Progressive application of knowledge sources in multistage speech recognition
First Claim
Patent Images
1. A method to be executed at least in part in a computing device for providing multistage speech recognition, the method comprising:
- receiving an utterance from a speaker;
performing a first pass speech recognition on the utterance employing at least one algorithm including a gender detection algorithm and a feature MLLR (fMLLR) algorithm executed sequentially;
adapting the at least one algorithm based on a result of the first pass speech recognition;
performing a second pass speech recognition on the utterance employing the adapted at least one algorithm wherein at least one algorithm is adapted in a progressive manner at the second pass based on a result of the first pass of speech recognition by constraining at least one of an acoustic model and a language model associated with the at least one algorithm, wherein constraining at least one of the acoustic model and the language model associated with the algorithms includes reducing a search space for the at least one algorithm; and
employing the adapted at least one algorithm for performing speech recognition on subsequent utterances from the same speaker.
2 Assignments
0 Petitions
Accused Products
Abstract
A speech recognition system is provided with iteratively refined multiple passes through the received data to enhance the accuracy of the results by introducing constraints and adaptation from initial passes into subsequent recognition operations. The multiple passes are performed on an initial utterance received from a user. The iteratively enhanced subsequent passes are also performed on following utterances received from the user increasing an overall system efficiency and accuracy.
56 Citations
16 Claims
-
1. A method to be executed at least in part in a computing device for providing multistage speech recognition, the method comprising:
-
receiving an utterance from a speaker; performing a first pass speech recognition on the utterance employing at least one algorithm including a gender detection algorithm and a feature MLLR (fMLLR) algorithm executed sequentially; adapting the at least one algorithm based on a result of the first pass speech recognition; performing a second pass speech recognition on the utterance employing the adapted at least one algorithm wherein at least one algorithm is adapted in a progressive manner at the second pass based on a result of the first pass of speech recognition by constraining at least one of an acoustic model and a language model associated with the at least one algorithm, wherein constraining at least one of the acoustic model and the language model associated with the algorithms includes reducing a search space for the at least one algorithm; and employing the adapted at least one algorithm for performing speech recognition on subsequent utterances from the same speaker. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
-
-
10. A speech recognition system for providing multistage speech recognition, the system comprising:
a computing device executing a speech recognition application, the application including; a speech recognition engine for executing a plurality of algorithms to recognize utterances provided by a speaker, the speech recognition engine configured to; receive an utterance from a speaker; perform a first pass speech recognition on the utterance employing a selected portion of the plurality of algorithms including a gender detection algorithm and an fMLLR algorithm executed sequentially; adapt the selected portion of the plurality of algorithms based on a result of the first pass speech recognition by modifying one of;
a constraint, a number, and an order of the selected portion of the plurality of algorithms;perform subsequent passes of speech recognition on the utterance employing the adapted algorithms, wherein the algorithms are further adapted in a progressive manner at each subsequent pass based on a result of a preceding pass of speech recognition by constraining at least one of an acoustic model and a language model associated with the algorithms, wherein constraining at least one of the acoustic model and the language model associated with the algorithms includes reducing a search space for the algorithms; employ the progressively adapted algorithms for performing speech recognition on subsequent utterances from the same speaker; and perform a Viterbi search within the search space based on the progressively adapted algorithms. - View Dependent Claims (11)
-
12. A computer-readable storage medium with instructions stored thereon for providing multistage speech recognition, the instructions comprising:
-
receiving a first utterance from a speaker; performing a first pass speech recognition on the first utterance employing a plurality of algorithms including a gender detection algorithm and a feature MLLR (fMLLR) algorithm executed sequentially; creating an index of allowed words and an associated time point based on a result of the first pass speech recognition; adapting the plurality of algorithms by computing a set of allowed states in a phonetic network associated with the algorithms based on the allowed words and reducing a search space for subsequent passes based on the allowed states; performing the subsequent passes of speech recognition on the first utterance employing the adapted algorithms, wherein the algorithms are further adapted in a progressive manner by reducing the search space at each subsequent pass based on a result of a preceding pass of speech recognition by constraining at least one of an acoustic model and a language model associated with the algorithms, and wherein constraining at least one of the acoustic model and the language model associated with the algorithms includes reducing a search space for the algorithm; employing the progressively adapted algorithms for performing speech recognition on subsequent utterances from the same speaker; and
performing a Viterbi search within the search space based on the progressively adapted algorithms. - View Dependent Claims (13, 14, 15, 16)
-
Specification