Method and system for retrieving documents with spoken queries

• US 7,542,966 B2
• Filed: 02/25/2005
• Issued: 06/02/2009
• Est. Priority Date: 04/25/2002
• Status: Active Grant

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Woelfel, Joseph K., Wolf, Peter P., Ramakrishnan, Bhiksha
• Primary Examiner(s): MIZRAHI, DIANE D
• Application Number:
  US11/067,165
• Publication Number:
  US 20050149516A1
• Time in Patent Office: 1,558 Days
• Field of Search: 707 2- 5, 715/201, 715/229, 715/234
• US Class Current: 1/1
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06F 16/3343 using phonetics
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching

Abstract

A method indexes and retrieves documents stored in a database. A document feature vector is extracted from each document and the documents are then indexed according to the feature vectors. A spoken query is converted to an intermediate representation representing likelihoods of possible sequential combinations of terms in the spoken query. A query certainty vector is generated from the intermediate representation. Other information is acquired. The other information is combined with the query certainty vector. The query vector and the other information are then compared to each of the document feature vectors to retrieve a ranked result set of documents.

Claims

1. A computer implemented method for indexing and retrieving documents stored in a database, comprising the steps of:

extracting a document feature vector from each of a plurality of documents;

indexing each of the plurality of documents according the associated document feature vector;

converting, using a processor, a spoken query to an intermediate representation representing possible sequential combinations of terms in the spoken query, the intermediate representation is selected from a group consisting of a lattice of terms, an n-best list, or combination thereof;

generating a query certainty vector from the intermediate representation; and

comparing the query certainty vector to each of the document feature vectors to retrieve a ranked result set of documents.

2. The method of claim 1, further comprising:

projecting each document feature vector and the query certainty vector to a low dimension.

3. The method of claim 1, in which the intermediate representation is the lattice of terms.

4. The method of claim 1, in which the terms are words.

5. The method of claim 1, in which the terms are parts of words.

6. The method of claim 1, in which the intermediate representation is the n-best list.

7. The method of claim 1, in which the terms are phonemes.

8. The method of claim 1, in which the documents describe points of interest, each point of interest having a location, and further comprising:

determining a position of a user generating the spoken query;

determining a distance between the position of the user and each point of interest matching the spoken query to produce location information;

combining the location information with the query certainty vector; and

ranking the set of result documents according to the location information and the query certainty vector.

9. The method of claim 1, further comprising:

converting a plurality of spoken queries to a plurality of intermediate representations, the plurality of spoken queries linked by logical operators;

generating a plurality of query certainty vectors from the plurality of intermediate representations; and

comparing the plurality of query certainty vectors to each of the document feature vectors according to the logical operators to retrieve the ranked result set of documents.

10. The method of claim 1, further comprising:

converting a spoken preference to a corresponding intermediate representation;

generating a preference vector from the corresponding intermediate representation; and

ranking the set of result documents according to the preference vector.

11. The method of claim 10, in which the preference vector is weighted by weights, and further comprising:

multiplying the query certainty vector by the preference vector.

12. The method of claim 11, in which the weights can be positive and negative, and further comprising:

comparing the query certainty vector and the preference vector to each of the document feature vectors to retrieve the ranked result set of documents.

13. A system for indexing and retrieving documents, comprising:

a plurality of documents, each document having an associated document feature vector;

a database indexing each of the plurality of documents according the associated document feature vector;

a speech recognition engine for converting a spoken query to an intermediate representation representing possible sequential combinations of terms in the spoken query;

a query module, comprising a processor, the query module configured for generating a query certainty vector from the intermediate representation; and

a comparator configured to compare the query certainty vector to each of the document feature vectors to retrieve a ranked result set of documents.