METHOD AND SYSTEM FOR REGISTERING VOICES AND METHOD AND SYSTEM FOR RECOGNIZING VOICES BASED ON THE SAME

摘要

PURPOSE: A method and a system for registering voices and a method and a system for recognizing voices based on the registering method and system are provided to consider the volume of the voice of a speaker when the voice of the speaker is learned for registration and the speaker is authenticated so as to enable more accurate authentication of the speaker. CONSTITUTION: The spectrum of a sound signal waveform input from the outside is analyzed(S20). A language unit previously set for speaker recognition is extracted from a voice signal waveform among the sound signal waveform(S50). The magnitude of volume of voice is measured for each of the extracted language units(S60). A voice database storing background speaker voice information including information on the volumes of a plurality of background speakers is constructed. It is judged if the volume with respect to each language unit is in a specific volume range previously set based on the voice database(S70). When the volumes of a predetermined number of language units among the language units belong to the specific volume range, each language unit is learned using a multilevel neural network(S90).