Voice recognition using a grammar or N-gram procedures
View Patent ↗A method for voice recognition, wherein a diagram method with integrated unequivocal syntax restriction is combined with an N-gram voice model with statistical word sequence evaluation in such a way that alternative recognition methods can be used in different segments of a word sequence.
1. A method for recognizing speech from a word sequence, the method comprising:
applying a first recognition procedure to a first segment of the word sequence, the first segment including a plurality of first words, the first recognition procedure including a digram recognition method with integrated unique syntax;
applying a second recognition procedure to a second segment of the word sequence, the second segment including a plurality of second words, the second recognition procedure including a statistical word sequence evaluation;
combining a last two words of the plurality of first words into a pseudoword upon a change from the first recognition procedure to the second recognition procedure; and
processing the pseudoword using a digram detection method.
2. The method as recited in claim 1 wherein at least one of the first and second segments is predefined in terms of at least one of a respective segment length and segment position.
3. The method as recited in claim 2 wherein at least one of the first and second segments is permanently allocated to one of the first and the second recognition procedure.
4. The method as recited in claim 3 wherein the first segment has a predefined length and is positioned at a beginning of the word sequence.
5. The method as recited in claim 1 wherein the second segment has a predefined length and is positioned at a beginning of the word sequence.
6. The method as recited in claim 1 wherein a change from the second recognition procedure to the first recognition procedure is performed based on a respective word detection or phrase detection.
7. The method as recited in claim 6 wherein the second recognition procedure is used as standard.
8. The method as recited in claim 1 wherein the first and second segments include successive word sequence segments.
9. The method as recited in claim 1 wherein each of the first and second words include a sound sequence that includes at least one of (a) a spoken letter, (b) a syllable and (c) a syllable sequence.
10. A method for recognizing speech from a word sequence, the method comprising:
applying a first recognition procedure that includes an integrated unique syntax procedure to a first segment of the word sequence, the first segment including a plurality of first words;
applying a second recognition procedure that includes a statistical word sequence procedure to a second segment of the word sequence, the second segment including a plurality of second words;
combining a last two words of the plurality of first words into a pseudoword upon a change from the first recognition procedure to the second recognition procedure; and
processing the pseudoword using a digram detection method;
wherein the first recognition procedure is a digram recognition procedure and the second recognition procedure is a trigram recognition procedure and wherein the second recognition procedure limits permissible series of second words in the second segment according to a statistical evaluation.
11. The method as recited in claim 10 wherein the applying the second recognition procedure includes:
recognizing a word triplet, the word triplet including three second words of the plurality of the second words; and
representing the word triplet as a pseudoword doublet, the pseudoword doublet including a second and a third pseudoword, the second pseudoword overlapping with the third pseudoword and each of the second and third pseudowords including two of the three second words of the word triplet.