Realtime acoustic adaptation using stability measures

• US 8,849,664 B1
• Filed: 07/16/2013
• Issued: 09/30/2014
• Est. Priority Date: 06/05/2012
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Lei, Xin, Aleksic, Petar
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US13/943,320
• Time in Patent Office: 441 Days
• Field of Search: 704/270.1, 704/251, 704/235, 704/243, 704/244, 704/260, 704/270, 704/211, 704/231, 704/254, 704/245, 704/246, 704/233, 704/236, 379/88.02, 379/88.01, 709/231, 709/224
• US Class Current: 704/246
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 17/14 Use of phonemic categorisation or speech recognition prior to speaker recognition or verification

Abstract

Methods, systems, and computer programs encoded on a computer storage medium for real-time acoustic adaptation using stability measures are disclosed. The methods include the actions of receiving a transcription of a first portion of a speech session, wherein the transcription of the first portion of the speech session is generated using a speaker adaptation profile. The actions further include receiving a stability measure for a segment of the transcription and determining that the stability measure for the segment satisfies a threshold. Additionally, the actions include triggering an update of the speaker adaptation profile using the segment, or using a portion of speech data that corresponds to the segment. And the actions include receiving a transcription of a second portion of the speech session, wherein the transcription of the second portion of the speech session is generated using the updated speaker adaptation profile.

Claims

1. A computer-implemented method comprising:

decoding a first portion of a speech session to generate a first speech recognition hypothesis using a speaker adaptation profile;

identifying a segment of the first speech recognition hypothesis;

determining that a stability measure for the identified segment satisfies a threshold;

triggering a real-time update of the speaker adaptation profile using the identified segment, or using a portion of speech data that corresponds to the identified segment; and

decoding a second portion of the speech session to generate a second speech recognition hypothesis using the updated speaker adaptation profile.

2. The method of claim 1, wherein the identified segment comprises a word, sub-word, or group of words.

3. The method of claim 1, further comprising outputting the first speech recognition hypothesis and the second speech recognition hypothesis.

4. The method of claim 1, wherein the stability measure is based on one or more of an age metric, a right context metric, and a regression.

5. The method of claim 1, wherein triggering a real-time update of the speaker adaptation profile comprises adding the identified segment to an adaptation queue.

6. The method of claim 1, further comprising updating the speaker adaptation profile.

7. The method of claim 6, wherein updating the speaker adaptation profile comprises modifying an acoustic model.

8. The method of claim 1, wherein the stability measure represents a probability.

9. The method of claim 1, wherein the speech session comprises one utterance.

10. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising;

decoding a first portion of a speech session to generate a first speech recognition hypothesis using a speaker adaptation profile;

identifying a segment of the first speech recognition hypothesis;

determining that a stability measure for the identified segment satisfies a threshold;

triggering a real-time update of the speaker adaptation profile using the identified segment, or using a portion of speech data that corresponds to the identified segment; and

decoding a second portion of the speech session to generate a second speech recognition hypothesis using the updated speaker adaptation profile.

11. The system of claim 10, wherein the identified segment comprises a word, sub-word, or group of words.

12. The system of claim 10, wherein the operations further comprise outputting the first speech recognition hypothesis and the second speech recognition hypothesis.

13. The system of claim 10, wherein the stability measure is based on one or more of an age metric, a right context metric, and a regression.

14. The system of claim 10, wherein triggering a real-time update of the speaker adaptation profile comprises adding the identified segment to an adaptation queue.

15. The system of claim 10, further comprising updating the speaker adaptation profile.

16. The system of claim 15, wherein updating the speaker adaptation profile comprises modifying an acoustic model.

17. The system of claim 10, wherein the stability measure represents a probability.

18. The system of claim 10, wherein the speech session comprises one utterance.

19. A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

decoding a first portion of a speech session to generate a first speech recognition hypothesis using a speaker adaptation profile;

identifying a segment of the first speech recognition hypothesis;

determining that a stability measure for the identified segment satisfies a threshold;

triggering a real-time update of the speaker adaptation profile using the identified segment, or using a portion of speech data that corresponds to the identified segment; and

decoding a second portion of the speech session to generate a second speech recognition hypothesis using the updated speaker adaptation profile.

20. The computer-readable medium of claim 19, wherein the stability measure is based on one or more of an age metric, a right context metric, and a regression.