| 摘要 |
The present invention relates to a method for measuring the resemblance between speech samples comprising a learning phase and a test phase each including a digital acquisition and processing step, an order p acoustic analysis step, and a step for computing the covariance matrix of size p x p of the resultant vector signal, the learning phase moreover comprising a step for inverting the reference covariance matrix X and a step for storing the matrix X<-1> thus obtained in a dictionary, the test phase comprising a step for multiplying the covariance matrix for the test Y and the reference matrix X<-1>, a step for extracting all or some of the p eigenvalues lambda k of this matrix product, and a computation step using a family of functions f, dubbed generalised sphericity functions, so as to obtain a numerical value measuring the resemblance beween the relevant test sample and the reference samples from the dictionary. The invention also relates to an apparatus for implementing this method. <IMAGE> |
