Audio-visual codebook dependent cepstral normalization

• US 7,319,955 B2
• Filed: 11/29/2002
• Issued: 01/15/2008
• Est. Priority Date: 11/29/2002
• Status: Active Grant

• Current Assignee: Cerence Operating Company (Cerence Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Deligne, Sabine, Potamianos, Gerasimos, Neti, Chalapathy V.
• Primary Examiner(s): Hudspeth; David
• Assistant Examiner(s): ALBERTALLI, BRIAN LOUIS
• Application Number:
  US10/307,164
• Publication Number:
  US 20040107098A1
• Time in Patent Office: 1,873 Days
• Field of Search: None
• US Class Current: 704/233
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 15/24 Speech recognition using non-acoustical features

Abstract

An arrangement for yielding enhanced audio features towards the provision of enhanced audio-visual features for speech recognition. Input is provided in the form of noisy audio-visual features and noisy audio features related to the noisy audio-visual features.

Claims

1. An apparatus for enhancing speech for speech recognition, said apparatus comprising:

a first input medium which obtains noisy audio-visual features;

a second input medium which obtains noisy audio features related to the noisy audio-visual features; and

a cepstral speech function output arrangement for combining the first and second inputs to yield enhanced audio features that are re-combined with visual features to yield enhanced audio-visual features used for speech recognition.

2. The apparatus according to claim 1, wherein said arrangement for yielding enhanced audio features is adapted to yield estimated clean speech features.

3. The apparatus according to claim 1, wherein said arrangement for yielding enhanced audio features comprises an arrangement for determining a posterior distribution based on the noisy audio-visual features.

4. The apparatus according to claim 3, wherein said arrangement for determining a posterior distribution is adapted to determine a posterior distribution based additionally on Gaussian parameters which model a probability density function related to the noisy audio-visual features.

5. The apparatus according to claim 1, wherein said arrangement for yielding enhanced audio features comprises an arrangement for estimating audio compensation codewords.

6. The apparatus according to claim 1, wherein said arrangement for yielding enhanced audio features comprises an arrangement for determining the difference between the noisy audio features and modified noisy audio-visual features.

7. The apparatus according to claim 1, wherein said arrangement for yielding enhanced audio features comprises:

an arrangement for determining a posterior distribution based on the noisy audio-visual features; and

an arrangement for estimating audio compensation codewords.

8. The apparatus according to claim 7, wherein said arrangement for yielding enhanced audio features comprises an arrangement for effecting a multiplication of the posterior distribution with the estimated audio compensation codewords.

9. The apparatus according to claim 8, wherein said arrangement for yielding enhanced audio features comprises an arrangement for determining the difference between the noisy audio features and the multiplication of the posterior distribution with the estimated audio compensation codewords.

10. The apparatus according to claim 1, wherein said first input medium is adapted to accept noisy audio-visual features which have resulted from the processing of normalized audio features and normalized videofeatures.

11. A method of enhancing speech for speech recognition, said method comprisingthe steps of:

obtaining noisy audio-visual features;

obtaining noisy audio features related to the noisy audio-visual features; and

using a cepstral speech function operating on the noisy audio features and the noisy audio-visual features to yield enhanced audio features that are re-combined with visual features to yield enhanced audio-visual features used for speech recognition.

12. The method according to claim 11, wherein said step of yielding enhanced audio features comprises yielding estimated clean speech features.

13. The method according to claim 11, wherein said step of yielding enhanced audio features comprises determining a posterior distribution based on the noisy audio-visual features.

14. The method according to claim 13, wherein said step of determining a posterior distribution comprises determining a posterior distribution based additionally on Gaussian parameters which model a probability density function related to the noisy audio-visual features.

15. The method according to claim 11, wherein said step of yielding enhanced audio features comprises estimating audio compensation codewords.

16. The method according to claim 11, wherein said step of yielding enhanced audio features comprises determining the difference between the noisy audio features and modified noisy audio-visual features.

17. The method according to claim 11, wherein said step of yielding enhanced audio features comprises:

determining a posterior distribution based on the noisy audio-visual features; and

estimating audio compensation codewords.

18. The method according to claim 17, wherein said step of yielding enhanced audio features comprises effecting a multiplication of the posterior distribution with the estimated audio compensation codewords.

19. The method according to claim 18, wherein said step of yielding enhanced audio features comprises determining the difference between the noisy audio features and the multiplication of the posterior distribution with the estimated audio compensation codewords.

20. The method according to claim 11, wherein said step of obtaining noisy audio-visual features comprises obtaining noisy audio-visual features which have resulted from the processing of normalized audio features and normalized video features.

21. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for enhancing speech for speech recognition, said method comprising the steps of:

obtaining noisy audio-visual features;

obtaining noisy audio features related to the noisy audio-visual features; and

using a cepstral speech function operating on the noisy audio features and the noisy audio-visual features to yield enhanced audio features that are re-combined with visual features to yield enhanced audio-visual features used for speech recognition.