• Front end
  • base features: filter bank amplitudes, cepstral coefficients, delta and acceleration coefficients
    
  • statistically-derived features: PCA, LDA, HLDA
    
  • prosodics, multi-modal features (lip trackers)
    
  
  
• Acoustic Model
  • structural model: markov models, neural nets
    
  • probabilistic model: gaussian, laplacian, support vector machines
    
  • continuous and semi-continuous; tied and untied
    
  
  
• Language Model
  • context free grammars, N-gram LMs, long-distance N-grams
    
  • dynamically switched LMs, cache LMs
    
  
  
• Decoder
  • depth-first (stack) search, breadth-first (viterbi) search
    
  • network-based (full expansion), dynamic (run-time expansion)
    
  • optimizations - data structures, pruning, two-pass decoding