Signal-Informed Temporal Routing for Vinyl Defect Regime Detection

Yi-Hung Kan; Homayoon Beigi

eess.ASSep 19, 2026

Signal-Informed Temporal Routing for Vinyl Defect Regime Detection

Authors: Yi-Hung Kan, Homayoon Beigi

Organizations: Nonlinear Control Laboratory, Dept. of Electrical Engineering, Dept. of Mechanical Engineering, Columbia University, New York, NY, USA · Recognition Technologies, Inc., South Salem, NY, USA

Abstract

Vinyl restoration systems must distinguish isolated clicks, short bursts, dense crackle, and overlapping damage before selecting a repair operation. We present a lightweight two-stage detector in which signal-informed sparse, burst, and dense experts produce complementary defect evidence, and a temporal backend converts that evidence into stable repair regimes. The backend factors the five-way decision hierarchically, applies a validation-only mixed regime gate, and decodes with validation-selected transition penalties that outperform a maximum-likelihood transition matrix on the same emissions. On a source-separated synthetic benchmark of 597 non-overlapping 15 s excerpts drawn from 21 recordings, the held-out system reaches 0.730 pooled five-class macro F1 (95% CI 0.694-0.761), an improvement of 0.043 over a flat frame-level router (paired bootstrap p=0.001). Clean and dense frames are highly reliable, while sparse, burst, and mixed frames remain far more ambiguous.

Figures & tables

Defect recipe	Sparse F1	Burst F1	Dense F1
Narrow clicks	0.605	0.000	0.000
Wide pops	0.927	0.899	0.000
Short scratches	0.000	0.634	0.000
Dense crackle	0.000	0.000	0.986
Mixed damage	0.577	0.378	0.975

TABLE I: Frozen expert results on held-out TEST. Sparse F1 is candidate/event level. Burst and dense F1 are frame level. A zero entry means the recipe contains no positives of that type.

	Five-class				4-cl.
	Macro	Wtd.	Micro	Mixed	Macro
System	F1	F1	F1	F1	F1	Trans.
Front-end evidence rule	0.464	0.926	0.895	0.051	0.597	14725
Flat five-class router	0.687	0.980	0.978	0.413	0.768	2439
+ hierarchy and mixed gate	0.702	0.982	0.981	0.335	0.775	2419
+ temporal consistency	0.713	0.983	0.982	0.367	0.785	2119

TABLE II: Held-out TEST performance under both taxonomies, cumulative backend stages, all with frozen expert evidence and validation-only selection. Final row 95% bootstrap CI: 0.694–0.761 five-class, 0.777–0.829 four-class.

Regime	Precision	Recall	F1	Support
Clean	0.995	0.987	0.991	92788
Sparse	0.599	0.534	0.565	442
Burst	0.627	0.770	0.691	478
Dense	0.973	0.987	0.980	46465
Mixed	0.413	0.432	0.422	437

TABLE III: Final Viterbi per-class metrics on held-out TEST.

Fig. 1: Row-normalized held-out TEST confusion matrix for the final Viterbi-decoded router. Clean and dense regimes are reliable. Mixed errors are dominated by confusion with dense, while sparse and burst remain partially ambiguous.

Source recording	Cases	5-class	4-class
Chopin, Ballade No. 4	50	0.712	0.803
Rachmaninoff, Piano Concerto No. 2	47	0.765	0.819
Fitzgerald and Armstrong, vocal jazz	12	0.656	0.740
Pooled	109	0.730	0.805

TABLE IV: Final system by held-out source recording.

Fault Detection

Fault detection identifies abnormal behavior in machines, sensors, infrastructure, software, and autonomous systems before it causes harm or disruption.

See all 105 papers

Restore, Separate, Restore: A Modular Framework for Music Source Restoration
Oct 6, 2026Tobias Morocutti, Emmanouil Karystinaios, Gerhard WidmerAudio Source Separation Music Source Separation
Institute of Computational Perception, Johannes Kepler University Linz, Austria · LIT Artificial Intelligence Lab, Linz, Austria
Multi-Stage Music Source Restoration with BandSplit-RoFormer Separation and HiFi++ GAN
Mar 4, 2026Tobias Morocutti, Emmanouil Karystinaios, Jonathan Greif +1Music Source Separation Audio Source Separation
Institute of Computational Perception (CP-JKU) Johannes Kepler University Linz, Austria
DTT-BSR+: A Generative-Regression Cascade for Music Source Restoration
Jun 23, 2026Youran Ni, Shihong Tan, Yuzhu Wang +1Audio Source Separation Music Source Separation
School of Electronic Information, Wuhan University, China · Signal Processing Research Center, Tampere University, Tampere, Finland

Oct 6, 2026J/K move · Enter open · S save

Restore, Separate, Restore: A Modular Framework for Music Source Restoration

Tobias Morocutti, Emmanouil Karystinaios, Gerhard Widmer

Institute of Computational Perception, Johannes Kepler University Linz, Austria · LIT Artificial Intelligence Lab, Linz, Austria

Audio Source Separation Music Source Separation

Open paper AskPDF arXiv

Music Source Restoration (MSR) seeks to recover original, unprocessed instrument stems from mixed, mastered, and possibly degraded recordings. Unlike conventional source separation, which treats the mixture as a linear sum of clean sources, MSR must additionally invert nonlinear production effects, such as equalization and compression, and transmission-related degradations, such as codec artifacts. We propose a three-stage framework built around this distinction: (1) a mixture restoration model that addresses degradation before separation, (2) a single model that separates the restored mixture into eight target stems (vocals, guitars, keyboards, synthesizers, bass, drums, percussion, and orchestra), and (3) stem-specific restoration experts fine-tuned on the separator's own residual artifacts. Each stage improves restoration quality over the previous one on the MSR Challenge test set. We release code and models to support future research in MSR at https://github.com/theMoro/music_source_restoration.

Abstract

Fault Detection

Explore similar work

Restore, Separate, Restore: A Modular Framework for Music Source Restoration

Multi-Stage Music Source Restoration with BandSplit-RoFormer Separation and HiFi++ GAN

DTT-BSR+: A Generative-Regression Cascade for Music Source Restoration