摘要 |
A method for reducing a voice recognition memory in a mobile terminal is provided to reduce a memory size necessary for a search space without the degradation of a recognition rate by reducing the number of arrangement generally needed in a tree-lexicon. Arrangement needed in a certain state to express a recognition object vocabulary as a tree-lexicon includes PDF_ID[] arrangement and Sibling[] arrangement. Certain state acoustic model information is stored in the PDF_ID[] arrangement. Connection information between the certain state and a next state is searched in the Sibling[] arrangement. LeftChild[] arrangement and WordID[] arrangement indicating whether the certain state is an end of a word are removed.
|