System and methods for using transcripts to train an automated directory assistance service

• US 7,447,636 B1
• Filed: 05/12/2005
• Issued: 11/04/2008
• Est. Priority Date: 05/12/2005
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: BBN Technologies Corporation (Rtx Corporation), Verizon Corporate Services Group Incorporated (Verizon Communications Inc.)
• Inventors: Shu, Han, Nguyen, Long, Schwartz, Richard Mark, Makhoul, John
• Primary Examiner(s): OPSASNICK, MICHAEL N
• Application Number:
  US11/129,270
• Time in Patent Office: 1,272 Days
• Field of Search: 704/4, 704/231, 704/251, 704/275, 379/88.01, 379/88.03, 379/213.01
• US Class Current: 704/275
• 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

An automated directory assistance (130) includes a training system (210). and a directory assistance system (220). The training system (210) trains and maintains the directory assistance system (220). The training system (210) includes a transcription module (310), a speech grammar estimation module (330), a listings statistics estimation module (340), and a required words determination module (350). The transcription module (310) obtains transcripts relating to directory service requests. The speech grammar estimation module (330) creates an n-gram grammar for multiple telephone numbers from the transcripts. The listings statistics estimation module (340) identifies words used to refer to each of the telephone numbers from the transcripts. The required words determination module (350) identifies at least one word that is required to request each of the telephone numbers from the transcripts. The directory assistance system (210) includes a speech recognition module (510), a listing retrieval module (520), and an accept/reject module (530). The speech recognition module (510) receives an audible request for a telephone number from a caller and generates a transcript from the audible request. The listing retrieval module (520) retrieves at least one listing corresponding to the audible request from a database using the transcript. The accept/reject module (530) determines whether to accept one or more of the listings retrieved by the listing retrieval module (520) and presents a telephone number corresponding to the accepted listing to the caller.

Claims

1. A system for generating transcripts used to configure an automated directory assistance service, comprising:

a grammar creation component configured to create a loose grammar for each of a plurality of telephone numbers using a plurality of grammar rules and a plurality of general phrases, the grammar creation component being configured to store the loose grammar including the plurality of grammar rules and the plurality of general phrases to a database, the plurality of general phrases including extraneous phrases for requesting any one of the plurality of telephone numbers; and

a speech recognition component configured to access the database and generate a transcript of a directory assistance call using the loose grammar,

2. The system of claim 1, wherein the grammar creation component is configured to create the loose grammar by identifying plausible words for referring to each of the telephone numbers.

3. The system of claim 1, wherein the speech recognition component is configured to recognize words spoken during the directory assistance call.

4. The system of claim 1, wherein the plurality of grammar rules includes alternate words for referring to the respective telephone number.

5. The system of claim 1, wherein the grammar creation component is configured to use a phone book to obtain and translate a plurality of listings corresponding with the plurality of telephone numbers.

6. The system of claim 5, wherein the plurality of grammar rules includes plausible words for referring to the listing for the telephone number corresponding to the directory assistance call.

7. The system of claim 1, wherein the directory assistance call includes a recorded request for the telephone number.

8. The system of claim 7, wherein the plurality of grammar rules includes a match to a first portion of the recorded request, and the plurality of general phrases includes a match to a second portion of the recorded request.

9. The system of claim 1, further comprising:

an accept/reject component configured to determine whether the generated transcript is acceptable.

10. The system of claim 9, wherein the accept/reject component uses confidence scores to determine whether to accept or reject the generated transcript.

11. The system of claim 9, further comprising:

a verification/correction module configured to present the generated transcript to a human for at least one of verification and modification.

12. A method for generating transcripts used to configure an automated directory assistance service, comprising:

creating a loose grammar for each of a plurality of telephone numbers using a plurality of grammar rules and a plurality of general phrases;

storing the loose grammar to a database;

generating a transcript of a directory assistance call using the loose grammar and knowledge of a telephone number that was given out by a human operator for the directory assistance call, the telephone number being used to index the loose grammar in the database;

producing a confidence score for the generated transcript; and

determining whether the generated transcript is acceptable based on the confidence score for the generated transcript.

13. The method of claim 12, wherein the creating includes:

identifying possible words or phrases for referring to each of the telephone numbers.

14. The method of claim 12, wherein the generating includes:

recognizing words spoken during the directory assistance call.

15. A computer-readable medium configured to store instructions executable by at least one processor to perform a method for generating transcripts used by an automated directory assistance service, comprising:

identifying plausible words for referring to each of a plurality of telephone numbers;

creating a loose grammar for each of the telephone numbers using the identified words and a plurality of general phrases;

storing the loose grammar to a database; and

generating a transcript of a directory assistance call using the loose grammar and a telephone number given out by a human operator for the directory assistance call, the telephone number being used to index the loose grammar in the database.

16. A system for generating transcripts used to configure an automated directory assistance service, comprising:

a grammar creation component configured to create a plurality of loose grammars for a plurality of telephone numbers using a plurality of grammar rules and a plurality of general phrases, the grammar creation component being configured to store the plurality of loose grammars including the plurality of grammar rules and the plurality of general phrases to a database, the plurality of general phrases including extraneous phrases for requesting any one of the plurality of telephone numbers; and

a speech recognition component configured to generate a transcript of a directory assistance call using one of the plurality of loose grammars.

17. The system of claim 16, wherein each of the plurality of the loose grammars is associated with a corresponding one of the plurality of telephone numbers.