Time-anchored posterior indexing of speech

• US 20070143110A1
• Filed: 12/15/2005
• Published: 06/21/2007
• Est. Priority Date: 12/15/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Chelba, Ciprian, Acero, Alejandro, Yu, Roger, Nguyen, Patrick, Seide, Frank Torsten, Gunawardana, Asela, Selberg, Erik
• Granted Patent: US 7,831,425 B2
• Time in Patent Office: Days
• US Class Current: 704/251
• CPC Class Codes: G10L 15/05 Word boundary detection
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A computer-implemented method of indexing a speech lattice for search of audio corresponding to the speech lattice is provided. The method includes identifying at least two speech recognition hypotheses for a word which have time ranges satisfying a criteria. The method further includes merging the at least two speech recognition hypotheses to generate a merged speech recognition hypothesis for the word.

Claims

1. A computer-implemented method of indexing a speech lattice for search of audio corresponding to the speech lattice, the method comprising:

identifying at least two speech recognition hypotheses for a word which have time ranges satisfying a criteria; and

merging the at least two speech recognition hypotheses to generate a merged speech recognition hypothesis for the word.

2. The computer-implemented method of claim 1, wherein each of the at least two speech recognition hypotheses for the word has an associated start time, an associated end time, and an associated probability, and wherein merging the at least two speech recognition hypotheses to generate the merged speech recognition hypothesis for the word further comprises combining the associated probabilities of the at least two speech recognition hypotheses which have time ranges satisfying the criteria.

3. The computer-implemented method of claim 2, wherein satisfying the criteria requires that the at least two speech recognition hypotheses for the word have identical start times and identical end times.

4. The computer-implemented method of claim 2, wherein satisfying the criteria requires that the at least two speech recognition hypotheses for the word have start times that are within a predetermined range of each other, and end times that are within a predetermined range of each other.

5. The computer-implemented method of claim 4, wherein merging the at least two speech recognition hypotheses to generate the merged speech recognition hypothesis for the word further comprises generating the merged speech recognition hypothesis for the word such that start and end times for the merged speech recognition hypothesis are the same as start and end times for a best of the at least two speech recognition hypotheses.

6. The computer-implemented method of claim 2, and further comprising storing an index entry to represent the merged speech recognition hypothesis for the word.

7. The computer-implemented method of claim 6, wherein storing the index entry to represent the merged speech recognition hypothesis for the word further comprises encoding the merged speech recognition hypothesis into an integer value represented by a predetermined number of bits, a first plurality of the predetermined number of bits representing a time range for the merged speech recognition hypothesis, a second plurality of the predetermined number of bits representing a quantized time duration and a probability of the merged speech recognition hypothesis.

8. The computer-implemented method of claim 7, wherein the time range for the merged speech recognition hypothesis is a center point between the start and end times of the merged speech recognition hypothesis.

9. The computer-implemented method of claim 7, wherein the time range for the merged speech recognition hypothesis is one of the start and end times of the merged speech recognition hypothesis.

10. The computer-implemented method of claim 7, and further comprising representing audio as a sequence of time ranges with at least one word hypothesis associated with each time range.

11. A computer-readable medium having computer-executable instructions for performing the steps of method claim 1.

12. A computer-implemented method of searching audio, the method comprising:

receiving a search query;

searching an index corresponding to the audio for entries associated with the search query;

for each of a plurality of speech signals, retrieving from the entry an identifier for the speech signal, time range information for a word hypothesis associated with the entry, and a probability of the word hypothesis appearing at the time range 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.

13. The computer-implemented method of claim 12, wherein the time range information for the word hypothesis associated with each entry is indicative of a start time and an end time of the word hypothesis.

14. The computer-implemented method of claim 13, wherein the search query is representative of a phrase containing first and second consecutive words, wherein searching the index for entries associated with the search query further comprises searching the index for entries associated with the first and second words in the query.

15. The computer-implemented method of claim 14, and further comprising:

identifying, from the time range information of the entries associated with the first and second words in the query, occurrences of the first and second words which are consecutive in time in a speech signal.

16. The computer-implemented method of claim 15, wherein identifying occurrences of the first and second words which are consecutive in time in a speech signal comprises identifying occurrences of the first and second words which are consecutive in time in a speech signal based upon an end time of the first word in the speech signal and a start time of the second word in the speech signal.

17. The computer-implemented method of claim 16, wherein identifying occurrences of the first and second words which are consecutive in time in a speech signal comprises identifying occurrences of the first and second words in a speech signal in which an end time of the first word in the speech signal is substantially identical to a start time of the second word in the speech signal.

18. The computer-implemented method of claim 16, wherein identifying occurrences of the first and second words which are consecutive in time in a speech signal comprises identifying occurrences of the first and second words in a speech signal in which an end time of the first word in the speech signal is within a predetermined time range of a start time of the second word in the speech signal.

19. The computer-implemented method of claim 18, wherein the end time of the first word in the speech signal is after the start time of the second word in the speech signal.