Word recognition apparatus and method

• US 4,069,393 A
• Filed: 12/11/1974
• Issued: 01/17/1978
• Est. Priority Date: 09/21/1972
• Status: Expired due to Term

• Current Assignee: Siemens Corporate Research & Support, Inc. (Siemens AG)
• Original Assignee: Threshold Technology, Inc.
• Inventors: Martin, Thomas B., Cox, Robert B., Herscher, Marvin B.
• Primary Examiner(s): Claffy, Kathleen H.
• Assistant Examiner(s): Kemeny, E. S.
• Application Number:
  US05/531,543
• Time in Patent Office: 1,133 Days
• Field of Search: 179/1 SD, 179/1 SA, 179/1 SB, 179/1 VS, 340/146.3 WD
• US Class Current: 704/255
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence

Abstract

An apparatus which receives spoken input training words and a subsequent spoken input command word and generates a correlation function is indicative of the resemblance of the command word to each training word. A feature extraction means processes received input words and generates digital feature output signals on particular ones of a number of feature output lines. Means are provided for storing, as a time dependent matrix, the status of the feature signals which occur during each training word. Means are also provided for storing as a time dependent matrix, the status of the feature signals which occur during the command word. Further means are provided for comparing, member by member, the command word matrix with each training word matrix and for generating a correlation figure which reflects each comparison. In a preferred embodiment of the invention, means are provided for time normalizing the matrices representative of the training words and the command word before the comparisons are made.

Claims

1. Apparatus which receives spoken input training words and a subsequent spoken input command word and generates a correlation function that is indicative of the resemblance of the command word to each training word, comprising:

a. feature extraction means for processing received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. means for storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during each training word;

c. means for storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during the command word; and

d. means for comparing, member by member, the command word matrix with each training word matrix and for generating a correlation figure which reflects each matrix comparison, the member comparisons taking account only of the presence of features in the matrices being compared.

2. Apparatus as defined by claim 1 further comprising means for time normalizing the training word matrices and the command word matrix before the comparisons thereof.

3. Apparatus as defined by claim 1 wherein the correlation figure for each matrix comparison is a sum of positive and negative member comparisons, a positive comparison being indicated by the presence of a feature at a member position in both of two matrices being compared, and a negative comparison being indicated by the presence of a feature at a member position in only one of two matrices being compared.

4. Apparatus as defined by claim 1 further comprising means for comparing, on a time shifted basis, the command word matrix with each training word matrix and for generating a supplementary correlation figure which reflects each such comparison.

11. Apparatus as defined by claim 2 wherein said means for time normalizing divides each training word and the command word into a predetermined number of time slots of substantially equal duration.

5. Apparatus as defined by claim 5 further comprising means for comparing, on a time shifted basis, the command word matrix with each training word matrix and for generating a supplementary correlation figure which reflects each such comparison.

6. Apparatus as defined by claim 5 further comprising means for time normalizing the training word matrices and the command word matrix before the comparisons thereof.

12. Apparatus as defined by claim 6 wherein said means for time normalizing divides each training word and the command word into a predetermined number of time slots of substantially equal duration.

7. A method for receiving spoken input training words and a subsequent spoken input command word and generating a correlation function that is indicative of the resemblance of the command word to each training word, comprising the steps of:

a. extracting features from received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during each training word;

c. storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during the command word; and

d. comparing, member by member, the command word matrix with each training word matrix and generating a correlation figure which reflects each matrix comparison, the member comparison taking account only of the presence of features in the matrices being compared.

8. The method as defined by claim 7 further comprising a step of time normalizing the training word matrices and the command word matrix before the comparisons thereof.

9. The method as defined by claim 7 further comprising the step of comparing, on a time shifted basis, the command word matrix with each training word matrix and generating a supplementary correlation figure which reflects each comparison.

10. The method as defined by claim 8 further comprising the step of comparing, on a time shifted basis, the command word matrix with each training word matrix and generating a supplementary correlation figure which reflects each comparison.

13. Apparatus which receives spoken input training words and the subsequent spoken input command word and generates a correlation function that is indicative of the resemblance of the command word to each training word, comprising:

a. feature extraction means for processing received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. means for storing, as a time dependent matrix, the status of the feature signals which occur during each training word;

c. means for storing, as a time dependent matrix, the status of the feature signals which occur during the command word;

d. means for comparing, member by member, the command word matrix with each training word matrix and for generating a correlation figure which reflects each comparison; and

e. means for comparing, on a time shifted basis, the command word matrix with each training word matrix and for generating a supplementary correlation figure which reflects each such comparison.

14. Apparatus as defined by claim 13 further comprising means for time normalizing the training word matrices and the command word matrix before the comparisons thereof.

15. Apparatus as defined by claim 14 wherein said means for time normalizing divides each training word and the command word into a predetermined number of time slots of substantially equal duration.

16. Apparatus as defined by claim 15 further comprising means responsive to said correlation figure and supplementary correlation figure for generating an occurrence indication indicative of the training word which corresponds most closely to the command word.

17. Apparatus as defined by claim 16 wherein said means for generating an occurrence indication includes means for determining whether said combined correlation figure and supplementary correlation figure exceeds a predetermined threshold.

18. Apparatus which receives spoken input training words and a subsequent spoken input command word and generates an occurrence indication indicative of the training word which corresponds most closely to the command word, said occurrence indication being suitable for control of a companion system, comprising:

a. feature extraction means for processing received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. means for storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during each training word;

c. means for storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during the command word; and

d. means for comparing, member by member, the command word matrix with each training word matrix, the member comparisons taking account only of the presence of features in the matrices being compared, and for generating said occurrence indication for the training word whose matrix corresponds with at least a predetermined degree of similarity to the command word matrix.

19. Apparatus as defined by claim 18 wherein each matrix comparison is a sum of positive and negative member comparisons, a positive comparison being indicated by the presence of a feature at a member position in both of two matrices being compared and a negative comparison being indicated by the presence of a feature at a member position in only one of two matrices being compared.

20. Apparatus as defined by claim 18 further comprising means for comparing, on a time shifted basis, the command word matrix with each training word matrix.

21. Apparatus as defined by claim 18 further comprising means for time normalizing training word matrices and the command word matrix before the comparison thereof.

22. Apparatus as defined by claim 21 wherein said means for time normalizing divides each training word and a command word into a predetermined number of time slots of substantially equal duration.

23. Apparatus which receives spoken input training words and a subsequent spoken input command word and generates an occurrence indication indicative of the training word which corresponds most closely to the command word, said occurrence indication being suitable for control of a companion system, comprising:

a. feature extraction means for processing received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. means for storing, as a time dependent matrix, the status of the feature signals which occur during each training word;

c. means for storing, as a time dependent matrix, the status of the feature signals which occur during the command word;

d. means for comparing, member by member, the command word matrix with each training word matrix; and

e. means for comparing, on a time shifted basis, the command word matrix with each training word matrix, and for generating said occurrence indication for the training word whose matrix corresponds, in accordance with the defined comparisons, with at least a predetermined degree of similarity to the command word matrix.

24. Apparatus as defined by claim 23 further comprising means for time normalizing the training word matrices and the command word matrix before the comparisons thereof.

25. Apparatus as defined by claim 24 wherein said means for time normalizing divides each training word and the command word into a predetermined number of time slots of substantially equal duration.

26. A method for receiving spoken input training words and a subsequent spoken input command word and generating an occurrence indication indicative of the training word which most closely resembles the command word, said occurrence indication being suitable for control of a companion system, comprising the steps of:

a. extracting features from received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during each training word;

c. storing, as a time dependent matrix, the presence or absence status of the feature signals which occur during each command word; and

d. comparing, member by member, the command word matrix with each training word matrix, the member comparisons taking account only of the presence of features in the matrices being compared, and generating said occurrence indication for the training word whose matrix corresponds with at least a predetermined degree of similarity to the command word matrix.

27. A method for receiving spoken input training words and a subsequent spoken input command word and generating an occurrence indication indicative of the training word which most closely resembles the command word, said occurrence indication being suitable for control of a companion system, comprising the steps of:

a. extracting features from received input words and generating digital feature output signals on particular ones of a number of feature output lines, the particular ones being dependent upon the features present in an input word;

b. storing, as a time dependent matrix, the status of the feature signals which occur during each training word;

c. storing, as a time dependent matrix, the status of the feature signals which occur during the command word;

d. comparing, member by member, the command word matrix with each training word matrix; and

e. comparing, on a time shifted basis, the command word matrix with each training word matrix, and generating said occurrence indication for the training word whose matrix corresponds, in accordance with the defined comparisons, with at least a predetermined degree of similarity to the command word matrix.