Class based language model RRS feed

  • Question

  • Hi to all,

    i would like to know if it is possible to implement a mechanism similar to the so called "class based language model". What i mean is the possibility to reduce the searching space (the space of solutions, or the set of all possible recognized sentences) when two or more different grammars are combined together by exploding each grammar at runtime during the sequential analysis of the input.

    For example, if I have grammarA (with N possible recognized utterances) and grammarB (with M), i can implement the grammar0 as following:

    Rule1 -> grammarA "and" grammarB

    Rule2 -> grammarA "then" grammarB

    the set of all possible utterances of grammar0 will be of (NxM)x2. What i want to have is the possibility to first expand at runtime grammarA, then check for the world "and" or "then", and then expand the grammarB, without generating the (NxM)x2 solution and risking an "out of stack" error. I know that other SR Engines have a mechanism like this, so, is this possible in the Microsoft Speech Environment?

    Thank you!

    Monday, September 17, 2012 10:47 AM