Large-vocabulary speech recognition using an integrated syntactic and semantic statistical language model
First Claim
1. A speech recognition system, comprising:
- a pre-processor receiving an acoustic signal and processing the acoustic signal to produce an acoustic feature sequence; and
a recognition processor receiving the acoustic feature sequence and processing the acoustic feature sequence using a multiple-span stochastic language model to form a linguistic message, wherein the multiple-span stochastic language model includes a local span providing an immediate word context and a large span providing a global word context.
2 Assignments
0 Petitions
Accused Products
Abstract
Methods and apparatus for performing large-vocabulary speech recognition employing an integrated syntactic and semantic statistical language model. In an exemplary embodiment, a stochastic language model is developed using a hybrid paradigm in which latent semantic analysis is combined with, and subordinated to, a conventional n-gram paradigm. The hybrid paradigm provides an estimate of the likelihood that a particular word, chosen from an underlying vocabulary will occur given a prevailing contextual history. The estimate is computed as a conditional probability that a word will occur given an "integrated" history combining an n-word, syntactic-type history with a semantic-type history based on a much larger contextual framework. Thus, the exemplary embodiment seamlessly blends local language structures with global usage patterns to provide, in a single language model, the proficiency of a short-horizon, syntactic model with the large-span effectiveness of semantic analysis.
-
Citations
45 Claims
-
1. A speech recognition system, comprising:
-
a pre-processor receiving an acoustic signal and processing the acoustic signal to produce an acoustic feature sequence; and a recognition processor receiving the acoustic feature sequence and processing the acoustic feature sequence using a multiple-span stochastic language model to form a linguistic message, wherein the multiple-span stochastic language model includes a local span providing an immediate word context and a large span providing a global word context. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A method for performing speech recognition, comprising the steps of:
-
receiving an acoustic signal; processing the acoustic signal to form an acoustic feature sequence; and processing the acoustic feature sequence using a multiple-span stochastic language model to form a linguistic message, wherein the multiple-span stochastic language model includes a local span providing an immediate word context and a large span providing a global word context. - View Dependent Claims (20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36)
-
-
37. A method for deriving a multiple-span stochastic language model, comprising the steps of:
-
reading at least one training document; using an n-gram paradigm to compute a set of local a priori probabilities based on said at least one training document; using a latent semantic paradigm to compute a set of global a priori probabilities based on said at least one training document; and combining the local a priori probabilities and the global a priori probabilities to provide the multiple-span stochastic language model. - View Dependent Claims (38, 39)
-
-
40. A computer readable medium containing program instructions for deriving a multiple-span stochastic language model, the program instructions including instructions for:
-
reading at least one training document; computing a set of local a priori probabilities, based on said at least one training document, using an n-gram paradigm; computing a set of global a priori probabilities, based on said at least one training document, using a latent semantic paradigm; and combining the local a priori probabilities and the global a priori probabilities to provide the multiple-span stochastic language model. - View Dependent Claims (41, 42)
-
-
43. A system for deriving a multiple-span stochastic language model, comprising:
-
at least one training document; and a processor receiving the at least one training document as input and (a) counting occurrences of local and global word sequences in said at least one training document, (b) computing a set of local a priori probabilities using an n-gram paradigm based on relative frequency counts of local word sequences in said at least one training document, (c) computing a set of global a priori probabilities using a latent semantic paradigm based on relative frequency counts of global word sequences in said at least one training document, and (d) combining the local a priori probabilities and the global a priori probabilities to provide the multiple-span stochastic language model. - View Dependent Claims (44, 45)
-
Specification