Method and apparatus for indexing speech

• US 20060265222A1
• Filed: 05/20/2005
• Published: 11/23/2006
• Est. Priority Date: 05/20/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Acero, Alejandro, Chelba, Ciprian I.
• Granted Patent: US 7,634,407 B2
• Time in Patent Office: Days
• US Class Current: 704/240
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

A method of indexing a speech segment includes identifying at least two alternative word sequences based on the speech segment. For each word in the alternative sequences, information is placed in an entry for the word in the index. The information indicates the position of the word in at least one of the alternative sequences.

Claims

1. A method of indexing a speech segment, the method comprising:

identifying at least two alternative speech unit sequences based on the speech segment;

for each speech unit in the at least two alternative speech unit sequences, placing information in an entry in the index that indicates a position of the speech unit in at least one of the two alternative speech unit sequences.

2. The method of claim 1 wherein the position is a relative word position.

3. The method of claim 1 wherein placing information in an entry further comprises placing a probability of the speech unit appearing at the position.

4. The method of claim 3 further comprising determining the probability of the speech unit appearing at the position by summing at least two probabilities, each probability associated with the speech unit appearing in a separate speech unit sequence.

5. The method of claim 4 wherein the probabilities that are summed are associated with a same position for the speech unit.

6. The method of claim 4 wherein identifying at least two alternative speech unit sequences comprises identifying a lattice of speech unit sequences and determining a probability of a speech unit comprises determining the probability of reaching a state in the lattice associated with the speech unit from the beginning of the lattice and determining the probability of reaching the end of the lattice from the state.

7. The method of claim 1 further comprising placing separate information in an entry for a speech unit to indicate a plurality of positions in the at least two alternative speech unit sequences where the speech unit appears.

8. The method of claim 1 wherein the speech unit comprises a word.

9. The method of claim 1 wherein the speech unit comprises a sub-word.

10. A computer-readable medium having computer-executable instructions for performing steps comprising:

receiving a search query;

searching an index for an entry associated with a word in the search query;

for each of a plurality of speech signals, retrieving from the entry an identifier for the speech signal, a position for the word, and a probability of the word appearing at the position given the speech signal;

using the probabilities to rank the speech signals relative to each other to form ranked speech signals; and

returning search results based on the ranked speech signals.

11. The computer-readable medium of claim 10 wherein using the probabilities to rank the speech signals comprises forming a bigram probability using a probability for a first word in the query associated with a first position in the speech signal and a probability for a second word in the query associated with a second position in the speech signal where the second position is after the first position.

12. The computer-readable medium of claim 11 further comprising forming a second bigram probability using the probability for the second word in the query associated with the second position in the speech signal and a probability for a third word in the query associated with a third position in the speech signal.

13. The computer-readable medium of claim 10 wherein using the probabilities to rank the speech signals comprises forming a score for a speech signal by forming a unigram score based on the probabilities and forming a bigram score based on the probabilities, and forming a weighted sum from the unigram score and the bigram score to form the score for the speech signal.

14. The computer-readable medium of claim 10 wherein a speech signal is divided into segments and retrieving a position comprises retrieving a position within a segment of the speech signal.

15. The computer-readable medium of claim 14 wherein forming a score for a speech signal comprises forming a unigram score by retrieving from the entry probabilities for each position in each segment of the speech signal where the word of the search query appears, and summing the retrieved probabilities.

16. The computer-readable medium of claim 10 further comprising modifying a word of the search query to form a modified word and searching the index for an entry associated with the modified word.

17. A method of searching for speech segments, the method comprising:

accessing an index containing probabilities of positions for words generated from a plurality of speech segments;

retrieving a set of probabilities for a word from the index;

and returning identifiers for speech segments that contain the word based on the set of probabilities.

18. The method of claim 17 wherein the index is formed from a plurality of speech recognition lattices, each lattice being generated from a separate speech segment and representing a plurality of possible word sequences.

19. The method of claim 18 wherein a probability of a position for a word in a speech segment represents a sum of probabilities of the word at multiple states in the lattice.

20. The method of claim 17 further comprising retrieving a set of probabilities for each word of a multi-word query and using the sets of probabilities to determine a multi-word n-gram score.