Partitioning speech rules by context for speech recognition
First Claim
1. A method of continuous speech recognition including dynamic generation of a current language model and interpretation in a speech recognition system according to a current context comprising the following steps:
- a. associating with each of a plurality of speech rules, a context wherein each of said speech rules will be active;
b. during initialization of said speech recognition system, determining common contexts for said plurality of speech rules, and partitioning said plurality of speech rules into a partition of speech rule sets according to said common contexts wherein each of said plurality of speech rules resides in only one of said speech rule sets of said partition;
c. upon the detection of speech, determining said current context of said speech recognition system;
d. determining all speech rule sets which each have a context matching said current context, and storing said matched speech rule sets as a context matched set of speech rules;
e. dynamically generating a current language model from said context matched set of speech rules in said partition for use by a speech recognizer;
f. said speech recognizer using said current language model to recognize words contained in said speech; and
g. an interpreter using said context matched set of speech to interpret and perform actions according to said words recognized by said speech recognizer.
2 Assignments
0 Petitions
Accused Products
Abstract
Partitioning speech recognition rules for generation of a current language model and interpretation in a speech recognition system. Contexts for each of speech recognition rules are determined when each of the speech rules will be active. At one interval (e.g. initialization of the system), common contexts for the speech rules are determined and grouped or partitioned into speech rule sets according to these common contexts. Rapid and efficient generation of a language model upon the detection of a current context at a second interval (e.g. upon the detection of speech in one embodiment) then may be performed. Subsequent to the generation of the language model, interpretation may be performed using the speech recognition rules grouped into these common contexts.
334 Citations
18 Claims
-
1. A method of continuous speech recognition including dynamic generation of a current language model and interpretation in a speech recognition system according to a current context comprising the following steps:
-
a. associating with each of a plurality of speech rules, a context wherein each of said speech rules will be active; b. during initialization of said speech recognition system, determining common contexts for said plurality of speech rules, and partitioning said plurality of speech rules into a partition of speech rule sets according to said common contexts wherein each of said plurality of speech rules resides in only one of said speech rule sets of said partition; c. upon the detection of speech, determining said current context of said speech recognition system; d. determining all speech rule sets which each have a context matching said current context, and storing said matched speech rule sets as a context matched set of speech rules; e. dynamically generating a current language model from said context matched set of speech rules in said partition for use by a speech recognizer; f. said speech recognizer using said current language model to recognize words contained in said speech; and g. an interpreter using said context matched set of speech to interpret and perform actions according to said words recognized by said speech recognizer. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8)
-
-
9. A method of speech recognition including dynamic generation of a current language model in a speech recognition system according to a current context comprising the following steps:
-
a. associating with each of a plurality of speech rules, a context wherein each of said speech rules will be active; b. determining common contexts for said plurality of speech rules; c. partitioning each of said plurality of speech rules into a partition of speech rule sets according to said common contexts, wherein each of said plurality of speech rules resides in only one of said speech rule sets of said partition, said partition of speech rule sets being used for dynamic language model generation upon the detection of speech according to said current context, each of said speech rule sets being included in said language model only upon determining that said common context for said speech rule set matches said current context; and d. using said partition of speech rule sets to dynamically generate said language model upon said detection of said speech, and recognize said speech using said language model.
-
-
10. An apparatus for speech recognition system according to a current context comprising:
-
a. means for associating with each of a plurality of speech rules, a context wherein each of said speech rules will be active; b. means for determining common contexts for said contexts associated with each of said plurality of speech rules; c. means for partitioning each of said plurality of speech rules into a partition of speech rule sets according to said common contexts, wherein each of said plurality of speech rules resides in only one said speech rule set of said partition, said partition of speech rule sets being used for dynamic language model generation upon the detection of speech according to said current context, each of said speech rule sets being included in said language model only upon determining that said common context for said speech rule set matches said current context; and d. means for using said partition of speech rule sets to dynamically generate said language model upon said detection of said speech, and recognize said speech using said language model.
-
-
11. An apparatus for speech recognition system according to a current context comprising:
-
a. association circuitry for associating with each of a plurality of speech rules, a context wherein each of said speech rules will be active; b. context determination circuitry for determining common contexts for said contexts associated with each of said plurality of speech rules; and c. partitioning circuitry for partitioning each of said plurality of speech rules into a partition of speech rule sets according to said common contexts, wherein each of said plurality of speech rules resides in only one said speech rule set of said partition, said partition of speech rule sets being used for dynamic language model generation upon the detection of speech according to said current context, each of said speech rule sets being included in said language model only upon determining that said common context for said speech rule set matches said current context; and d. model generation and recognition circuitry for using said partition of speech rule sets to dynamically generate said language model upon said detection of said speech, and recognize said speech using said language model. - View Dependent Claims (12, 13, 14, 15, 16, 17, 18)
-
Specification