FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2587459042> ?p ?o ?g. }

• W2587459042 abstract "Recently, the factorized hidden layer (FHL) adaptation method is proposed for speaker adaptation of deep neural network (DNN) acoustic models. An FHL contains a speaker-dependent (SD) transformation matrix using a linear combination of rank-1 matrices and an SD bias using a linear combination of vectors, in addition to the standard affine transformation. On the other hand, full-rank bases are used with a similar DNN adaptation method which is based on cluster adaptive training (CAT). Therefore, it is interesting to investigate the effect of the rank of the bases used for adaptation. The increase of the rank of the bases improves the speaker subspace representation, without increasing the number of learnable speaker parameters. In this work, we investigate the effect of using various ranks for the bases of the SD transformation of FHLs on Aurora 4, AMI IHM and AMI SDM tasks. Experimental results have shown that when one FHL layer is used, it is optimal to use low-ranked bases of rank-50, instead of full-rank bases. Furthermore, when multiple FHLs are used, rank-1 bases are sufficient." @default.
• W2587459042 created "2017-02-17" @default.
• W2587459042 creator A5004424826 @default.
• W2587459042 creator A5032798707 @default.
• W2587459042 date "2016-12-01" @default.
• W2587459042 modified "2023-09-25" @default.
• W2587459042 title "Low-rank bases for factorized hidden layer adaptation of DNN acoustic models" @default.
• W2587459042 cites W1492775261 @default.
• W2587459042 cites W1513820424 @default.
• W2587459042 cites W1513862252 @default.
• W2587459042 cites W1524333225 @default.
• W2587459042 cites W1537275613 @default.
• W2587459042 cites W1567690964 @default.
• W2587459042 cites W1569447338 @default.
• W2587459042 cites W1592672653 @default.
• W2587459042 cites W1599512239 @default.
• W2587459042 cites W1891007208 @default.
• W2587459042 cites W1985371235 @default.
• W2587459042 cites W1989549063 @default.
• W2587459042 cites W1993409002 @default.
• W2587459042 cites W2002342963 @default.
• W2587459042 cites W2010362084 @default.
• W2587459042 cites W2015633636 @default.
• W2587459042 cites W2056738732 @default.
• W2587459042 cites W2069631319 @default.
• W2587459042 cites W2079623482 @default.
• W2587459042 cites W2083751884 @default.
• W2587459042 cites W2094147890 @default.
• W2587459042 cites W2100969003 @default.
• W2587459042 cites W2106119541 @default.
• W2587459042 cites W2106554350 @default.
• W2587459042 cites W2117824967 @default.
• W2587459042 cites W2146871184 @default.
• W2587459042 cites W2155648385 @default.
• W2587459042 cites W2157711090 @default.
• W2587459042 cites W2160306971 @default.
• W2587459042 cites W2181607856 @default.
• W2587459042 cites W2292517305 @default.
• W2587459042 cites W2399832792 @default.
• W2587459042 cites W2399979637 @default.
• W2587459042 cites W2400957076 @default.
• W2587459042 cites W2402201734 @default.
• W2587459042 cites W2403797310 @default.
• W2587459042 cites W2508162385 @default.
• W2587459042 cites W2616139854 @default.
• W2587459042 cites W3162418253 @default.
• W2587459042 cites W330298975 @default.
• W2587459042 cites W82936479 @default.
• W2587459042 doi "https://doi.org/10.1109/slt.2016.7846332" @default.
• W2587459042 hasPublicationYear "2016" @default.
• W2587459042 type Work @default.
• W2587459042 sameAs 2587459042 @default.
• W2587459042 citedByCount "3" @default.
• W2587459042 countsByYear W25874590422017 @default.
• W2587459042 countsByYear W25874590422018 @default.
• W2587459042 crossrefType "proceedings-article" @default.
• W2587459042 hasAuthorship W2587459042A5004424826 @default.
• W2587459042 hasAuthorship W2587459042A5032798707 @default.
• W2587459042 hasConcept C104317684 @default.
• W2587459042 hasConcept C106487976 @default.
• W2587459042 hasConcept C11413529 @default.
• W2587459042 hasConcept C114614502 @default.
• W2587459042 hasConcept C120665830 @default.
• W2587459042 hasConcept C121332964 @default.
• W2587459042 hasConcept C139807058 @default.
• W2587459042 hasConcept C153180895 @default.
• W2587459042 hasConcept C154945302 @default.
• W2587459042 hasConcept C159985019 @default.
• W2587459042 hasConcept C164226766 @default.
• W2587459042 hasConcept C165443888 @default.
• W2587459042 hasConcept C17744445 @default.
• W2587459042 hasConcept C185592680 @default.
• W2587459042 hasConcept C192562407 @default.
• W2587459042 hasConcept C199539241 @default.
• W2587459042 hasConcept C202444582 @default.
• W2587459042 hasConcept C204241405 @default.
• W2587459042 hasConcept C2776359362 @default.
• W2587459042 hasConcept C28490314 @default.
• W2587459042 hasConcept C32834561 @default.
• W2587459042 hasConcept C33923547 @default.
• W2587459042 hasConcept C39920418 @default.
• W2587459042 hasConcept C41008148 @default.
• W2587459042 hasConcept C50644808 @default.
• W2587459042 hasConcept C55493867 @default.
• W2587459042 hasConcept C74650414 @default.
• W2587459042 hasConcept C92757383 @default.
• W2587459042 hasConcept C94625758 @default.
• W2587459042 hasConceptScore W2587459042C104317684 @default.
• W2587459042 hasConceptScore W2587459042C106487976 @default.
• W2587459042 hasConceptScore W2587459042C11413529 @default.
• W2587459042 hasConceptScore W2587459042C114614502 @default.
• W2587459042 hasConceptScore W2587459042C120665830 @default.
• W2587459042 hasConceptScore W2587459042C121332964 @default.
• W2587459042 hasConceptScore W2587459042C139807058 @default.
• W2587459042 hasConceptScore W2587459042C153180895 @default.
• W2587459042 hasConceptScore W2587459042C154945302 @default.
• W2587459042 hasConceptScore W2587459042C159985019 @default.
• W2587459042 hasConceptScore W2587459042C164226766 @default.
• W2587459042 hasConceptScore W2587459042C165443888 @default.
• W2587459042 hasConceptScore W2587459042C17744445 @default.

• next