System and method for improving robustness of speech recognition using vocal tract length normalization codebooks

• US 8,160,875 B2
• Filed: 08/26/2010
• Issued: 04/17/2012
• Est. Priority Date: 06/20/2007
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property II LP (AT&T, Inc.)
• Inventors: Gilbert, Mazin
• Primary Examiner(s): Azad, Abul
• Application Number:
  US12/869,039
• Publication Number:
  US 20100324893A1
• Time in Patent Office: 600 Days
• Field of Search: 704200-278
• US Class Current: 704/234
• CPC Class Codes: G10L 15/07 to the speaker

Abstract

Disclosed are systems, methods, and computer readable media for performing speech recognition. The method embodiment comprises selecting a codebook from a plurality of codebooks with a minimal acoustic distance to a received speech sample, the plurality of codebooks generated by a process of (a) computing a vocal tract length for a each of a plurality of speakers, (b) for each of the plurality of speakers, clustering speech vectors, and (c) creating a codebook for each speaker, the codebook containing entries for the respective speaker'"'"'s vocal tract length, speech vectors, and an optional vector weight for each speech vector, (2) applying the respective vocal tract length associated with the selected codebook to normalize the received speech sample for use in speech recognition, and (3) recognizing the received speech sample based on the respective vocal tract length associated with the selected codebook.

Claims

1. A method of performing speech recognition, the method comprising:

selecting a codebook, from a plurality of codebooks, indicating an initial vocal tract length based on an acoustic distance to a received speech, wherein the plurality of codebooks comprises codebooks for each of a plurality of speakers and is generated based on a respective vocal tract length for each of the plurality of speakers;

initiating a computation of a computed vocal tract length from the received speech;

normalizing the received speech based on at least one of the initial vocal tract length and the computed vocal tract length to yield a normalized received speech; and

recognizing the normalized received speech, wherein the recognizing uses normalized received speech that is normalized using the initial vocal tract length until the computation of the computed vocal tract length is complete.

2. The method of claim 1, further comprising:

identifying an additional speech vector in the received speech; and

selecting an additional codebook indicating an additional vocal tract length from the plurality of codebooks based on the additional speech vector.

3. The method of claim 2, further comprising:

normalizing the received speech based at least on the additional vocal tract length to yield additional normalized received speech; and

recognizing the additional normalized received speech.

4. The method of claim 1, wherein the method is performed frame by frame.

5. The method of claim 1, wherein recognizing the normalized received speech occurs in real time.

6. The method of claim 1, wherein the plurality of codebooks covers vocal tract lengths from approximately 0.8 to 1.2 times an ideal vocal tract length.

7. The method of claim 1, wherein selecting the codebook from the plurality of codebooks is based on a likelihood calculation.

20. The non-transitory computer-readable storage medium of claim 1, wherein selecting the codebook from the plurality of codebooks is based on a likelihood calculation.

8. A system for performing speech recognition, the system comprising:

a processor;

a first module configured to control the processor to select a codebook, from a plurality of codebooks, indicating an initial vocal tract length based on an acoustic distance to a received speech, wherein the plurality of codebooks comprises codebooks for each of a plurality of speakers and is generated based on a respective vocal tract length for each of the plurality of speakers;

a second module configured to control the processor to initiate a computation of a computed vocal tract length from the received speech;

a third module configured to control the processor to normalize the received speech based on at least the initial vocal tract length and the computed vocal tract length to yield a normalized received speech; and

a fourth module configured to control the processor to recognize the normalized received speech, wherein the recognizing uses normalized received speech that is normalized using the initial vocal tract length until the computation of the computed vocal tract length is complete.

9. The system of claim 8, further comprising:

a fifth module configured to control the processor to identify an additional speech vector in the received speech; and

a sixth module configured to control the processor to select an additional codebook indicating an additional vocal tract length from the plurality of codebooks based on the additional speech vector.

10. The system of claim 9, further comprising:

a seventh module configured to control the processor to normalize the received speech based at least on the additional vocal tract length to yield additional normalized received speech; and

a seventh module configured to control the processor to recognize the additional normalized received speech.

11. The system of claim 8, wherein the system operates frame by frame.

12. The system of claim 8, wherein the fourth module is further configured to control the processor to recognize the normalized received speech in real time.

13. The system of claim 8, wherein the plurality of codebooks covers vocal tract lengths from approximately 0.8 to 1.2 times an ideal vocal tract length.

14. The system of claim 8, wherein the first module is further configured to control the processor to select the codebook from the plurality of codebooks based on a likelihood calculation.

15. A non-transitory computer-readable storage medium storing instructions which, when executed by a computing device, cause the computing device to perform speech recognition, the instructions comprising:

selecting a codebook, from a plurality of codebooks, indicating an initial vocal tract length based on an acoustic distance to a received speech, wherein the plurality of codebooks comprises codebooks for each of a plurality of speakers and is generated based on a respective vocal tract length for each of the plurality of speakers;

initiating a computation of a computed vocal tract length from the received speech;

normalizing the received speech based on at least one of the initial vocal tract length and the computed vocal tract length to yield a normalized received speech; and

recognizing the normalized received speech, wherein the recognizing uses normalized received speech that is normalized using the initial vocal tract length until the computation of the computed vocal tract length is complete.

16. The non-transitory computer-readable storage medium of claim 15, further comprising:

identifying an additional speech vector in the received speech;

selecting an additional codebook indicating an additional vocal tract length from the plurality of codebooks based on the additional speech vector.

17. The non-transitory computer-readable storage medium of claim 16, further comprising:

normalizing the received speech based at least on the additional vocal tract length to yield additional normalized received speech;

recognizing the additional normalized received speech.

18. The non-transitory computer-readable storage medium of claim 15, wherein recognizing the normalized speech is performed frame by frame.

19. The non-transitory computer-readable storage medium of claim 15, wherein recognizing the normalized received speech occurs in real time.