PROCEDURE FOR DETERMINING THE PROBABILITY OF THE APPEARANCE OF A SEQUENCE OF AT LEAST TWO WORDS DURING A VOICE RECOGNITION.

摘要

Procedure to determine the probability of the appearance of a sequence of at least two words during a speech recognition by means of a computer, a) in which a language presents linguistic classes: (C1, ..., Ck) = F ((f1, v11, ..., v1j) ... (fm, vm1, ..., vmj)), where f1, ... fm designates a linguistic characteristic, m designates the number of characteristics linguistics, vm1 ... vmj designate the linguistic values of the linguistic characteristic fm, j designates the number of linguistic values, C1, ..., Ck designates the linguistic classes, k designates The number of language classes, F designates a representation specification (classifier) of linguistic characteristics and linguistic values on language classes. b) in which a word is assigned at least one of the linguistic classes; c) in which the probability of the appearance of the sequence of at least two words is determined by: P (W) ni = 1 P Ci P Ci-1 P (wijCi) xP (Ci j Ci-1) xP (Ci-1 j wi-1) where P (W) designates the probability of the occurrence of the sequence of at least two words W designates the sequence of at least two words wi designates the word i of the sequence W with (i = 1..n), n designates the number of the words wi of the sequence W, Ci designates a linguistic class C, which belongs to a word wi, Ci-1 designates a linguistic class, which belongs to a word wi-1, Ci designates the sum of all linguistic classes C, which belong to a word wi, P (wijCi) designates the conditional probability of the word, P (Ci j Ci-1) designates the probability of bigramas ( also: bigramas of classes probability of bigramas) P (Ci-1 j wi-1) designates the conditioned probability of the classes.