System for recognizing spoken sounds from continuous speech and method of using same

• US 5,734,793 A
• Filed: 08/29/1996
• Issued: 03/31/1998
• Est. Priority Date: 09/07/1994
• Status: Expired due to Term

• Current Assignee: Google Technology Holdings LLC (Alphabet Inc.)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Wang, Shay-Ping T.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US08/707,451
• Time in Patent Office: 579 Days
• Field of Search: 395/2.41, 395/2.4, 395/23, 395/24, 395/25
• US Class Current: 704/232
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• 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 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems
• CPC Class Codes: G10L 15/34 Adaptation of a single recogniser for parallel processing, e.g. by use of multiple processors or cloud computing
• CPC Class Codes: G10L 25/27 characterised by the analysis technique

Abstract

A system for recognizing spoken sounds from continuous speech includes a plurality of classifiers and a selector. Each of the classifiers implements a discriminant function which is based on a polynomial expansion. By determining the polynomial coefficients of a discriminant function, the corresponding classifier is tuned to classify a specific spoken sound. The selector utilizes the classifier outputs to identify the spoken sounds. A method of using the system is also provided.

Claims

1. A method of recognizing a spoken sound from continuously spoken speech, the continuously spoken speech including a plurality of spoken sounds, the method comprising the steps of:

(a) extracting at least one feature from an interval of the continuously spoken speech;

(b) distributing the at least one feature to a plurality of classifiers, each of the classifiers generating a classifier output signal as a function of a polynomial expansion that includes the at least one feature as an independent variable, whereby producing a plurality of classifier output signals; and

(c) identifying the spoken sound by comparing the plurality of classifier output signals to one another in order to determine which of the plurality of classifier output signals has a largest magnitude.

2. The method of claim 1, wherein each of the plurality of classifiers implements a respective polynomial expansion comprising a plurality of coefficients which have been adjusted during a prior training phase to classifiy a respective one of the spoken sounds.

3. The method of claim 1 wherein the at least one feature is selected from a group consisting of cepstral coefficients, predictive coefficients, and Fourier coefficients.

4. The method of claim 1, wherein the polynomial expansion has the form:

##EQU3## wherein y represents the classifier output signal;

wherein i, m, and n are integers;

wherein w_i-1 represents the coefficient for the ith term;

wherein x₁, x₂, . . . , x_n represent one or more independent variables; and

wherein g_1i, . . . , g_ni represent the exponents for the ith term in the expansion which are applied to the one or more independent variables.

5. The method of claim 1, wherein steps (a)-(c) are repeated for a sequence of intervals of time to identify a sequence of spoken sounds from the continuously spoken speech.

6. The method of claim 5 wherein each interval of time has a duration of 10 milliseconds.

7. The method of claim 1 wherein the spoken sound is a word, character, or phoneme.

8. The method of claim 1 wherein the method is implemented by a computer program.

9. A system for recognizing a spoken sound from continuously spoken speech having a plurality of spoken sounds, the system comprising:

a plurality of classifiers for generating a plurality of classifier output signals, each of the classifiers receiving at least one feature extracted from an interval of the continuously spoken speech and generating a classifier output signal as a function of a polynomial expansion that includes the at least one feature as an independent variable; and

a selector for identifying the spoken sound by comparing the plurality of classifier output signals to one another in order to determine which of the plurality of classifier output signals has a largest magnitude.

10. The system recited in claim 9 further comprising:

a converter which receives the continuously spoken speech and generates a speech signal from the continuously spoken speech;

a feature extractor for extracting the at least one feature from the speech signal; and

a means for distributing the at least one feature to the plurality of classifiers.

11. The system of claim 9, wherein the at least one feature is selected from a group consisting of cepstral coefficients, predictive coefficients, and Fourier coefficients.

12. The system of claim 9, wherein each of the plurality of classifiers implements a respective polynomial expansion comprising a plurality of coefficients which have been adjusted during a prior training phase to classifiy a respective one of the spoken sounds.

13. The system of claim 9, wherein the polynomial expansion has the form:

##EQU4## wherein y represents the classifier output signal;

wherein i, m, and n are integers;

wherein w_i-1 represents the coefficient for the ith term;

wherein x₁, x₂, . . . , x_n represent one or more independent variables; and

wherein g_1i, . . . , g_ni represent the exponents for the ith term in the expansion which are applied to the one or more independent variables.

14. The system of claim 9 wherein the spoken sound is a word, character, or phoneme.

15. The system of claim 9 wherein the system is implemented in a computer program.

16. The system of claim 9 wherein the plurality of classifiers and the selector are implemented using at least one integrated circuit.

17. The system of claim 9 wherein the selector identifies a spoken sound corresponding to a sequence of intervals of time.

18. The system of claim 17 wherein each of the intervals of time has a duration of 10 milliseconds.

19. A method of recognizing a spoken sound from continuously spoken speech, the continuously spoken speech including a plurality of spoken sounds, the method comprising the steps of:

(a) extracting at least one feature from an interval of the continuously spoken speech;

(b) distributing the at least one feature to a plurality of classifiers, each of the classifiers generating a classifier output signal based on a polynomial expansion having a form ##EQU5## wherein y represents the classifier output signal, w_i-1 represents a coefficient for the ith term, i, m, and n are integers, x₁, x₂, . . . , x_n represent the at least one feature, and g_1i, . . . , g_ni represent the exponents for the ith term in the expansion, whereby the plurality of classifiers produce a plurality of classifier output signals; and

(c) identifying the spoken sound by comparing the plurality of classifier output signals to one another in order to determine which of the plurality of classifier output signals has a largest magnitude.

20. The method of claim 19, wherein at least one term of said polynomial expansion includes at least two indepent variables corresponding to non-zero exponents and representing different dimensions in a feature space.

21. The method of claim 19, wherein each of the plurality of classifiers implements a respective polynomial expansion comprising a plurality of coefficients which have been adjusted during a prior training phase to classifiy a respective one of the spoken sounds.

22. The method of claim 19, wherein the at least one feature is selected from a group consisting of cepstral coefficients, predictive coefficients, and Fourier coefficients.