Speech coding apparatus with single-dimension acoustic prototypes for a speech recognizer

• US 5,280,562 A
• Filed: 10/03/1991
• Issued: 01/18/1994
• Est. Priority Date: 10/03/1991
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Lucassen, John M., Bellegarda, Jerome R., Nahamoo, David, Picheny, Michael A., Epstein, Edward A., Bahl, Lalit R.
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Doerrler, Michelle
• Application Number:
  US07/770,495
• Time in Patent Office: 838 Days
• Field of Search: 381/29-52, 395/2
• US Class Current: 704/200
• CPC Class Codes: G10L 19/038 Vector quantisation, e.g. TwinVQ audio
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94

Abstract

In speech recognition and speech coding, the values of at least two features of an utterance are measured during a series of time intervals to produce a series of feature vector signals. A plurality of single-dimension prototype vector signals having only one parameter value are stored. At least two single-dimension prototype vector signals having parameter values representing first feature values, and at least two other single-dimension prototype vector signals have parameter values representing second feature values. A plurality of compound-dimension prototype vector signals have unique identification values and comprise one first-dimension and one second-dimension prototype vector signal. At least two compound-dimension prototype vector signals comprise the same first-dimension prototype vector signal. The feature values of each feature vector signal are compared to the parameter values of the compound-dimension prototype vector signals to obtain prototype match scores. The identification values of the compound-dimension prototype vector signals having the best prototype match scores for the feature vectors signals are output as a sequence of coded representations of an utterance to be recognized. A match score, comprising an estimate of the closeness of a match between a speech unit and the sequence of coded representations of the utterance, is generated for each of a plurality of speech units. At least one speech subunit, of one or more best candidate speech units having the best match scores, is displayed.

Claims

1. A speech coding apparatus comprising:

means for measuring the values of at least first and second different features of an utterance during each of a series of successive time intervals to produce a series of feature vector signals representing the feature values;

means for storing a plurality of single-dimension prototype vector signals, each single-dimension prototype vector signal having at least one parameter value, at least two single-dimension prototype vector signals being first-dimension prototype vector signals having parameter values representing first feature values, at least two other single-dimension prototype vector signals being second-dimension prototype vector signals having parameter values representing second feature values;

means for storing a plurality of compound-dimension prototype vector signals, each compound-dimension prototype vector signal having a unique identification value, each compound-dimension prototype vector signal comprising one first-dimension prototype vector signal and one second-dimension prototype vector signal, at least two compound-dimension prototype vector signals comprising the same first-dimension prototype vector signal;

means for comparing the closeness of the feature values of a feature vector signal to the parameter values of the compound-dimension prototype vector signals to obtain prototype match scores for the feature vector signal and each compound-dimension prototype vector signal; and

means for outputting at least the identification value of the compound-dimension prototype vector signal having the best prototype match score as a coded representation signal of the feature vector signal.

2. A speech coding apparatus as claimed in claim 1, characterized in that the means for comparing comprises:

means for comparing the value of the first feature of a feature vector signal with the parameter value of each first-dimension prototype vector signal exactly one time only; and

means for comparing the value of the second feature of a feature vector signal with the parameter value of each second-dimension prototype vector signal exactly one time only.

3. A speech coding apparatus as claimed in claim 2, characterized in that a plurality of single-dimension prototype vector signals are contained in two or more compound-dimension prototype vector signals.

4. A speech coding apparatus as claimed in claim 3, characterized in that each compound-dimension prototype vector signal contains at least one single-dimension prototype vector signal which is also contained in at least some other compound-dimension prototype vector signal.

5. A speech coding apparatus as claimed in claim 4, characterized in that for each dimension, the number of single-dimension prototype vector signals is less than one-half the number of compound-dimension prototype vector signals.

6. A speech coding apparatus as claimed in claim 1, characterized in that the measuring means comprises a microphone.

7. A speech coding apparatus as claimed in claim 1, characterized in that the comparison means comprises:

means for comparing the closeness of the feature values of a first feature vector signal to the parameter values of a first compound-dimension prototype vector signal to obtain a prototype match score for the feature vector signal and the first compound-dimension prototype vector signal;

means for comparing the closeness between the parameter values of a first part of a second compound-dimension prototype vector signal and the feature values of a corresponding first part of the first feature vector signal to obtain a partial prototype match score for the second compound-dimension prototype vector signal and the first feature vector signal; and

means for selecting the first compound-dimension prototype vector signal as better than the second compound-dimension prototype vector signal if the prototype match score for the first feature vector signal and the first compound-dimension prototype vector signal is better than the partial prototype match score for the second compound-dimension prototype vector signal and the first feature vector signal.

8. A speech coding apparatus as claimed in Claim 7, characterized in that the comparison means further comprises:

means for estimating a best remainder prototype match score for the feature values of a second part of the feature vector signal different from the first part of the feature vector signal; and

means for selecting the first compound-dimension prototype vector signal as better than the second compound-dimension prototype vector signal if the prototype match score for the first feature vector signal and the first compound-dimension prototype vector signal is better than the sum of the partial prototype match score and the best remainder prototype match score.

9. A speech coding apparatus as claimed in Claim 8, characterized in that the means for estimating the best remainder prototype match score comprises:

means for comparing the closeness between the second part of the first feature vector signal and the closest corresponding single-dimension prototype vector signals to obtain a first minimum remainder prototype match score;

means for comparing the closeness between the second part of a second feature vector signal, corresponding to the second part of the first feature vector signal, and the closest corresponding single-dimension prototype vector signals to obtain a second minimum remainder prototype match score;

means for comparing the closeness between the second part of the second feature vector signal and a corresponding second part of a second compound-dimension prototype vector signal having the best prototype match score with the second feature vector signal to obtain a prior reference remainder prototype match score;

means for combining, by weighted combination, the first minimum remainder prototype match score with the excess of the prior reference remainder match score over the second minimum remainder prototype match score to estimate the best remainder match score.

10. A method of speech coding comprising the steps of:

measuring the values of at least first and second different features of an utterance during each of a series of successive time intervals to produce a series of feature vector signals representing the feature values;

storing a plurality of single-dimension prototype vector signals, each single-dimension prototype vector signal having at least one parameter value, at least two single-dimension prototype vector signals being first-dimension prototype vector signals having parameter values representing first feature values, at least two other single-dimension prototype vector signals being second-dimension prototype vector signals having parameter values representing second feature values;

storing a plurality of compound-dimension prototype vector signals, each compound-dimension prototype vector signal having a unique identification value, each compound-dimension prototype vector signal comprising one first-dimension prototype vector signal and one second-dimension prototype vector signal, at least two compound-dimension prototype vector signals comprising the same first-dimension prototype vector signal;

comparing the closeness of the feature values of a feature vector signal to the parameter values of the compound-dimension prototype vector signals to obtain prototype match scores for the feature vector signal and each compound-dimension prototype vector signal; and

outputting at least the identification value of the compound-dimension prototype vector signal having the best prototype match score as a coded representation signal of the feature vector signal.

11. A speech coding method as claimed in Claim 10, characterized in that the step of comparing comprises:

comparing the value of the first feature of a feature vector signal with the parameter value of each first-dimension prototype vector signal exactly one time only; and

comparing the value of the second feature of a feature vector signal with the parameter value of each second-dimension prototype vector signal exactly one time only.

12. A speech coding apparatus as claimed in Claim 11, characterized in that a plurality of single-dimension prototype vector signals are contained in two or more compound-dimension prototype vector signals.

13. A speech coding method as claimed in Claim 12, characterized characterized in that each compound-dimension prototype vector signal contains at least one single-dimension prototype vector signal which is also contained in at least some other compound-dimension prototype vector signal.

14. A speech coding method as claimed in Claim 13, characterized in that for each dimension, the number of single-dimension prototype vector signals is less than one-half the number of compound-dimension prototype vector signals.

15. A speech recognition apparatus comprising:

means for measuring the values of at least first and second different features of an utterance during each of a series of successive time intervals to produce a series of feature vector signals representing the feature values;

means for storing a plurality of single-dimension prototype vector signals, each single-dimension prototype vector signal having at least one parameter value, at least two single-dimension prototype vector signals being first-dimension prototype vector signals having parameter values representing first feature values, at least two other single-dimension prototype vector signals being second-dimension prototype vector signals having parameter values representing second feature values;

means for storing a plurality of compound-dimension prototype vector signals, each compound-dimension prototype vector signal having a unique identification value, each compound-dimension prototype vector signal comprising one first-dimension prototype vector signal and one second-dimension prototype vector signal, at least two compound-dimension prototype vector signals comprising the same first-dimension prototype vector signal;

means for comparing the closeness of the feature values of each feature vector signal to the parameter values of the compound-dimension prototype vector signals to obtain prototype match scores for each feature vector signal and each compound-dimension prototype vector signal;

means for outputting the identification values of the compound-dimension prototype vector signals having the best prototype match scores for the feature vector signals as a sequence of coded representations of an utterance to be recognized;

means for generating a match score for each of a plurality of speech units, each match score comprising an estimate of the closeness of a match between the speech unit and the sequence of coded representations of the utterance, each speech unit comprising one or more speech subunits;

means for identifying one or more best candidate speech units having the best match scores; and

means for outputting at least one speech subunit of one or more of the best candidate speech units.

16. A speech recognition apparatus as claimed in Claim 15, characterized in that the means for comparing comprises:

means for comparing the value of the first feature of a feature vector signal with the parameter value of each first-dimension prototype vector signal exactly one time only; and

means for comparing the value of the second feature of a feature vector signal with the parameter value of each second-dimension prototype vector signal exactly one time only.

17. A speech recognition apparatus as claimed in Claim 16, characterized in that a plurality of single-dimension prototype vector signals are contained in two or more compound-dimension prototype vector signals.

18. A speech recognition apparatus as claimed in Claim 17, characterized in that each compound-dimension prototype vector signal contains at least one single-dimension prototype vector signal which is also contained in at least some other compound-dimension prototype vector signal.

19. A speech recognition apparatus as claimed in Claim 18, characterized in that for each dimension, the number of single-dimension prototype vector signals is less than one-half the number of compound-dimension prototype vector signals.

20. A speech recognition apparatus as claimed in Claim 15, characterized in that the measuring means comprises a microphone.

21. A speech recognition apparatus as claimed in Claim 15, characterized in that the means for outputting at least one speech subunit comprises a display.

22. A speech recognition apparatus as claimed in Claim 15, characterized in that the means for outputting at least one speech subunit comprises a speech synthesizer.

23. A speech recognition method comprising:

measuring the values of at least first and second different features of an utterance during each of a series of successive time intervals to produce a series of feature vector signals representing the feature values;

storing a plurality of single-dimension prototype vector signals, each single-dimension prototype vector signal having at least one parameter value, at least two single-dimension prototype vector signals being first-dimension prototype vector signals having parameter values representing first feature values, at least two other single-dimension prototype vector signals being second-dimension prototype vector signals having parameter values representing second feature values;

storing a plurality of compound-dimension prototype vector signals, each compound-dimension prototype vector signal having a unique identification value, each compound-dimension prototype vector signal comprising one first-dimension prototype vector signal and one second-dimension prototype vector signal, at least two compound-dimension prototype vector signals comprising the same first-dimension prototype vector signal;

comparing the closeness of the feature values of each feature vector signal to the parameter values of the compound-dimension prototype vector signals to obtain prototype match scores for each feature vector signal and each compound-dimension prototype vector signal;

outputting the identification values of the compound-dimension prototype vector signals having the best prototype match scores for the feature vector signals as a sequence of coded representations of an utterance to be recognized;

generating a match score for each of a plurality of speech units, each match score comprising an estimate of the closeness of a match between the speech unit and the sequence of coded representations of the utterance, each speech unit comprising one or more speech subunits;

identifying one or more best candidate speech units having the best match scores; and

outputting at least one speech submit of one or more of the best candidate speech units.

24. A speech recognition method as claimed in Claim 23, characterized in that the step of comparing comprises:

comparing the value of the first feature of a feature vector signal with the parameter value of each first-dimension prototype vector signal exactly one time only; and

comparing the value of the second feature of a feature vector signal with the parameter value of each second-dimension prototype vector signal exactly one time only.

25. A speech recognition method as claimed in Claim 24, characterized in that a plurality of single-dimension prototype vector signals are contained in two or more compound-dimension prototype vector signals.

26. A speech recognition method as claimed in Claim 25, characterized in that each compound-dimension prototype vector signal contains at least one single-dimension prototype vector signal which is also contained in at least some other compound-dimension prototype vector signal.

27. A speech recognition method as claimed in Claim 26, characterized in that for each dimension, the number of single-dimension prototype vector signals is less than one-half the number of compound-dimension prototype vector signals.

28. A speech coding apparatus comprising:

means for measuring the value of at least a first feature of an utterance during each of a series of successive time intervals to produce a series of feature vector signals representing the feature values;

means for storing a plurality of prototype vector signals, each prototype vector signal having at least one parameter value representing a first feature value and having a unique identification value;

means for comparing the closeness of the feature value of a feature vector signal to the parameter values of the prototype vector signals to obtain prototype match scores for the feature vector signal and each prototype vector signal; and

means for outputting at least the identification value of the prototype vector signal having the best prototype match score as a coded representation signal of the feature vector signal;

characterized in that the comparison means comprises means for comparing the closeness of the feature value of an uncoded feature vector signal to the parameter value of the prototype vector signal having the best prototype match score for an immediately preceding feature vector signal prior to comparing the feature value of the uncoded feature vector signal to the parameter values of other prototype vector signals.