APPARATUS AND METHOD FOR GENERATING PSEUDOMORPHEME-BASED SPEECH RECOGNITION UNITS BY UNSUPERVISED SEGMENTATION AND MERGING

摘要

An apparatus for generating pseudomorpheme voice recognition units of the present invention includes: a sub-morpheme dividing unit which extracts smaller morpheme units from an input syntactic word unit group; and a sub-morpheme merge unit which generates pseudomorpheme units by pairing sub-morphemes with the highest frequency from the sub-morpheme units. According to the present invention, the pseudomorpheme units are generated by unsupervised segmentation and merge; a rate of words other than vocabulary is reduced by segmenting unregistered words such as a proper noun, a loan word, a compound word, which are hard to analyze with an existing morpheme analyzer, by an unsupervised method; and recognition performance can be increased by merging words with high frequencies.