a Basic ldea Human Cognition -during recognition different feature components are used for different objects to be distinguished .s Pattern Recognition- different feature components for different speech recognition unit(SRU) subsets(model subsets) 4 XI >80% B B→X80% AM80% aB&c srus XX&X: Feature vectors Center of speech Technology, Tsinghua University Slide 16
Center of Speech Technology, Tsinghua University Slide 16 ❑ Basic Idea ❖ Human Cognition - during recognition, different feature components are used for different objects to be distinguished ❖ Pattern Recognition - different feature components for different speech recognition unit (SRU) subsets (model subsets) X A B C 80% X1 X2 A A B C >80% >80% A,B&C: SRUs X,X1&X2 : Feature vectors
a Definition For any feature vector X-1,x2,.,xpI, and a choosing vector W=[w1, w2,...,wpI, we define the choosing operation as WoX-XOW=x wi,x2 w2,.,xp wpl . In the figure on last page, Xi and x2 can be regarded as x1=XWi=1,2 where i∈{0,1},i=1,2,andd+=1,2,…,D If Choosing" generalized into weighting n∈[O,1,c=1,2,…,D Center of speech Technology, Tsinghua University Slide 17
Center of Speech Technology, Tsinghua University Slide 17 ❑ Definition ❖ For any feature vector X=[x1 , x2 , …, xD] T , and a choosing vector W=[w1 , w2 , …, wD] T , we define the choosing operation as WX=XW=[x1·w1 , x2·w2 , …, xD·wD] T ❖ In the figure on last page, X1 and X2 can be regarded as Xi = XWi , i=1,2, where wid{0,1}, i=1,2, and d=1,2,…,D ❖ If “Choosing” generalized into “weighting”: wd[0,1], d=1,2,…,D
a Feature Weighting Input signal> Feature extraction {X1,X2,…,Xn} 1912··9T Recognizer Y(s=RoWs s is an sru subset Center of speech Technology, Tsinghua University Slide 18
Center of Speech Technology, Tsinghua University Slide 18 { X1 , X2 , …, XT } {Y1 , Y2 , …, YT } Yt (s)=Xt Wts Recognizer Feature Extraction Input Signal s is an SRU subset ❑ Feature Weighting
口 Problems How to divide the whole sru set(or model set into subsets S? ,s How to train the weighting vector for each model subset W? The model set division should be based on a minimum classification error(MCe) criterion Center of speech Technology, Tsinghua University Slide 19
Center of Speech Technology, Tsinghua University Slide 19 ❑ Problems ❖ How to divide the whole SRU set (or model set) into subsets {s} ? ❖ How to train the weighting vector for each model subset {Ws}? ❖ The model set division should be based on a minimum classification error (MCE) criterion
E(S W: error count for model W(S: optimal weight for subset s given weight W subset s Model S Space s Sy E(SW(S)) D Mce Based model set division( ,s Goal: to find an optimal weight for set s Center of speech Technology, Tsinghua University Slide 20
Center of Speech Technology, Tsinghua University Slide 20 S E(S|W(S)) Model Space S E(S|W): error count for model subset S given weight W W(S): optimal weight for subset S ❑ MCE Based Model Set Division (0) ❖ Goal: to find an optimal weight for set S