STAR_PAK: A SIGNAL PROCESSING PACKAGE FOR ACOUSTIC-PHONETIC ANALYSIS OF SPEECH
G DUNCAN, Jonathan M. Dalby, M.A. Jack · 2024
Star_pak (Signal Transform Analysis for Recognition PAcKage) is a software suite which provides all front end signal processing requirements for the Edinburgh speech recognition system based on feature extraction techniques.The design of the package is based on processing a speech "frame", Le. a short-time segment of the order of 25.6ms, muo-is processed by an array of signal processing techniques which provide a rich acoustic description of the speech signal.The star_.pak processing strategy can be viewed as applying firstly a set of kernel transformations, such as Fourier transformations and autocorrelation.which provide a set of primary transform descriptors.Certain of the outputs from these transforms are then further combined to provide secondary descriptions of signal transform domains (see Fig. 1), such as high frequency dominance, voiced] unvoiced, etc.The resulting primary and secondary signal descriptors are stored on a frame-by-frame basis into an output file which is related to the_original input signal file and is referred to as a "star map" ("map" = Multiple Acoustic ' mmA'Vol a Part Mm) , a: