S-SPADE Done Right: Detailed Study of the Sparse Audio Declipper Algorithms

2018/09/26 by Pavel Záviška, Záviška, Pavel, Ondřej Mokrý +3
Computer Science · Engineering · Mathematics · #Advanced Adaptive Filtering Techniques #Blind Source Separation Techniques #Speech and Audio Processing #eess.AS #eess.SP #math.OC

paper · pdf · doi:10.48550/arxiv.1809.09847

arxiv created 2020/01/16 · arxiv updated 2020/01/17

Abstract

This technical report shows and discusses in detail how Sparse Audio Declipper (SPADE) algorithms are derived from the signal model using the ADMM approach. The analysis version (A-SPADE) of Kitić et. al. (LVA/ICA 2015) is derived and justified. The synthesis version (S-SPADE) of the same research team is shown to solve a different optimization task than intended. This issue is corrected in this report, leading to the new S-SPADE algorithm which is in line to A-SPADE.

Related

• APPLADE: Adjustable Plug-and-play Audio Declipper Combining DNN with Sparse Optimization
• Sparsity-based audio declipping methods: selected overview, new algorithms, and large-scale evaluation
• A modeling and algorithmic framework for (non)social (co)sparse audio restoration
• Consistent Iterative Hard Thresholding For Signal Declipping
• Benchmarks and leaderboards for sound demixing tasks