Technologies for improved keyword spotting

Patent metadata

Field	Value
Publication number	US-10217458-B2
Application number	US-201615274498-A
Country	US
Kind code	B2
Filing date	Sep 23, 2016
Priority date	Sep 23, 2016
Publication date	Feb 26, 2019
Grant date	Feb 26, 2019

Abstract

Technologies for improved keyword spotting are disclosed. A compute device may capture speech data from a user of the compute device, and perform automatic speech recognition on the captured speech data. The automatic speech recognition algorithm is configured to both spot keywords as well as provide a full transcription of the captured speech data. The automatic speech recognition algorithm may preferentially match the keywords compared to similar words. The recognized keywords may be used to improve parsing of the transcribed speech data or to improve an assistive agent in holding a dialog with a user of the compute device.

First claim

The invention claimed is:

1. A compute device for automatic speech recognition, the compute device comprising:
an automatic speech recognition algorithm trainer to acquire a statistical language model for an automatic speech recognition algorithm, wherein the statistical language model comprises a large-vocabulary language model that has been modified such that the large-vocabulary language model preferentially matches words present in a plurality of keywords;
a speech data capturer to receive speech data of a user of the compute device;
and an automatic speech recognizer to perform the automatic speech recognition algorithm on the speech data to produce an output transcript, wherein the output transcript comprises one or more keywords of the plurality of keywords and one or more words not in the plurality of keywords.

2. The compute device of claim 1 , wherein the large-vocabulary language model that has been modified to preferentially match words present in the plurality of keywords comprises a first hidden Markov model to match words present in a large vocabulary and a second hidden Markov model to match words present in the plurality of keywords.

3. The compute device of claim 2 , wherein weightings of the plurality of keywords are higher than corresponding weightings of the rest of the statistical language model such that the statistical language model preferentially matches the plurality of keywords.

4. The compute device of claim 2 , wherein the statistical language model is formed by a linear interpolation of the large-vocabulary language model and a keyword language model.

5. The compute device of claim 1 , further comprising a speech parser to:
identify, based on the one or more keywords, a context of a portion of the output transcript; and
parse the output transcript based on the context of the portion of the output transcript.

6. The compute device of claim 1 , wherein to acquire the statistical language model for the automatic speech recognition algorithm comprises to train a statistical language model for a large vocabulary and augment the statistical language model with a keyword language model such that the statistical language model preferentially matches the plurality of keywords.

7. The compute device of claim 6 , wherein the statistical language model has been trained using domain-specific training data.

8. The compute device of claim 1 , further comprising an assistive agent to update a belief state of the assistive agent in response to a match of the one or more keywords.

9. The compute device of claim 8 , wherein to update the belief state in response to matching the one or more keywords comprises to search a word lattice of the automatic speech recognition algorithm and to find a better match of the word lattice to the speech data based on the one or more keywords.

10. The compute device of claim 1 , wherein at least one of the keywords of the plurality of keywords is a keyphrase comprising two or more words.

11. A method for automatic speech recognition by a compute device, the method comprising:
acquiring, by the compute device, a statistical language model for an automatic speech recognition algorithm, wherein the statistical language model comprises a large-vocabulary language model that has been modified such that the large-vocabulary language model preferentially matches words present in a plurality of keywords;
receiving, by the compute device, speech data of a user of the compute device; and
performing, by the compute device, the automatic speech recognition algorithm on the speech data to produce an output transcript, wherein the output transcript comprises one or more keywords of the plurality of keywords and one or more words not in the plurality of keywords.

12. The method of claim 11 , wherein the large-vocabulary language model that has been modified to preferentially match words present in the plurality of keywords comprises a first hidden Markov model to match words present in a large vocabulary and a second hidden Markov model to match words present in the plurality of keywords.

13. The method of claim 12 , wherein weightings of the plurality of keywords are higher than corresponding weightings of the rest of the statistical language model such that the statistical language model preferentially matches the plurality of keywords.

14. The method of claim 12 , wherein the statistical language model is formed by a linear interpolation of the large-vocabulary language model and a keyword language model.

15. The method of claim 11 , wherein acquiring the statistical language model for the automatic speech recognition algorithm comprises training a statistical language model for a large vocabulary and augmenting the statistical language model with a keyword language model such that the statistical language model preferentially matches the plurality of keywords.

16. The method of claim 11 , further comprising updating, by an assistive agent of the compute device, a belief state of the assistive agent in response to matching the one or more keywords.

17. The method of claim 16 , wherein updating, by the assistive agent, the belief state in response to matching the one or more keywords comprises searching a word lattice of the automatic speech recognition algorithm and finding a better match of the word lattice to the speech data based on the one or more keywords.

18. One or more non-transitory, computer-readable media comprising a plurality of instructions thereon that, when executed, causes a compute device to:
acquire a statistical language model for an automatic speech recognition algorithm, wherein the statistical language model comprises a large-vocabulary language model that has been modified such that the large-vocabulary language model preferentially matches words present in a plurality of keywords;
receive speech data of a user of the compute device; and
perform the automatic speech recognition algorithm on the speech data to produce an output transcript, wherein the output transcript comprises one or more keywords of the plurality of keywords and one or more words not in the plurality of keywords.

19. The one or more non-transitory, computer-readable media of claim 18 , wherein the large-vocabulary language model that has been modified to preferentially match words present in the plurality of keywords comprises a first hidden Markov model to match words present in a large vocabulary and a second hidden Markov model to match words present in the plurality of keywords.

20. The one or more non-transitory, computer-readable media of claim 19 , wherein weightings of the plurality of keywords are higher than corresponding weightings of the rest of the statistical language model such that the statistical language model preferentially matches the plurality of keywords.

21. The one or more non-transitory, computer-readable media of claim 19 , wherein the statistical language model is formed by a linear interpolation of the large-vocabulary language model and a keyword language model.

22. The one or more non-transitory, computer-readable media of claim 18 , wherein the plurality of instructions further causes the compute device to:
identify, based on the one or more keywords, a context of a portion of the output transcript; and
parse the output transcript based on the context of the portion of the output transcript.

23. The one or more non-transitory, computer-readable media of claim 18 , wherein to acquire the statistica

Assignees

• Intel Corp

Inventors

• Mangalath Praful
• Bauer Josef G
• Stemmer Georg

Classifications

• G10L15/144

  Training of HMMs · CPC title

• G10L2015/228

  of application context · CPC title

• G10L15/142 Primary

  Hidden Markov Models [HMMs] · CPC title

• G10L15/22

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title

• G06F40/205

  Parsing · CPC title