Method for processing initially recognized speech in a speech recognition session
AssigneeInternational Business Machines Corporation
InventorsKimberlee A. Kemble, James R. Lewis, Vanessa V. Michelini, Margarita Zabolotskaya
Life of this patent
Estimating the term…
Abstract
In a speech recognition session, a method of speech processing utilizing a plurality of grammars can include a series of steps. The method can include spawning a separate thread of execution corresponding to each one of a plurality grammars and allocating a post-processing recognition task to each separate thread of execution. Each post-processing recognition task can produce a possible recognition result using the corresponding grammar. A possible recognition result produced by one of the post-processing recognition tasks can be identified.
Claims16
In a speech recognition session, a method of speech processing utilizing a plurality of grammars, said method comprising:
- spawning a separate thread of execution corresponding to each one of said plurality of grammars;
- allocating a speech recognition task to each separate thread of execution for processing a user-spoken utterance to support a pre-selected application, each said speech recognition task producing a possible recognition result using said corresponding grammar;
- assigning to each of said plurality of grammars a unique priority code based on a contextual relevance of each grammar to the pre-selected application, said unique priority code being an ordinal measure of the likelihood that a particular word accurately reflects a corresponding user-spoken utterance;
- and
- identifying a possible recognition result produced by one of said speech recognition tasks.
A machine readable storage for use in a speech recognition session, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:
- for a plurality of grammars, spawning a separate thread of execution corresponding to each one of said plurality grammars;
- allocating a speech recognition task to each separate thread of execution for processing a user-spoken utterance to support a pre-selected application, each said speech recognition task producing a possible recognition result using said corresponding grammar;
- assigning to each of said plurality of grammars a unique priority code based on a contextual relevance of each grammar to the pre-selected application, said unique priority code being an ordinal measure of the likelihood that a particular word accurately reflects a corresponding user-spoken utterance;
- and
- identifying a possible recognition result produced by one of said speech recognition tasks.
In a speech recognition session, a method of speech processing utilizing a plurality of grammars, the method comprising:
- spawning a separate thread of execution corresponding to each one of the plurality of grammars;
- allocating a speech recognition task to each separate thread of execution, each speech recognition task producing a possible recognition result based on a corresponding grammar;
- assigning a unique priority code to each of the plurality of grammars, the unique priority code being an ordinal measure of the likelihood that a particular recognition result reflects a corresponding user-spoken utterance;
- determining a confidence score for each of the possible recognition results, each confidence score corresponding to a likelihood that a possible recognition result correctly corresponds to the user-spoken utterance;
- and
- selecting a likely recognition result produced by one of the speech recognition tasks by identifying a recognition result having a highest confidence score, wherein, if a difference between any two or more of the recognition results is within a predetermined threshold, a recognition result produced using a grammar having a highest priority code.
Dependencies are inferred from the claim text.