Model enrollment method for speech or speaker recognition

The voice and registration method, comprising the following steps: a (a) please user say a word; the (b) for detecting the first pronounced (354); (c) intersection the user to say the word; (d) detecting second pronounced (358); (e) for measuring the first similarity the first pronunciation and second pronunciation (362); the includes the first similarity is less than a predetermined similarity, intersection the user to say the word; (g) and detecting pronounced (366); (h) for measuring the second similarity the first pronunciation and third pronunciation (370); And ()), wherein the second similarity is larger than or equal to the similarity, generating a reference (364).