METHOD FOR MODIFYING ERROR OF SPEECH RECOGNITION AND APPARATUS FOR PERFORMING THE METHOD

摘要

A method for modifying a speech recognition error and a device thereof are disclosed. The method for modifying a speech recognition error comprises the following steps of: performing speech recognition about a user speech to convert the user speech into a text; detecting an error word in the text by using speech part information and pronunciation information about each word configuring the text; generating a first word alignment pattern including an error word in the text and matching the first word alignment pattern with a second word alignment pattern, which is extracted from a pre-built word alignment model, to modify the error word included in the text; and modifying the error word included in the text modified based on a pre-built self neural network model. Therefore, the present invention improves a performance of speech recognition and is easily applied to a variety of systems used for speech recognition.