Method for the computation of an eigenspace for the representation of a plurality of training speakers

摘要

The method involves developing speaker-dependent model sets for individual training speakers. A common speaker-independent model set is developed for the training speakers using training speaker data and the independent set is adapted to each speaker to form the speaker-dependent model set by using the individual training speaker data. The method involves developing speaker-dependent model sets for individual training speakers using training speaker data, whereby the models of a set are each described by several model parameters, forming a combined model per speaker in many-dimensional vector space by chaining a number of parameters for the model set of each training speaker to form a super vector and performing a transformation to derive Eigen space base vectors involving reducing the dimension of the model space. A common speaker-independent model set is developed for the training speakers using training speaker data and the independent set is adapted to each speaker to form the speaker-dependent model set by using the individual training speaker data