Pattern recognition device, pattern recognition method and program

摘要: This pattern recognition device converts an inputted signal into a feature vector, and performs on the by matching said vector against dictionary. A dictionary (10) comprises subspace basis which represents of dimension less than dimensional number multiple probabilization parameters for converting degree similarity calculated from likelihood. The is provided with unit (3) calculates means quadratic polynomial value inner product likelihood exponential function linear sum parameters. In (10), learning performed expectation-maximization algorithm that utilizes constraint conditions between