METHOD AND APPARATUS FOR LOCATING SPEECH KEYWORD AND SPEECH RECOGNITION SYSTEM

• US 20100094626A1
• Filed: 09/27/2007
• Published: 04/15/2010
• Est. Priority Date: 09/27/2006
• Status: Active Grant

• Current Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Original Assignee: Sharp Kabushiki Kaisha (Hon Hai Precision Industry Co., Ltd.)
• Inventors: Li, Fengqin, Yang, Qinqtao, Chen, Chen, Wu, Yadong
• Granted Patent: US 8,255,215 B2
• Time in Patent Office: Days
• US Class Current: 704/234
• 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 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

It is an object of the present invention to provide a method and apparatus for locating a keyword of a speech and a speech recognition system. The method includes the steps of: by extracting feature parameters from frames constituting the recognition target speech, forming a feature parameter vector sequence that represents the recognition target speech; by normalizing of the feature parameter vector sequence with use of a codebook containing a plurality of codebook vectors, obtaining a feature trace of the recognition target speech in a vector space; and specifying the position of a keyword by matching prestored keyword template traces with the feature trace. According to the present invention, a keyword template trace and a feature space trace of a target speech are drawn in accordance with an identical codebook. This causes resampling to be unnecessary in performing linear movement matching of speech wave frames having similar phonological feature structures. This makes it possible to improve the speed of location and recognition while ensuring the precision of recognition.

Claims

1. A method for locating a keyword contained in a recognition target speech, comprising the steps of:

by extracting feature parameters from frames constituting the recognition target speech, forming a feature parameter vector sequence that represents the recognition target speech;

by normalizing of the feature parameter vector sequence with use of a codebook containing a plurality of codebook vectors, obtaining a feature trace of the recognition target speech in a vector space; and

specifying the position of a keyword by matching prestored keyword template traces with the feature trace.

2. The method as set forth in claim 1, wherein the normalizing step includes the step of displaying the feature trace by extracting representative feature parameter vectors from the feature parameter vector sequence.

3. The method as set forth in claim 2, wherein the step of extracting the representative feature parameter vectors includes the steps of:

searching the codebook for codebook vectors that are closest to the feature parameter vectors;

displaying the feature parameter vectors with use of numbers of the codebook vectors thus found; and

displaying the feature parameters by combining feature parameter vectors represented by continuously identical codebook vectors.

4. The method as set forth in claim 2, wherein the matching step includes the steps of:

by using each of the keyword template traces, sequentially calculating, with respect to each trace frame, distances between (i) template feature vectors that represent the keyword template trace and (ii) the representative feature parameter vectors that represent the feature trace, the trace frame representing the feature trace on a time axis;

specifying a minimum value of the distances; and

specifying, as the position of the keyword, a position that a keyword template trace corresponding to the minimum value occupies in the vector space.

5. The method as set forth in claim 4, wherein the feature parameter vectors include at least one parameter selected from the group consisting of a cepstral parameter, an energy value of the cepstral parameter, a first derivative of the cepstral parameter, a second derivative of the cepstral parameter, a first derivative of the energy value, and a second derivative of the energy value.

6. The method as set forth in claim 4, wherein each of the distances is displayed with use of a codebook vector distance, the code book vector distance being a distance between any two of the codebook vectors.

7. The method as set forth in claim 6, wherein the codebook vector distance is stored in advance in a matrix form.

8. The method as set forth in claim 1, wherein the keyword template traces are generated in accordance with the codebook vectors.

9. The method as set forth in claim 8, wherein each of the keyword template traces is obtained by the steps of:

obtaining a phonemic name of each syllable of the keyword in accordance with correspondence between syllables and phonemic models;

obtaining each state of the keyword in accordance with correspondence between the phonemic models and states and forming a state vector;

displaying a codebook vector sequence number of a phonemic model with use of a number corresponding to the state vector contained in the codebook;

combining codebook vector sequence numbers of phonemic models represented by continuously identical codebook vector numbers; and

obtaining the keyword template trace after concatenating the codebook vector sequence numbers.

10. The method as set forth in claim 8, wherein each of the keyword template traces is obtained by the steps of:

forming a parameter vector sequence of an input keyword by dividing a speech wave of the input keyword into speech wave frames and then by extracting feature parameter vectors from the speech wave frames; and

by using the codebook to normalize the feature parameter vector sequence, forming a template vector sequence that represents the keyword template trace.

11. The method as set forth in claim 4, wherein the codebook is obtained by the steps of:

extracting, from a phonetic model based on a hidden Markov model, statistical parameters that represent states of each phonemic model and then forming a feature vector of each of the states; and

foil ling each of the codebook vectors of the codebook by displaying the feature vector of each of the states by number.

12. The method as set forth in claim 11, wherein the phonemic model is generated by a learning corpus of a plurality of application regions.

19. The method as set forth in claim 3, wherein the matching step includes the steps of:

by using each of the keyword template traces, sequentially calculating, with respect to each trace frame, distances between (i) template feature vectors that represent the keyword template trace and (ii) the representative feature parameter vectors that represent the feature trace, the trace frame representing the feature trace on a time axis;

specifying a minimum value of the distances; and

specifying, as the position of the keyword, a position that a keyword template trace corresponding to the minimum value occupies in the vector space.

13. An apparatus for locating a keyword contained in a recognition target speech, comprising:

a feature extracting device for, by extracting feature parameters from frames constituting the recognition target speech, forming a feature parameter vector sequence that represents the recognition target speech;

a normalizing device for, by normalizing of the feature parameter vector sequence with use of a codebook containing a plurality of codebook vectors, obtaining a feature trace of the recognition target speech in a vector space; and

a matching device for specifying the position of a keyword by matching prestored keyword template traces with the feature trace.

14. The apparatus as set forth in claim 13, wherein the normalizing device displays the feature trace by extracting representative feature parameter vectors from the feature parameter vector sequence.

15. The apparatus as set forth in claim 14, wherein the normalizing device searches the codebook for codebook vectors that are closest to the feature parameter vectors, displays the feature parameter vectors with use of numbers of the codebook vectors thus found, and displays the feature parameters by combining feature parameter vectors represented by continuously identical codebook vectors.

16. The apparatus as set forth in claim 14, wherein the matching device uses each of the keyword template traces to sequentially calculate, with respect to each trace frame, distances between (i) template feature vectors that represent the keyword template trace and (ii) the representative feature parameter vectors that represent the feature trace, the trace frame representing the feature trace on a time axis;

specifies a minimum value of the distances; and

specifies, as the position of the keyword, a position that a keyword template trace corresponding to the minimum value occupies in the vector space.

17. A speech recognition system comprising:

an apparatus as set forth in claim 13; and

a speech recognition device for recognizing the content of a keyword in accordance with the position of the keyword, the keyword being specified by the apparatus.

18. The speech recognition system as set forth in claim 17, wherein the speech recognition device performs recognition in accordance with a hidden Markov model.

20. The apparatus as set forth in claim 15, wherein the matching device uses each of the keyword template traces to sequentially calculate, with respect to each trace frame, distances between (i) template feature vectors that represent the keyword template trace and (ii) the representative feature parameter vectors that represent the feature trace, the trace frame representing the feature trace on a time axis;

specifies a minimum value of the distances; and

specifies, as the position of the keyword, a position that a keyword template trace corresponding to the minimum value occupies in the vector space.