SPEECH RECOGNITION

• US 20110125499A1
• Filed: 11/24/2009
• Published: 05/26/2011
• Est. Priority Date: 11/24/2009
• Status: Active Grant

• Current Assignee: Nexidia Incorporated (Nice Ltd)
• Original Assignee: Nexidia Incorporated (Nice Ltd)
• Inventors: Griggs, Kenneth King, Arrowood, Jon A.
• Granted Patent: US 9,275,640 B2
• Time in Patent Office: Days
• US Class Current: 704/251
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

Systems, methods, and apparatus, including computer program products for accepting a predetermined vocabulary-dependent characterization of a set of audio signals, the predetermined characterization including an identification of putative occurrences of each of a plurality of vocabulary items in the set of audio signals, the plurality of vocabulary items included in the vocabulary; accepting a new vocabulary item not included in the vocabulary; accepting putative occurrences of the new vocabulary item in the set of audio signals; and generating, by an analysis engine of a speech processing system, an augmented characterization of the set of audio signals based on the identified putative occurrences of the new vocabulary item.

Claims

1. A computer-implemented method comprising:

accepting a predetermined vocabulary-dependent characterization of a set of audio signals, the predetermined characterization including an identification of putative occurrences of each of a plurality of vocabulary items in the set of audio signals, the plurality of vocabulary items included in the vocabulary;

accepting a new vocabulary item not included in the vocabulary;

accepting putative occurrences of the new vocabulary item in the set of audio signals; and

generating, by an analysis engine of a speech processing system, an augmented characterization of the set of audio signals based on the identified putative occurrences of the new vocabulary item.

2. The method of claim 1, wherein the analysis engine comprises a wordspotting engine.

3. The method of claim 1, further comprising generating, by a user interface engine, a visual representation of the augmented characterization and causing the visual representation to be displayed on a display terminal.

4. The method of claim 1, wherein the predetermined vocabulary-dependent characterization of the set of audio signals includes quantities representative of confidence scores associated with the putative occurrences of each vocabulary item.

5. The method of claim 1, wherein the predetermined vocabulary-dependent characterization of the set of audio signals comprises a transcript that is generated by a speech recognition engine of the speech processing system.

6. The method of claim 1, wherein the predetermined vocabulary-dependent characterization of the set of audio signals comprises a transcript that is manually generated by a human transcriptionist.

7. The method of claim 1, wherein accepting the new vocabulary item includes accepting a term that is present in a user-provided text corpus.

8. The method of claim 1, wherein accepting the new vocabulary item includes accepting a term that is present in a result set of an automatic discovery action.

9. The method of claim 8, wherein the automatic discovery action comprises a web crawler action.

10. The method of claim 8, wherein the automatic discovery action comprises a speech recognition action that is performed on a subset of the set of audio signals or a disjoint set of the set of audio signals using a different vocabulary.

11. The method of claim 1, further comprising identifying putative occurrences of the new vocabulary item.

12. The method of claim 11, wherein the putative occurrences of the new vocabulary item are identified by the analysis engine.

13. The method of claim 11, wherein the putative occurrences of the new vocabulary item are identified by a human user.

14. The method of claim 11, wherein identifying the putative occurrences includes determining quantities representative of confidence scores associated with the putative occurrences of the new vocabulary item.

15. The method of claim 11, wherein identifying the putative occurrences includes identifying a duration of each putative occurrence of the new vocabulary item.

16. The method of claim 1, wherein generating the augmented characterization includes comparing the confidence scores associated with the putative occurrences of the new vocabulary item with confidence scores associated with putative occurrences of each vocabulary item in the predetermined characterization.

17. The method of claim 1, wherein generating the augmented characterization includes:

updating the vocabulary to include the new vocabulary item; and

using the updated vocabulary, processing at least a portion of the set of audio signals including the duration of each putative occurrence of the new vocabulary item.

18. The method of claim 1, wherein generating the augmented characterization includes inserting the new vocabulary item into the predetermined characterization.

19. The method of claim 1, wherein generating the augmented characterization includes deleting a vocabulary item occurring in the predetermined characterization.

20. The method of claim 1, wherein generating the augmented characterization includes replacing a vocabulary item occurring in the predetermined characterization with the new vocabulary item.

21. The method of claim 1, wherein generating the augmented characterization includes augmenting a word lattice representative of the set of audio signals.

22. The method of claim 1, wherein generating the augmented characterization includes updating a transcription of the set of audio signals.

23. The method of claim 1, wherein accepting the new vocabulary item includes accepting the new vocabulary item from a user.

24. The method of claim 1, wherein the new vocabulary item is in a language different that that of at least one of the plurality of vocabulary items included in the vocabulary.

25. The method of claim 1, further comprising:

generating the predetermined vocabulary-dependent characterization of the set of audio signals, wherein the generating includes aligning, by a transcript alignment engine of the speech processing system, the set of audio signals and a transcript.

26. The method of claim 1, wherein the set of audio signals comprises one or more of the following:

a live audio stream, a legal deposition, a telephone call, and broadcast media.

27. The method of claim 26, wherein at least a first audio signal of the set of audio signals is associated with video.

28. A computer-readable medium storing instructions for causing a processor of a speech processing system to:

accept a predetermined vocabulary-dependent characterization of a set of audio signals, the predetermined characterization including an identification of putative occurrences of each of a plurality of vocabulary items in the set of audio signals, the plurality of vocabulary items included in the vocabulary;

accept a new vocabulary item not included in the vocabulary;

accept putative occurrences of the new vocabulary item in the set of audio signals; and

generate an augmented characterization of the set of audio signals based on the identified putative occurrences of the new vocabulary item.