Open AccessPerceptual orthogonal matching pursuit for speech sparse modelling
Author(s) -
Bouchhima B.,
Amara R.,
Turki HadjAlouane M.
Publication year - 2017
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2017.1608
Subject(s) - matching pursuit , weighting , computer science , speech recognition , coding (social sciences) , pattern recognition (psychology) , filter (signal processing) , artificial intelligence , perception , matching (statistics) , neural coding , speech coding , speech processing , computer vision , mathematics , medicine , statistics , compressed sensing , neuroscience , biology , radiology
The perceptual orthogonal matching pursuit (POMP), a sparse approximation algorithm built upon the known orthogonal matching pursuit (OMP), is introduced. It is designed for speech processing and can be of great use in speech coding applications. It can handle all types of real dictionaries, including predefined and adaptive dictionaries. Being a suboptimal method, POMP performs a series of local updates where it minimises a perceptual distortion measure involving a perceptual weighting filter. This filter is tailored for speech signals and is used in AMR 3GPP coders. Experiments show that POMP outperforms the standard OMP for predefined and adaptive dictionaries.