dispatch

  Finally, probability analysis, based on the chosen database, is performed and judgment whether the person is recognized is pronounced. Two different modeling techniques can be suggested for purposes of person identification or verification. The approach based on Hidden Markov Speaker Models (HMMs) with the mixtures of Gaussian distribution is text-dependent one. For this technology a speaker is described by a set of HMMs, constructing the set of speech groups spoken by that person. The speech groups may be phonemes or words [7]. The text-independent technology is rested upon Gaussian Mixture Speaker Models. For this technology each speaker is defined by only one model describing all utterances of that person. Identification in the range of both approaches is executed by evaluation of maximum a posteriori probability [7].