DISCUSSION: HOW DO WE APPLY
DYNAMIC PROGRAMMING TO SPEECH RECOGNITION?

• The optimal solution is only as good as the cost function. What is an appropriate cost function for speech recognition? What features should we use?
  
  
• How should we train our reference models ("templates")?
  
  
• How can we attempt continuous speech recognition using this approach?
  (Hints: Word Spotting, Level Building, Bridle Algorithm)