VOICE RECOGNITION METHOD USING MACHINE LEARNING

摘要

A voice recognition method using machine learning, according to the present invention, comprises the steps of: generating voice recognition result information including rankings of information on multiple candidate sentences and each candidate sentence by receiving voice signals; re-ranking by changing the ranking of information on multiple candidate sentences by processing, according to a machine learning model pre-learned, the information on multiple candidate sentences included in the voice recognition result information; and selecting top-ranked candidate sentence information among the information on multiple candidate sentences re-ranked. Particularly, the machine learning mode is generated by receiving voice signals and answer sentence information, recognizing voice for the voice signals so as to generate voice recognition result information including sentence scores, which refer to the rankings of information on multiple candidate sentence and each candidate sentence, adding the answer sentence information to the voice recognition result information so as to generate answer set information, extracting feature information of the information on candidate sentences and answer sentences, included in the answer set information so as to generate learning data, and carrying out machine learning for changing the rankings of the information on candidate sentences based on the learning data and differences in the feature information between the candidate sentence information and the answer sentence information.