Procedure for the recognition of a keyword in spoken language, a) in which the keyword W is represented by a sequence of states of Hidden-Markov models (101) b) in which the spoken language is sampled with a frequency predetermined and at each sampling time t is created (102), from the spoken language, a characteristic vector Ot for a speech signal, which belongs to the sampling instant t, c) in which a sequence O of vectors is projected of characteristics Ot by means of a Viterbi algorithm on the sequence of states, being calculated (103) a measure of local confidence for a state with the help of an emission measure, d) in which the Viterbi algorithm supplies (104 ) a measure of global trust, e) in which the keyword is recognized in the spoken language, when C (W, O) T, (105) where C () designates the measure of trust is applied, W designates the keyword, represented as a sequence of states os, O designates the sequence of feature vectors Ot, T designates a predetermined threshold value. f) in which, in another case, the keyword is not recognized in the spoken language.
展开▼
