Granted Patent B2

US 7,499,858 · App. 11/465,584 · Granted Mar 3, 2009

Abstract

Claims (33)

1. A method of information retrieval, comprising:

parsing a plurality of documents into a hierarchical set of phrases, each of the plurality of documents including text;

tokenizing the phrases;

constructing a language model using the phrases;

constructing an index for the phrases;

receiving a spoken query;

ranking each of the plurality of documents based on the spoken query; and

updating a current view displayed to a user according to the ranking.

2. The method of claim 1 wherein the step of tokenizing the phrases includes tokenizing pauses.

3. The method of claim 2 wherein the pauses are treated as start phrase tokens.

4. The method of claim 2 wherein the pauses are treated as end phrase tokens.

5. The method of claim 2 wherein the documents represent songs.

6. The method of claim 2 wherein the step of ranking includes scoring each of the plurality of documents based on the spoken query and randomizing order of documents having identical scores.

7. The method of claim 2 wherein the step of ranking each of the plurality of documents based on the spoken query includes representing the spoken query by a vector of weighted N-grams.

8. The method of claim 7 wherein the step of ranking each of the plurality of documents based on the spoken query further includes calculating a posteriori probabilities for each term in the spoken query and using the a posteriori or probabilities as weights.

9. The method of claim 8 wherein the step of ranking each of the plurality of documents based on the spoken query further includes determining a ranking for each of the plurality of documents based on the a posteriori probabilities of each term in the spoken query.

10. A method of information retrieval, comprising:

parsing a plurality of documents into a hierarchical set of phrases and tokenizing the phrases, each of the plurality of documents including text;

adding at least one start phrase marker or end phrase marker and tokenizing the at least one start phrase marker or end phrase marker;

constructing a language model and an index using the tokenized phrases and the tokenized at least one start or end phrase marker;

receiving a spoken query from a user of an information retrieval system;

retrieving one or more of the documents based on the user's spoken query and by using the language model and the index; and

displaying to the user an updated view that includes at least some of the one or more retrieved documents.

11. The method of claim 10 wherein the index comprises an inverted index.

12. The method of claim 10 wherein a step between the receiving and retrieving steps comprises using a speech recognizer to process the spoken query.

13. The method of claim 12 wherein the retrieving step comprises using the processed spoken query, the language model, and the index to retrieve the one or more retrieved documents.

14. The method of claim 10 wherein the spoken query comprises one or more terms and further comprising the step of calculating a posteriori probabilities related to the terms.

15. The method of claim 14 further comprising the step of modifying at least one of the calculated a posteriori probabilities.

16. The method of claim 15 wherein the modifying step comprises reducing at least one of the calculated a posteriori probabilities.

17. An information retrieval method, comprising:

receiving a spoken query from a user of an information retrieval system, the system including a language model and an index constructed from tokenized phrases and at least one tokenized start phrase marker or end phrase marker, the phrases created by parsing a plurality of documents into a hierarchical set of phrases, each of the plurality of documents including text;

retrieving one or more of the documents based on the user's spoken query and by using the language model and the index; and

displaying to the user an updated view that includes at least some of the one or more retrieved documents.

Assignments (8)