Use of periodicity and jitter for automatic speech recognition

• US 6,055,499 A
• Filed: 05/01/1998
• Issued: 04/25/2000
• Est. Priority Date: 05/01/1998
• Status: Expired due to Fees

• Current Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Original Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Inventors: Chengalvarayan, Rathinavelu, Thomson, David Lynn
• Primary Examiner(s): Dorvil, Richemond
• Application Number:
  US09/071,214
• Time in Patent Office: 725 Days
• Field of Search: 704/250, 704/205, 704/206, 704/207, 704/208, 704/217, 704/216, 704/256, 704/255
• US Class Current: 704/250
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit

Abstract

A class of features related to voicing parameters that indicate whether the vocal chords are vibrating. Features describing voicing characteristics of speech signals are integrated with an existing 38-dimensional feature vector consisting of first and second order time derivatives of the frame energy and of the cepstral coefficients with their first and second derivatives. Hidden Markov Model (HMM)-based connected digit recognition experiments comparing the traditional and extended feature sets show that voicing features and spectral information are complementary and that improved speech recognition performance is obtained by combining the two sources of information.

Claims

1. A method of speech recognition comprising the step of:

a. starting with a standard feature vector;

b. including at least one voicing feature and at least one time derivative of at least one voicing feature with said standard feature vector; and

c. using said standard feature vector with said included features to recognize speech.

2. The method of claim 1, wherein said standard feature vector includes time derivatives of energy.

3. The method of claim 1, wherein said standard feature vector includes spectral features.

4. The method of claim 3, wherein said spectral features include cepstral coefficients.

5. The method of claim 1, after step b and before step c further comprising the step of including an energy feature with said standard feature vector and said voicing features.

6. The method of claim 5, wherein said standard feature vector has 38 features, said energy feature has one feature and said voicing features has five features.

7. The method of claim 1, wherein a sum of the features of said standard feature vector and said voicing features is 44.

8. The method of claim 1, wherein said voicing features are from a group of features including periodicity features and jitter features.

9. The method of claim 1, wherein said voicing features includes periodicity features.

10. The method of claim 1, wherein said voicing features includes jitter features.

11. The method of claim 1, wherein said voicing features includes periodicity features and jitter features.

12. The method of claim 1, wherein step b further includes the steps of:

normalizing an autocorrelation function with a peak at m=0 so that a ratio lies in a normalized range between 0 and 1; and

choosing a largest peak in the normalized autocorrelation function as the estimate of a pitch period and a corresponding frequency value of said pitch period is a measure of a periodicity feature.

13. An apparatus for speech recognition, comprising:

means for determining a standard feature vector;

means for storing a standard feature vector after it has been determined;

means for including at least one voicing feature and at least one time derivative of at least one voicing feature with said stored standard feature vector; and

means for using said stored standard feature vector and said included voicing features to recognize speech;

wherein an error rate for speech recognition is reduced because of a robustness resulting from including the at least one voicing feature and the at least one time derivative of at least one voicing feature.

14. The apparatus for speech recognition according to claim 13, wherein said speech recognition is less sensitive to differences in transmission conditions because of including said at least one voicing feature and said at least one time derivative of at least one voicing feature.

15. The apparatus for speech recognition according to claim 14, wherein said at least one voicing feature includes periodicity features and jitter features.

16. The apparatus for speech recognition according to claim 13, wherein said at least one voicing feature includes periodicity features and jitter features.

17. The apparatus for speech recognition according to claim 13, wherein said apparatus has been trained to recognize a string of words using a minimum string error training.

18. The apparatus for speech recognition according to claim 13, wherein said apparatus has been trained to recognize at least one word using maximum likelihood training.