SPEAKER-INDEPENDENT VARIABLE-WORD KEYWORD SPOTTING SYSTEM INCLUDING GARBAGE MODELING UNIT USING DECISION TREE-BASED STATE CLUSTERING AND METHOD THEREOF

摘要

A speaker-independent variable-word keyword spotting system including a garbage modeling unit using decision tree-based state clustering and a method thereof are provided to replace a garbage modeling unit of an existing keyword spotting system with a garbage model unit having a superior function for removing a garbage part and a background noise part, thereby easily improving performance of the keyword spotting system. A pre-processor(100) detects an inputted voice section and extracts characteristics of the detected voice. A variable-word keyword recognizing system(200) uses the voice characteristics from the pre-processor and detects a keyword through a keyword model unit(240), a garbage model unit(260) and a tying model unit(265). The garbage modeling unit is a garbage modeling unit using decision tree-based state clustering using both inter-phoneme acoustic information and statistical information.