Grammar generation for voice-based searches

• US 6,973,429 B2
• Filed: 12/04/2000
• Issued: 12/06/2005
• Est. Priority Date: 12/04/2000
• Status: Active Grant

• Current Assignee: A9.com Incorporated (Amazon.com, Inc.)
• Original Assignee: A9.com Incorporated (Amazon.com, Inc.)
• Inventors: Smith, Kenneth
• Primary Examiner(s): McFadden, Susan
• Assistant Examiner(s): Vo, Huyen X.
• Application Number:
  US09/729,646
• Publication Number:
  US 20020069059A1
• Time in Patent Office: 1,828 Days
• Field of Search: 704/257, 704/270.1, 704/9, 704/245, 704/2, 704/260, 707/4
• US Class Current: 704/257
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks

Abstract

A grammar generation process generates a speech recognition grammar for interpreting search queries or a domain of items. The grammar comprises both single-term and multi-term utterances derived from the texts of the items (preferably the item titles). The utterances are derived in-part by expanding phrases selected from the item text into their individual terms plus all forward combinations of such terms. The forward combinations and individual terms that are deemed not useful to the search process are filtered out of the grammar. The process tends to produce a grammar containing the utterances that are most likely to occur within voice queries for the items, while maintaining a grammar size that is sufficiently small to provide reliable speech recognition.

Claims

1. A method of specifying to a speech recognition system a set of valid utterances for interpreting voice-based queries for items within a domain of items, the method comprising:

extracting phrases from at least some of the items within the domain;

expanding each phrase into a set consisting of individual terms of the phrase and forward combinations of terms within the phrase, to thereby generate a set of utterances which includes both single-term and multiple-term utterances;

incorporating at least some of the single-term and multiple-term utterances into the speech recognition grammar; and

providing the speech recognition grammar to the speech recognition system.

2. The method as defined in claim 1, wherein extracting phrases comprises extracting titles of the items.

3. The method as defined in claim 2, wherein extracting phrases further comprises dividing a title having more than a predefined number of terms into multiple phrases.

4. The method as defined in claim 2, further comprising using the speech recognition grammar and the voice recognition system to interpret of voice-based query of a title search.

5. The method as defined in claim 1, further comprising extracting individual terms from at least some of the items, and incorporating at least some of the individual terms into the grammar.

6. The method as defined in claim 1, wherein incorporating at least some of the single-term and multiple-term utterances into the speech recognition grammar comprises removing at least some of the utterances according to a set of heuristics.

7. The method as defined in claim 1, further comprising converting a number within a multiple-term utterance produced by said phrase expansion into a multiple-word counterpart.

8. The method as defined in claim 1, further comprising expanding an acronym within a multiple-term utterance produced by said phrase expansion into a multiple-term counterpart.

9. A method of enabling an item to be located by a voice-based search query, the method comprising:

extracting a phrase from text of the item;

translating the phrase into a set of utterances consisting of (a) individual terms of the phrase, and (b) all ordered combinations of two or more consecutive terms of the phrase; and

storing at least some of the utterances of the set, including both single-term and multi-term utterances, within a speech recognition grammar used to interpret the voice-based search query.

10. The method as defined in claim 9, wherein extracting a phrase comprises extracting the phrase from a title of the item.

11. The method as defined in claim 9, wherein storing at least some of the utterances comprises filtering out at least one utterance according to a set of heuristics.

12. The method as defined in claim 9, further comprising removing a duplicate phrase within the title prior to translation into the set of utterances.

13. The method as defined in claim 9, further comprising converting a number within the set of utterances into a word counterpart.

14. The method as defined in claim 9, further comprising expanding an acronym within the set of utterances.

15. The method as defined in claim 9, wherein the phrase comprises at least three terms.

16. A system for conducting voice based searches within a domain of items, comprising:

a speech recognition system that interprets voice search queries from users; and

a grammar which specifies to the speech recognition system valid utterances for interpreting the voice search queries, wherein the grammar comprises both single-term and multi-term utterances derived from the items within the domain, and said multi-term utterances consist primarily of forward combinations derived from phrases within text of the items, said grammar generated using a computer system that translates a phrase into a set of utterances containing (a) individual terms of the phrase, and (b) all ordered combinations of two or more consecutive terms of the phrase.

17. The system as in claim 16, wherein the forward combinations are derived from titles of the items.

18. The system as in claim 17, wherein the speech recognition system uses the grammar to interpret voice queries of title searches.