chosen for its release control, not for a measured advantage. The production engine reproduces the reference B5 maps exactly: 3 of the 6 test files × 2 settings, 0 differing frame-bins. On this core sit three renderings of auditory memory: AGE (pitch blur growing with age, levels forgotten first, a colour path into the complement), GIST (the summary statistics of the last two bars) and MOTION (representational momentum or the spectral motion aftereffect). DISINTEGRATE adds an artistic decaying loop. A TEXTURE switch, six forms and JOURNEY are reported as conventional sound-design additions. On the author’s Intel Core i9-12900H at 48 kHz stereo with 512-sample blocks, the engine used 2.8 % of one core with every layer off and 14.8–17.5 % with every layer on. No deadline was missed and nothing was allocated in the audio callback. It misses deadlines at 64-sample buffers, and at 96 kHz in stereo with every layer on. Seven removed or demoted layers are documented. No listening test was run; four experiments are proposed. Files: the technical note (PDF); a reproducibility archive (ZIP) with the C++17 engine (v1.7.1), its tests (98 checks with an optional data file, 97 without), the benchmark and the author’s results, the identity verifier with its reference data, the frozen Spectral Latch records, and a SHA-256 manifest. Licences: the note, data, results and records are CC BY 4.0; the source code is under the PolyForm Noncommercial License 1.0.0 (see LICENSE-CODE.md in the archive).

Zenodo (CERN European Organization for Nuclear Research)
Openalex Percentile: Top 2%
Music Technology and Sound Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon