Method for speaker verification using at least one speech signal spoken by a speaker, by a computer

摘要

The method involves using a computer (R) for the verification of a speaker (SPR) by at least one a speech signal (SPS). At least one characteristic vector (MV) is formed (101) for the speech signal, which is compared (102) to at least one reference vector (RMV), which was formed from at least one speech signal of a speaker to be verified. The characteristic vector is compared (103) with at least one opposite vector (GMV), which was formed from at least one speech signal of a further speaker, who is not the speaker to be verified. A similarity value is formed (104) from the comparisons, with which a similarity between the characteristic vector and the reference vector and a similarity between characteristic vector and the opposite vector is described. The speaker is classified (105) as the speaker to be verified if the similarity value deviates within a pre-settable value around a pre-settable area.