| 摘要 |
PURPOSE: A speaker recognition device using a voice signal and a method thereof are provided to perform calculation for a part set of a feature that a recognition ratio is the highest by extracting the feature of the voice signal while using a loss feature theory, thereby reducing the amount of calculation. CONSTITUTION: A feature extraction unit(120) converts a received voice signal into a frequency area and extracts a feature from the voice signal. A speaker matching unit(130) calculates a marginal likelihood score through selection of the number of M part sets that a recognition ratio weight is higher than a predetermined recognition ratio by applying the recognition ratio weight to the number of N part sets including the feature. The speaker matching unit recognizes a speaker who generates the voice signal based on the maximum likelihood score. [Reference numerals] (110) Voice receiving unit; (120) Feature extraction unit; (130) Speaker matching unit; (140) Device control unit; (AA) Voice input; (BB) Control signal output |
