 
 19 Nov
 1999
 
 Sphinx
 Speech Group, CMU-SCS (rkm@cs.cmu.edu)
 
  
  
  
  
  
  
 Cross-Word Triphone Modeling
 
 ·Sphinx
 uses “triphone” or “phoneme-in-context” HMMs
 ·
 ·
 ·
 ·
 ·
 ·
 ·
 ·
 ·Cross-word
 transitions use appropriate exit-model, and inject left-context into entry state 
 
  
   
  
   
  
   
  
   
   
  
   
   
  
   
   
   
  
   
   
  
   
   
  
   
   
  
   
  
   
   
   
  
   
   
   
  
   
  
   
  
   
    
    
    
   
   
   
    
   
   
  
   
   
  
   
   
  
   
   
  
   
  
   
   
  
   
   
  
   
   
  
   
  
   
   
  
   
   
  
   
   
  
   
  
   
   
  
   
   
  
   
   
  
   
  
   
   
  
   
   
  
   
   
  
   
   
   
  
   
   
  
   
   
  
   
   
  
   
   
  
   
   
  
   
  
   
   
 
 
 
 AH
 ONE
 W
 N
 Context- dependent AH 
 HMM
 Separate N 
 HMM instances for each possible right context
 
 
 
 
 
 Multiplexed W  HMM; inherited left context propagated along with path-scores, and
 dynamically
 modifies the state model (Replication is too expensive)