A speakel iecognition system. using a modified f o r m of feedf o r w a r d n e u r a l n e t w o r k b a s e d on radial basis functions, is pres e n t e d . Each p e r s o n to b e recognised has their own n e u r a l I J I~~C~. t h a t is ti ained t o recognise s p e c t r a l f e a t u r e vectors r e pi ( w~i i t a t i v f~ of tlieu \peecIr a t e d heie w i t h a 40 speakel databa.se foi iifica.tioii. T h e modified iieulal s t r u c t u i c ly o u t p e i f o i u i b o t h a vcxtoi qiiantisntioil a i d a i cJ niultilayci pci i e p t i o i i c h s s i f i c~.Atlditioiidl a d v a n t a g e s incliidr e x t r r m r l y qiiic k a d a p t i v e trnins t a n d a r d niultilayri p e i c e p t r o n a p p r o a c h , a n d t h e ability to e x t r x t froin t h e trained n e t w o r k t h e s p e c t r a of those sounds t h a t c o n t r i b u t e niost t o t h e discrirnination process. m g . witti a t Irast d l 1 01 dc1 of nlngllltude llllplovelllellt OVCI tllP
COMPUTE MATCHING M C O D E M SCOllE CODEMMI: FOR FOR PERSON . . . . PERSON '3' 'w' Bcpsrbmtd work on speech andymi. and feature extraction hu tended to concmtrate on a p p l k a t h in cmnmunieatlonm and nubmatic speech recognition. H a t , rault. of an experimental study and the optkniration of featurea for a eonwationd VQ eodebook-bued automatic apeake idcntiflcation (AM) system M pramkd. Standard LPC and a paceptually weighted feature tarn14 PLP are compared wing apppriate distance namely the Iog-Wdhood, and three c e p atrd rulnt.: conatant weighting, the mot-powa-nun, and the inprne w t n c e . PLP features combinsed with a weighted ceptrd m e m e are f d to be condmtmtly the bed in a number of Mermt digitiudepmdent AS1 experiment.. armltr ~p p o r t the h y p o t h d a that the higha ovdcn of PLP (> 5) contain midcant mpeake mpeclllc WurmatAon, with AS1 pafornunee improving rapidly up to order 8 , and thm far more mlowly yet eonml.tmtly up to order 16. A dmilu p a t h i. . .~1 for codebook .ire, with fMt bnprarementm up to she 84, with more gradual g a h thae&a.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.