• transform sharing provides a means for dealing with small amounts of adaptation data
  
• under this scheme, even components that are not observed in the adaptation data can be adapted
  
• a common approach is the use of a binary regression class tree as shown below
  

• the leaves of the tree are termed as the "base regression classes"
  
• each Gaussian mixture component of a model set belongs to a single base class.
• the above tree has four base classes, C₄, C₅, C₆, and C₇
  
• during adaptation, occupation counts are accumulated for each of the base classes
• the dashed circles indicate clusters which have insufficient adaptation observations. the unbroken circles have sufficient adaptation data. Sufficiency is user-defined
  
• three transforms are created:
  W₂ => C₅
  W₃ => C₆, C₇
  W₄ => C₄