THE METHODOLOGY TO IMPLEMENT A SYSTEM CONVERTING VOICE INPUT TO TEXT STRING BASED ON BUILDING SCORE MATRIX AND SEARCHING BEST PATH OF PLU SEQUENCE FOR VOICE RECOGNITION BASED ON ANALYZING A PHONEME UNIT

摘要

PURPOSE: An implementation method of a device for converting a voice input into a text string and a construction method of a score matrix are provided to convert characters based on PLUs and to detect standard data which matches with input voices using a score matrix. CONSTITUTION: An implementation method of a device for converting a voice input into a text string comprises the following steps: extracting input words or sentences from a start point of voice signal to an end point of voice signal in a frequency domain by PLU(phone like unit); transmitting input PLUs to a character string domain; deciding a syllable boundary by analyzing the characteristic of PLUs; arranging the input PLUs in sequence; composing the frame of score matrix by arranging standard PLUs in sequence; composing a score matrix by allocating a score to an element of the score matrix; calculating an optimal path average score by detecting an optimal path; calculating a comparative score based on the difference between the standard PLUs and the input PLUs; detecting an optimal path through which the standard PLUs matches with the input PLUs.