The invention discloses a method for the voice recognition of a speaker using a q-order predictive model, comprising a step for extracting the statistical characteristics including a step of digital acquisition of the speaker's voice sample, corresponding to one or several utterances, a step of converting this voice sample into a sequence of vectors of size p, obtained from a series of analysis windows of average size T and an average spacing I, and a step of determining q+1 matrices from this vector sequence. The average size T is of duration of less than 10 ms and the average spacing I of a duration of less than 4.5 ms. The invention is useful in a sound lock including an electroacoustical conversion system (HP,6) and a recorded program implementation system (5) for the method.
展开▼