Business listing search

• US 7,890,326 B2
• Filed: 10/13/2006
• Issued: 02/15/2011
• Est. Priority Date: 10/13/2006
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Byrne, William J., Beaufays, Francoise, Strope, Brian
• Primary Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US11/549,486
• Publication Number:
  US 20080091435A1
• Time in Patent Office: 1,586 Days
• Field of Search: 704/251
• US Class Current: 704/251
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A method of operating a voice-enabled business directory search system includes selecting a subset of speech recognition language models from a larger set of speech recognition language models based on a type of business provided by a user, identifying weight values for the selected language models, and recognizing an identifier of a specific business in a speech input from the user based on the selected language models and the weight values.

Claims

1. A computer implemented method comprising:

selecting a subset of speech recognition language models from a larger set of speech recognition language models, the selection being based on a type of business specified in a speech input provided by a user;

selecting weight values for the selected language models based on information provided in the user speech input; and

recognizing, by a data processor, an identifier of a specific business in the speech input from the user based on the selected language models and the weight values.

2. The method of claim 1 wherein each language model is associated with a set of one or more types of businesses.

3. The method of claim 2 wherein selecting a subset of speech recognition language models comprises selecting a language model that is associated with a set of one or more types of businesses that are the most similar to the type of business provided by the user.

4. The method of claim 1 wherein each of some of the language models is associated with a geographical location, and the selecting of a subset of language models is also based on a geographical location provided by the user.

5. The method of claim 4 wherein selecting a subset of speech recognition language models comprises selecting a generic language model that is not associated with a particular geographical location and a location-specific language model that is associated with a particular geographical location.

6. The method of claim 5 wherein the generic language model and the location-specific language model are both associated with the same one or more types of businesses.

7. The method of claim 1 wherein selecting weight values comprises retrieving pre-computed weight values.

8. The method of claim 1, further comprising identifying one or more machines for recognizing the identifier of the specific business, each of the machines being associated with one or more language models.

9. The method of claim 1, further comprising evaluating a hierarchical tree of nodes, each node associated with one of the speech recognition language models.

10. The method of claim 9, comprising identifying a particular node in the hierarchical tree that is most relevant to the specific business, and wherein selecting weight values comprises assigning higher weight values to language models associated with ancestor nodes that are closer to the particular node than language models associated with ancestor nodes that are farther away from the particular node.

11. The method of claim 1 in which selecting weight values comprises assigning a higher weight value to a location specific language model than a generic language model.

12. The method of claim 1 in which selecting weight values comprises assigning a higher weight value to a language model for recognizing businesses in a geographical location closer to a particular geographical location provided by the user than language models associated with geographical locations that are farther away.

13. A computer implemented method comprising:

establishing speech recognition language models, each of some language models being associated with a geographical location and a set of types of businesses, each set comprising at least one type of business, each type of business being associated with one or more specific businesses; and

using, by a computer, one or more of the speech recognition language models to recognize an identifier of a specific business based on the geographical location and the type of business;

wherein a language model associated with a geographical location is biased towards recognizing specific businesses located in the geographical location relative to another language model associated with another geographical location, and

a language model associated with a set of types of businesses is biased towards recognizing specific businesses associated with the set of types of businesses relative to another language model associated with another set of types of businesses.

14. The method of claim 13, further comprising receiving a speech input having information about a geographical location, a type of business, and a specific business.

15. The method of claim 14, further comprising selecting one or more speech recognition language models for use in recognizing the specific business based on similarities between the types of businesses associated with the language models and the type of business in the speech input.

16. The method of claim 15 wherein the selection of the one or more language models is also based on a similarity or proximity between the geographical location associated with the language model and the geographical location in the speech input.

17. The method of claim 13 wherein the language models comprises higher-level language models each being constructed from a combination of lower-level language models.

18. The method of claim 17, further comprising storing values used in the combinations to enable subsequent recognition of identifiers of specific businesses to use the stored values to construct the higher-level language models from the lower-level language models without re-computing the values.

19. The method of claim 17 wherein some higher-level language models are each constructed from a combination of at least two lower-level language models that are associated with different geographical locations.

20. The method of claim 17 wherein some higher-level language models are each constructed from a combination of at least two lower-level language models that are associated with different sets of types of businesses.

21. An apparatus comprising:

one or more storage devices to store speech recognition language models;

a mapping module to select a subset of the language models based on a type of business specified in a speech input provided by a user and select weight values for the selected language models based on information provided in the user speech input; and

a speech recognition engine to recognize an identifier of a specific business in the speech input provided by the user based on the selected language models and the weight values.

22. The apparatus of claim 21 wherein each language model is associated with a set of types of businesses, each set comprising at least one type of business.

23. The apparatus of claim 22 wherein the mapping module selects a language model that is associated with a set of types of businesses that are the most similar to the type of business provided by the user.

24. The apparatus of claim 23 wherein each of some language models is associated with a geographical location, and the mapping module selects the subset of language models also based on a geographical location provided by the user.

25. The apparatus of claim 21 wherein selecting weight values comprises at least one of computing the weight values and retrieving pre-computed weight values.

26. The apparatus of claim 21 wherein the speech recognition engine comprises a distributed speech recognition engine running on a plurality of machines, each component being associated with one or more language models and the weight values.

27. An apparatus comprising:

a storage device to store speech recognition language models, each language model being associated with a geographical location and a set of types of businesses, each set comprising at least one type of business, each type of business being associated with one or more specific businesses; and

a speech recognition engine to use one or more of the speech recognition language models to recognize an identifier of a specific business based on the geographical location and the type of business;

wherein a language model associated with a geographical location is biased towards recognizing specific businesses located in the geographical location relative to another language model associated with another geographical location, and

a language model associated with a set of types of businesses is biased towards recognizing specific businesses associated with the set of types of businesses relative to another language model associated with another set of types of businesses.

28. An apparatus comprising:

a storage device to store speech recognition language models;

means for selecting a subset of speech recognition language models from a larger set of speech recognition language models, the selection being based on a type of business specified in a speech input provided by a user,

selecting weight values for the selected language models based on information provided in the user speech input, and

recognizing an identifier of a specific business in the speech input from the user based on the selected language models and the weight values.