Generating speech recognition grammars from a large corpus of data

• US 7,567,902 B2
• Filed: 09/18/2002
• Issued: 07/28/2009
• Est. Priority Date: 09/18/2002
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Davis, Brent L., Reich, David E.
• Primary Examiner(s): Armstrong; Angela A
• Application Number:
  US10/246,901
• Publication Number:
  US 20040054530A1
• Time in Patent Office: 2,505 Days
• Field of Search: 704/1, 704/9, 704/255, 704/257
• US Class Current: 704/251
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules

Abstract

A method of generating a speech recognition grammar for use with a speech recognition system can include parsing the corpus of data to identify grammatical structures within the corpus of data. The identified grammatical structures can be compared with grammar generation rules to determine particular ones of the identified grammatical structures to include within the speech recognition grammar. The grammar generation rules can designate which grammatical structures are to be included within the speech recognition grammar. The grammatical structures which have been identified in the parsing step and which also have been designated by the grammar generation rules can be included in the speech recognition grammar.

Claims

1. A method of generating an expandable speech recognition grammar for use with a speech recognition engine comprising:

parsing a corpus of data using a processor to generate an annotated corpus of data identifying grammatical structures and grammatical parts of speech within the corpus of data, wherein the corpus of data comprises a plurality of well formed sentences, and wherein the parsing comprises providing for each identified grammatical structure and grammatical part of speech a tag labeling each identified grammatical structure and grammatical part of speech accordingly;

using the processor to compare the identified grammatical structures and the identified grammatical parts of speech within the annotated corpus of data with grammar generation rules to designate particular ones of the identified grammatical structures and the identified grammatical parts of speech to include within a speech recognition grammar to be generated, wherein the grammar generation rules further designate, independently of a context of the corpus of data and any words already included in the expandable speech grammar, particular grammatical parts of speech and grammatical structures to be included within the expandable speech recognition grammar to be generated; and

using the processor to include within the expandable speech recognition grammar one or more words associated with the grammatical structures within the annotated corpus of data which have been identified in said parsing step and designated in said comparing step by the grammar generation rules, exclusive of words already included in the expandable speech recognition grammar, wherein the grammar is generated without use of counter-examples associated with the corpus of data.

2. The method of claim 1, further comprising adapting the processor for:

receiving a spoken query to search the corpus of data;

speech recognizing a portion of the spoken query which is specified in the expandable speech recognition grammar; and

searching the corpus of data for the recognized portion of the spoken query.

3. The method of claim 1, further comprising adapting the processor for:

identifying an additional corpus of data necessary for expansion of the expandable speech recognition grammar, wherein the additional corpus of data comprises one or more well formed sentences;

parsing the additional corpus of data to generate an annotated additional corpus of data identifying grammatical structures within the additional corpus of data;

comparing the identified grammatical structures within the annotated additional corpus of data with the grammar generation rules to designate particular ones of the identified grammatical structures within the annotated additional corpus of data to include within the expandable speech recognition grammar, wherein the grammar generation rules designate, independently of the context of the corpus of data and any words already included in the expandable speech grammar, which grammatical structures within the annotated additional corpus of data are to be included within the expandable speech recognition grammar; and

including one or more words associated with the grammatical structures within the annotated additional corpus of data within the expandable speech recognition grammar which have been identified during said parsing of the additional corpus of data and which also are designated by the grammar generation rules, exclusive of words already included in the expandable speech recognition grammar.

4. The method of claim 1, further comprising adapting the processor to generate the grammar based on a proximity of words in one of the well formed sentences of the corpus of data independent of the content of the one well formed sentence.