The present invention relates to a voice authentication system, in particular context request type speaker-independent authentication (Text-Prompted Speaker Independent Verification) anti on the context in which the input is applied - automatically generating a model independent authenticating the utterance voice password speaker authentication to a system and method. In speaker-independent authentication system of the present invention injeungryul is improved while the the basis of the phonetic not to require additional memory for storing the trained model generates the code data for the additional training requirements and training like a conventional speaker authentication system requires password It has the ability to automatically generate an anti-model for. The present invention is the word being entered as text data, analyzed by decomposition to a phoneme unit of this and, corresponding to find anti-phoneme in the matching look-up table model and anti-context analysis to automatically create the model unit, the utterance voice password pre-processing unit which extracts for the feature data, the feature in order to interpret the data in order to output the calculated phonemes unit Lai Clichy hood score between the voice password models and anti-models step of running the Viterbi decoder based on a hidden Markov model certification portion is configured two-part authentication step for performing a reliability test for determining whether the authentication is successful. ; Speaker authentication, type authentication context request, speaker-independent authentication, anti-model
展开▼
