SYSTEM AND METHOD FOR SPEEECH RECOGNITION USING ONLINE UNSUPERVISED FEATURE-SPACE ADAPTATION

摘要

PURPOSE: A voice recognition system using an online unsupervised characteristic vector adaptation method and a method thereof are provided to offer a recognition performance by using conversion parameters which convert the characteristic vector of voice data. CONSTITUTION: A voice recognition system extracts a characteristic vector from inputted voice data(410). The voice recognition system selects a GMM(Gaussian Mixture Model) which is close to the characteristic vector through a tress search method(430). The voice recognition system calculates conversion parameters by using the selected GMM(440). The voice recognition system converts the characteristic vector by using the calculated conversion parameters(450). The voice recognition system recognizes the voice of a user(460). [Reference numerals] (410) Extracting a characteristic vector from voice data; (420) Selecting the characteristic vector as much as a window size; (430) Selecting the most similar GMM through tree searching; (440) Measuring conversion parameters; (450) Converting the characteristic vector using the conversion parameters; (460) Recognizing the voice of a user; (AA) Start; (BB) End