60 likes | 285 Views
Phoneme. Sentence. Word. Speech : “Track down and neutralize Terrorists.”. P(down|track). Left (). P(and|down). …. and (&). P(left|track). Track (TR). “[t]”. Down (↓). “Track[t/r/ae/k]”. P(now|down). <HMM>. P(right|track). …. turn. now. Right (). <HMM sequence>.
E N D
Phoneme Sentence Word Speech : “Track down and neutralize Terrorists.” P(down|track) Left () P(and|down) … and (&) P(left|track) Track (TR) “[t]” Down (↓) “Track[t/r/ae/k]” P(now|down) <HMM> P(right|track) … turn now Right () <HMM sequence> “[t]” “[r]” “[ae]” “[k]” . . . . . . . . . Hand signal: “TR /↓/ & / NEU / TER” <Language model> … P(quickly||down) P(here|track) come 1st StateT 2nd StateT 3rd StateT … here quickly Direction: ↓ “&” “↓” “TR” “T” P(there|track) there Isolated symbol Connected symbol Sentence “T” “R” “Track down and neutralize terrorists.” “neutralize” “and” “down” “terrorists” “track”
…… …… HMM WORD # M …… HMM UG # 2 …… HMM UG # N HMM UG # 2 UG HMM # 3 UG HMM # 3 UG HMM # 3 UG HMM # 3 …… …… WORD HMM # 1 WORD HMM # M WORD HMM # 2 …… UG training DB based on moveme …… …… UG HMM # 11 UG HMM # 11 UG HMM # 1 UG HMM # 2 UG HMM # 11 UG HMM # N UG HMM # 11
Word network Image sequence P(down|track) Left () P(and|down) … and (&) UG HMM # 3 UG HMM # 3 UG HMM # 3 UG HMM # 3 P(left|track) Track (TR) Down (↓) • Pre-processing • Tracking hand • Feature extraction • Recognition • Decoding word sequence P(now|down) P(right|track) … turn now Right () . . . . . . . . . P(quickly||down) P(here|track) come … here quickly …… …… WORD HMM # M WORD HMM # 2 WORD HMM # 1 …… P(there|track) UG training DB based on moveme there …… …… UG HMM # N UG HMM # 1 UG HMM # 11 UG HMM # 11 UG HMM # 11 UG HMM # 2 UG HMM # 11
Recognized command: COME ME QUIET … Word network WD01 WD23 WD09 … Image sequence P(down|track) Left () P(and|down) … and (&) UG HMM # 3 UG HMM # 3 UG HMM # 3 UG HMM # 3 P(left|track) Track (TR) Down (↓) • Pre-processing • Tracking hand • Feature extraction • Recognition • Decoding word sequence P(now|down) P(right|track) … turn now Right () . . . . . . . . . P(quickly||down) P(here|track) come … here quickly …… …… WORD HMM # 1 WORD HMM # M WORD HMM # 2 …… P(there|track) there Word description UG training DB based on moveme …… …… UG HMM # 11 UG HMM # 11 UG HMM # 1 UG HMM # 11 UG HMM # 2 UG HMM # N UG HMM # 11
Recognized command: COME ME QUIET … Word network WD01 WD23 WD09 … P(down|track) Left () P(and|down) … and (&) P(left|track) Image sequence Track (TR) Down (↓) P(now|down) P(right|track) … turn now Right () . . . . . . . . . • Pre-processing • Tracking hand • Feature extraction • Recognition • Decoding word sequence P(quickly||down) P(here|track) come … here quickly P(there|track) there …… …… WORD HMM # M …… UG HMM # N Word description UG training DB based on moveme UG HMM # 3 UG HMM # 11 WORD HMM # 2 UG HMM # 3 UG HMM # 11 UG HMM # 2 WORD HMM # 1 UG HMM # 3 UG HMM # 11 UG HMM # 1 UG HMM # 3 UG HMM # 11 …… ……
Recognized command: COME ME QUIET … Word network WD01 WD23 WD09 … P(down|track) Left () P(and|down) … and (&) P(left|track) Image sequence Track (TR) Down (↓) P(now|down) P(right|track) … turn now Right () . . . . . . . . . • Pre-processing • Tracking hand • Feature extraction • Recognition • Decoding word sequence P(quickly||down) P(here|track) come … here quickly P(there|track) there …… …… WORD HMM # M …… UG HMM # N Word description UG training DB based on moveme UG HMM # 3 UG HMM # 11 WORD HMM # 2 UG HMM # 3 UG HMM # 11 UG HMM # 2 WORD HMM # 1 UG HMM # 3 UG HMM # 11 UG HMM # 1 UG HMM # 3 UG HMM # 11 …… ……