SYSTEM AND METHOD FOR ROBUST PERSONALIZATION OF SPEECH RECOGNITION

• US 20140136210A1
• Filed: 11/14/2012
• Published: 05/15/2014
• Est. Priority Date: 11/14/2012
• Status: Abandoned Application

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: JOHNSTON, Michael J.
• Application Number:
  US13/676,531
• Publication Number:
  US 20140136210A1
• Time in Patent Office: Days
• US Class Current: 704/275
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models

Abstract

Personalization of speech recognition while maintaining privacy of user data is achieved by transmitting data associated with received speech to a speech recognition service and receiving a result from the speech recognition service. The speech recognition service result is generated from a general purpose speech language model. The system generates an input finite state machine from the speech recognition result and composes the input finite state machine with a phone edit finite state machine, to yield a resulting finite state machine. The system composes the resulting finite state machine with a user data finite state machine to yield a second resulting finite state machine, and uses a best path through the second resulting finite state machine to yield a user specific speech recognition result.

Claims

1. A method comprising:

transmitting data associated with received speech to a speech recognition service;

receiving a speech recognition result from the speech recognition service, wherein the speech recognition result is generated from a general purpose language model;

generating an input finite state machine based on the speech recognition result;

composing the input finite state machine with a phone edit finite state machine to yield a resulting finite state machine; and

using the resulting finite state machine with user specific data to generate a user-specific speech recognition result from the speech recognition result.

2. The method of claim 1, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to insert data.

3. The method of claim 1, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to delete data.

4. The method of claim 1, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to substitute phoneme data.

5. The method of claim 1, further comprising:

composing the resulting finite state machine with a user data finite state machine to yield a second resulting finite state machine.

6. The method of claim 5, wherein using the resulting finite state machine with user specific data comprises identifying a best path through the second resulting finite state machine to yield the user-specific speech recognition result.

7. The method of claim 1, wherein using the resulting finite state machine to generate a user specific speech recognition result is performed on a smartphone.

8. A computer-readable medium having instructions stored which, when executed by a processor, perform a method comprising:

transmitting data associated with received speech to a speech recognition service;

receiving a speech recognition result from the speech recognition service, wherein the speech recognition result is generated from a general purpose language model;

generating an input finite state machine based on the speech recognition result;

composing the input finite state machine with a phone edit finite state machine to yield a resulting finite state machine; and

using the resulting finite state machine with user specific data to generate a user-specific speech recognition result from the speech recognition result.

9. The computer-readable medium of claim 8, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to insert data.

10. The computer-readable medium of claim 8, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to delete data.

11. The computer-readable medium of claim 8, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to substitute phoneme data.

12. The computer-readable medium of claim 8, the computer-readable medium having additional instructions stored which result in the method further comprising:

composing the resulting finite state machine with a user data finite state machine to yield a second resulting finite state machine.

13. The computer-readable medium of claim 12, wherein using the resulting finite state machine with user specific data comprises identifying a best path through the second resulting finite state machine to yield the user-specific speech recognition result.

14. The computer-readable medium of claim 8, wherein the processor is a part of a smartphone.

15. A system comprising:

a processor; and

a computer-readable medium having instructions stored which, when executed by the processor, perform a method comprising;

transmitting data associated with received speech to a speech recognition service;

receiving a speech recognition result from the speech recognition service, wherein the speech recognition result is generated from a general purpose language model;

generating an input finite state machine based on the speech recognition result;

composing the input finite state machine with a phone edit finite state machine to yield a resulting finite state machine; and

using the resulting finite state machine with user specific data to generate a user-specific speech recognition result from the speech recognition result.

16. The system of claim 15, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to insert data.

17. The system of claim 15, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to delete data.

18. The system of claim 15, wherein composing the input finite state machine with a phone edit finite state machine further comprises assigning a cost to an operation that results in a manipulation of an input sequence in the input finite state machine to substitute phoneme data.

19. The system of claim 15, the computer-readable medium having additional instructions stored which result in the method further comprising:

composing the resulting finite state machine with a user data finite state machine to yield a second resulting finite state machine.

20. The system of claim 19, wherein using the resulting finite state machine with user specific data comprises identifying a best path through the second resulting finite state to yield the user-specific speech recognition result.