Continuous mandarin chinese speech recognition system having an integrated tone classifier

• US 5,602,960 A
• Filed: 09/30/1994
• Issued: 02/11/1997
• Est. Priority Date: 09/30/1994
• Status: Expired due to Term

• Current Assignee: Apple Inc.
• Original Assignee: Apple Computer Incorporated (Apple Inc.)
• Inventors: Hon, Hsiao-Wuen, Lee, Kai-Fu, Chow, Yen-Lu
• Primary Examiner(s): Tung, Kee M.
• Application Number:
  US08/316,257
• Time in Patent Office: 865 Days
• Field of Search: 395/2, 395/2.14-2.16, 395/2.6-2.66, 381/41-43, 381/36-38
• US Class Current: 704/207
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 2015/027 Syllables being the recognition units
• CPC Class Codes: G10L 25/15 the extracted parameters being formant information

Abstract

A speech recognition system for continuous Mandarin Chinese speech comprises a microphone, an A/D converter, a syllable recognition system, an integrated tone classifier, and a confidence score augmentor. The syllable recognition system generates N-best theories with initial confidence scores. The integrated tone classifier has a pitch estimator to estimate the pitch of the input once and a long-term tone analyzer to segment the estimated pitch according to the syllables of each of the N-best theories. The long-term tone analyzer performs long-term tonal analysis on the segmented, estimated pitch and generates a long-term tonal confidence signal. The confidence score augmentor receives the initial confidence scores and the long-term tonal confidence signals, modifies each initial confidence score according to the corresponding long-term tonal confidence signal, re-ranks the N-best theories according to the augmented confidence scores, and outputs the N-best theories.

Claims

1. An integrated tone classifier for performing long-term tonal analysis of an input signal of continuous speech of a tonal language, the integrated tone classifier comprising:

a pitch estimator, having an input coupled to receive the input signal and an output, for estimating the pitch contour of the input signal; and

a long-term tone analyzer, having an input coupled to the output of the pitch estimator and an output forming an output of the integrated tone classifier, for segmenting an estimated pitch contour generated by the pitch estimator into units and for performing long-term tonal analysis on the units of the segmented, estimated pitch.

2. The system of claim 1, wherein the pitch estimator comprises

a memory device, coupled to a processor, the memory device having a first portion for storing program instruction steps for execution by the processor to estimate the pitch contour of the input signal.

3. The system of claim 2, wherein the long-term tone analyzer comprises a second portion of the memory device, the second portion of the memory device for storing instruction steps for execution by the processor to segment an estimate pitch contour generated by the pitch estimator into units and to perform long-term tonal analysis of the units of the segment, estimated pitch.

4. The system of claim 1, wherein the pitch estimator uses fourier analysis to determine the pitch contour of the input signal.

5. The system of claim 1, wherein the pitch estimator uses a low pass filter to determine the pitch contour of the input signal.

6. The system of claim 1, wherein the pitch estimator determines a pitch only once for the entire input signal.

8. The system of claim 4, wherein the integrated tone classifier comprises:

a pitch estimator, having an input coupled to receive the input signal and an output, for estimating the pitch contour of the input signal; and

a long-term tone analyzer, having a first input coupled to an output of the pitch estimator and a second input coupled to an output of the syllable recognition system and an output forming the output of the integrated tone classifier, for segmenting an estimated pitch contour generated by the pitch estimator into units according to an output of the syllable recognition system and for performing long-term tonal analysis on the units of the segmented, estimated pitch.

9. The system of claim 8, wherein the long-term tone analyzer generates a long-term tonal confidence signal indicating how well the units of a segmented, estimated pitch contour match the tones of the tonal language.

10. The system of claim 9, further comprising a confidence score augmentor, having a first input coupled to the output of the integrated tone classifier and a second input coupled to the output of the syllable recognition system, for modifying an initial confidence score generated by the syllable recognition system according to a long-term tonal confidence signal generated by the integrated tone classifier to generate an augmented confidence score.

11. The system of claim 10, wherein the confidence score augmentor comprises:

a first portion of a memory device, the memory device coupled to a processor, the first portion of the memory device for storing program instruction steps for execution by the processor to modify an initial confidence score generated by the syllable recognition system according to a long-term tonal confidence signal generated by the tone classifier to generate an augmented confidence score.

12. The system of claim 11, wherein the syllable recognition system comprises a second portion of the memory device, the second portion of the memory device for storing program instruction steps for execution by the processor to perform short-term tonal analysis of the input signal and for generating N-best theories with initial confidence scores.

13. The system of claim 12, wherein the integrated tone classifier comprises a third portion of the memory device, the third portion of the memory device for storing program instruction steps for execution by the processor to receive the N-Best theories with initial confidence scores, to perform long-term tonal analysis to determine the tone of syllables of the theories, and to generate a long-term tonal confidence signal for each theory.

14. The system of claim 10, wherein the confidence score augmentor determines an augmented confidence score by adding a long-term confidence score signal and an initial confidence score.

15. The system of claim 10, wherein the confidence score augmentor re-ranks the N-best theories according to the augmented confidence scores.

7. A system for recognizing an input signal of continuous speech of a tonal language, the system comprising:

a syllable recognition system, having an input and an output, for determining phonetic structures of syllables, for performing short-term tonal analysis of the input, and for generating N-best theories with initial confidence scores, the input of the syllable recognition system coupled to receive the input signal; and

an integrated tone classifier, having a first and second input and an output, the first input coupled to receive the input signal and the second input coupled to the output of the syllable recognition system to receive the N-best theories with initial confidence scores, for performing long-term tonal analysis to determine the tone of syllables of the theories and for generating a long-term tonal confidence signal for each theory.

16. A method for recognizing an input signal of continuous speech, having a pitch, of a tonal language, the method comprising the steps of:

determining N-best theories with initial confidence scores;

estimating the pitch contour of the input signal;

segmenting the pitch contour into units according to each of the N-best theories;

comparing the units to the tones of the tonal language;

generating a long-term confidence signal for each theory, the long-term confidence signal indicating how well the units of a theory match the tones of the tonal language; and

modifying an initial confidence score according to a long-term tonal confidence signal to generate an augmented confidence score.

17. The method of claim 16 further comprising the steps of:

ranking the N-best theories according to the augmented confidence scores; and

generating an output signal of the N-best theories with augmented initial confidence scores.

18. A system for recognizing an input signal of continuous speech of a tonal language comprising:

means for determining N-best theories with initial confidence scores;

means for estimating the pitch contour of the input signal;

means for segmenting the pitch contour into units according to each of the N-best theories;

means for comparing the units to the tones of the tonal language; and

means for modifying an initial confidence score according to a long-term tonal confidence signal to generate an augmented confidence score.

19. The system of claim 18 further comprising means for generating a long-term tonal confidence signal for each theory.

20. The system of claim 19 further comprising:

means for ranking the N-best theories according to augmented confidence scores; and

means for generating an output signal of the N-best theories with augmented confidence scores.