SYSTEM FOR RECOGNIZING DYNAMIC TIME-WARPING ISOLATED WORD USING VOICED, UNVOICED, AND SILENCE SOUND INFORMATION FOUNDATION

摘要

PURPOSE: A system for recognizing a dynamic time-warping(DTW) isolated word is provided to use voiced/unvoiced/silence sound information extracted from recognition-targeted voice signals to perform a DTW algorithm, so as to reduce calculating amounts consumed in a pattern matching. CONSTITUTION: A database(10) separates voiced/unvoiced/silence sound code sections relating to each recognition-targeted voice, and classifies the voices as standard voice patterns according to code word patterns, composed of combinations of the separated code section. The database(10) stores the classified voices. A pre-emphasis processor(20) extracts first characteristic variables from an inputted original voice signal to perform a pre-emphasis process for an original voice, and extracts second characteristic variables to generate a test voice pattern of the original voice signal by using the first/the second characteristic variables, then extracts boundary information according to each voiced/unvoiced/silence sound code section from the original voice signal. A code word classifier(30) forms code words divided into voiced/unvoiced/silence sound sections by using the first/the second variables, and retrieves standard voice patterns having code word patterns corresponding to the formed code words from the database(10). A pattern matching unit(40) partially applies a dynamic time-warping(DTW) algorithm to the retrieved standard voice patterns by using the boundary information, and performs pattern matching processes to generate a recognition result.