000218112 001__ 218112
000218112 005__ 20190317000427.0
000218112 020__ $$a978-1-4799-9988-0
000218112 022__ $$a1520-6149
000218112 02470 $$2ISI$$a000388373405168
000218112 037__ $$aCONF
000218112 245__ $$aExploiting Low-dimensional Structures to Enhance DNN based Acoustic Modeling in Speech Recognition
000218112 269__ $$a2016
000218112 260__ $$bIEEE$$c2016$$aNew York
000218112 300__ $$a5
000218112 336__ $$aConference Papers
000218112 490__ $$aInternational Conference on Acoustics Speech and Signal Processing ICASSP
000218112 520__ $$aWe propose to model the acoustic space of deep neural network (DNN) class-conditional posterior probabilities as a union of low- dimensional subspaces. To that end, the training posteriors are used for dictionary learning and sparse coding. Sparse representation of the test posteriors using this dictionary enables projection to the space of training data. Relying on the fact that the intrinsic di- mensions of the posterior subspaces are indeed very small and the matrix of all posteriors belonging to a class has a very low rank, we demonstrate how low-dimensional structures enable further en- hancement of the posteriors and rectify the spurious errors due to mismatch conditions. The enhanced acoustic modeling method leads to improvements in continuous speech recognition task using hybrid DNN-HMM (hidden Markov model) framework in both clean and noisy conditions, where upto 15.4% relative reduction in word error rate (WER) is achieved.
000218112 6531_ $$aSparse coding
000218112 6531_ $$aDictionary learning
000218112 6531_ $$aDeep neural network
000218112 6531_ $$aUnion of Low Dimensional Subspaces
000218112 6531_ $$aAcoustic modeling
000218112 700__ $$aDighe, Pranay
000218112 700__ $$aLuyet, Gil
000218112 700__ $$0243353$$g188259$$aAsaei, Afsaneh
000218112 700__ $$aBourlard, Hervé$$g117014$$0243348
000218112 7112_ $$cShanghai$$aProceedings of 2016 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2016)
000218112 773__ $$t2016 Ieee International Conference On Acoustics, Speech And Signal Processing Proceedings$$q5690-5694
000218112 8564_ $$uhttps://infoscience.epfl.ch/record/218112/files/Dighe_ICASSP_2016.pdf$$zn/a$$s313967$$yn/a
000218112 909C0 $$xU10381$$0252189$$pLIDIAP
000218112 909CO $$qGLOBAL_SET$$pconf$$ooai:infoscience.tind.io:218112$$pSTI
000218112 937__ $$aEPFL-CONF-218112
000218112 970__ $$aDighe_ICASSP_2016/LIDIAP
000218112 973__ $$rREVIEWED$$sPUBLISHED$$aEPFL
000218112 980__ $$aCONF