Vinyl defect detector improves repair decision accuracy with two-stage signal analysis
Signal-Informed Temporal Routing for Vinyl Defect Regime Detection
Machine Learning
Summary
Vinyl restoration needs to know what kind of scratch or noise is on the record to fix it correctly. The authors created a two-step system that looks closely at the sounds to find different types of defects, then uses a smart decision process to figure out the best repair approach. Their method works better than simpler systems on a test set of music recordings. Some types of defects like clean and very dense noise are easy for the system, but others like sparse scratches remain harder to classify.
What this means in practice
- •For audio restoration engineers: Detect and classify different types of vinyl record defects to guide automated repair tools in audio restoration workflows.
- •For digital archiving teams: Improve the quality control process by automatically identifying damage and noise regimes in digital transfers of vinyl records.
Authors
Yi-Hung Kan, Homayoon Beigi
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.