Method of analyzing input speech and speech analysis apparatus therefor

• US 4,827,516 A
• Filed: 10/10/1986
• Issued: 05/02/1989
• Est. Priority Date: 10/16/1985
• Status: Expired due to Fees

• Current Assignee: Toppan Printing Co., Ltd. (TOPPAN Holdings, Inc.)
• Original Assignee: Toppan Printing Co., Ltd. (TOPPAN Holdings, Inc.)
• Inventors: Tanabe, Masao, Yamaguchi, Mikio, Tsukahara, Yusuke, Masuda, Hitoshi
• Primary Examiner(s): Salce, Patrick R.
• Assistant Examiner(s): VOELTZ, EMANUEL T
• Application Number:
  US06/917,509
• Time in Patent Office: 935 Days
• Field of Search: 381/36-41, 381/45, 381/48-50, 364/513.5
• US Class Current: 704/224
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence

Abstract

A speech analysis apparatus according to the invention, comprising a transforming section for receiving a spectrum envelope, for transforming the spectrum envelope such magnitude data thereof becomes suitable, and for generating a transformed spectrum envelope, an integrator for receiving the transformed spectrum envelope output from the transforming section, for integrating the input spectrum envelope with respect to a predetermined variable, and for outputting an integrated spectrum envelope, and a projection circuit for receiving the transformed spectrum envelope from the transform circuit and the integrated spectrum envelope from the integrator, and for projecting the spectrum envelope with respect to the integrated data. Therefore, the analysis result inherent to the phoneme can be obtained regardless of vocal tract lengths. The spectrum envelope to be projected can be integrated by the integrator, along the frequency axis or the mel axis. The analysis apparatus further includes a spectrum envelope-extractor for obtaining the spectrum envelope, by using cepstrum analysis and smoothing the resultant spectrum envelope. A spectrum envelope in the transition from a consonant to a vowel can be obtained.

Claims

1. A speech analysis apparatus capable of obtaining an analysis result inherent to a phoneme, regardless of vocal tract lengths, comprising:

transforming means for receiving a spectrum envelope representing intensity data of a spectrum corresponding to a speech signal, and for proportionally transforming the received spectrum envelope such that maximum intensity data of the spectrum envelope is substantially equal to a predetermined intensity data;

integrating means for integrating the transformed spectrum envelope from said transforming means with respect to a predetermined variable from a first value of the variable to a second value of the variable obtain integral data at the second value; and

projecting means for projecting the transformed spectrum envelope with respect to the integrated data by varying the second value from the first value to a third value of the variable, and for obtaining a relationship between intensity data of the transformed spectrum envelope at the second value and the integral data at the second value.

2. The apparatus according to claim 1, wherein said transforming means comprises:

logarithm means for taking a logarithm of the intensity data of the received spectrum envelope to obtain a logarithmic spectrum envelope; and

normalizing means for normalizing the logarithmic spectrum envelope according to the predetermined intensity data to obtain the transformed spectrum envelope.

3. The apparatus according to claim 1, wherein said transforming means comprises:

normalizing means for normalizing the received spectrum envelope according to the predetermined intensity data; and

logarithm means for taking a logarithm of the intensity data of the normalized spectrum envelope, such that the maximum intensity data of the normalized spectrum envelope is substantially equal to the predetermined intensity data to obtain the transformed spectrum envelope.

4. The apparatus according to claim 1, wherein said transforming means comprises means for taking a logarithm of the received spectrum envelope such that the maximum intensity data of the spectrum envelope is substantially equal to the predetermined intensity data, to obtain the transformed spectrum envelope.

5. The apparatus according to claim 1, wherein said transforming means comprises means for equalizing the spectrum envelope, such that a total power of the received envelope is substantially equal to a predetermined power, to obtain the transformed spectrum envelope.

6. The apparatus according to claim 1, wherein said variable is frequency.

7. The apparatus according to claim 6, wherein the first value is a value ranging from 10 to 100 Hz.

8. The apparatus according to claim 1, wherein said variable is mels.

9. The apparatus according to claim 1, wherein the variable is a logarithmic frequency.

10. The apparatus according to claim 9, wherein the first value is a logarithmic value ranging from 10 to 100 Hz.

11. The apparatus according to claim 8, wherein the first value is zero.

12. The apparatus according to claim 1, further comprising spectrum envelope extracting means for extracting the spectrum envelope from an input speech signal, and for outputting the spectrum envelope to said transforming means.

13. The apparatus according to claim 12, wherein said spectrum envelope extracting means comprises:

Fourier transforming means for Fourier transforming the speech signal to obtain power spectra;

pitch frequency detecting means for detecting a pitch frequency of the speech signal from the power spectra;

spectrum extracting means for extracting from the power spectra each partial power spectrum within a frequency range defined as;

space="preserve" listing-type="equation">fP×

n-fP/2<

fA<

fP×

n+fP/2,

where fp is the detected pitch frequency and n is 0 or a positive integer; and

inverse Fourier transforming means for performing inverse Fourier transformation of each partial power spectrum extracted by said spectrum extracted means.

14. The apparatus according to claim 13, wherein said pitch frequency detecting means includes cepstrum analysis means for performing cepstrum analysis of the power spectra from said Fourier transforming means, to detect the pitch frequency.

15. The apparatus according to claim 13, wherein said spectrum envelope extracting means further comprises smoothing means for smoothing the spectrum envelope from said inverse Fourier transforming means.

16. The apparatus according to claim 12, wherein said spectrum envelope extracting means comprises:

high band enhancing means for enhancing the speech signal, such that a level of a high-frequency component of the speech signal is substantially equal to that of a low-frequency component of the speech signal;

a plurality of extracting means each for passing a portion of the high band enhanced speech signal within a bandwidth limited in accordance with input first control data, for full-wave rectifying the passed portion of the enhanced speech signal, and for extracting a low frequency component of the passed portion in accordance with the first control data;

a plurality of low-pass filter means respectively coupled to said plurality of rectifying means, for detecting a temporal change in peak value of the full-wave rectified part from said plurality of rectifying means, in accordance with the first multiplexing means for sequentially outputting the low frequency components from said plurality of extracting means in accordance with input second control data to obtain the spectrum envelope; and

control means for outputting the first control data to said plurality of extracting means and the second control data to said multiplexing means in response to an input speech analysis instruction.

17. The apparatus according to claim 16, wherein said control means includes means for outputting the first control data to said plurality of extracting means such that the spectrum envelope with respect to frequency is obtained.

18. The apparatus according to claim 16, wherein said control means includes means for outputting the first control data to said plurality of extracting means such that the spectrum envelope with respect to mels is obtained.

19. The apparatus according to claim 16, wherein said control means further comprises means for generating third control data in response to the speech analysis instruction;

and

said relation obtaining means comprises;

first A/D converting means converting the integral data from said integrating means, into digital integral data;

second A/D converting means for converting the intensity data of the transformed spectrum envelope from said transforming means, into digital intensity data; and

storage means for storing the digital intensity data in accordance with the digital integral data.

20. The apparatus according to claim 19, further comprising display means for displaying the digital intensity data stored in said storage means, with respect to the digital integral data.

21. A method of obtaining an analysis result inherent to a phoneme, regardless of vocal tract lengths comprising:

receiving a spectrum envelope to proportionally transform the spectrum envelope such that maximum intensity data of the received spectrum envelope is substantially equal to predetermined intensity data;

integrating the transformed spectrum envelope with respect to a predetermined variable from a first value of the variable to a second value of the variable to obtain data at the second value;

varying the second value from the first value to a third value of the variable to obtain a relationship between intensity data of the transformed spectrum envelope at the second value and the integral data at the second value.

22. The method according to claim 21, wherein said transforming of the spectrum envelope comprises taking a logarithm of the received spectrum envelope.

23. The method according to claim 21, wherein the variable is either frequency or mels.

24. The method according to claim 21, further comprising extracting the spectrum envelope from an input speech signal.

25. The method according to claim 24, wherein said extracting of the spectrum envelope comprises:

Fourier-transforming the speech signal to obtain power spectra;

detecting a pitch frequency of the speech signal from the obtained power spectra;

extracting from the power spectra each partial power spectrum within a frequency range defined as;

space="preserve" listing-type="equation">fP×

n-fP/2<

fA<

fP×

n+fP/2

where fP is the detected pitch frequency and n is 0 or a positive integer; and

performing inverse Fourier transformation of each partial power spectrum to obtain the spectrum envelope.

26. The method according to claim 25, wherein said detecting of the pitch frequency includes performing cepstrum analysis of the power spectra.

27. The method according to claim 25, wherein said extracting of the spectrum envelope further comprises smoothing the spectrum envelope.

28. An apparatus capable of extracting a spectrum envelope from an input speech signal, comprising:

Fourier transforming means for Fourier transforming the speech signal to obtain a power spectra;

pitch frequency detecting means for detecting a pitch frequency of the speech signal from the obtained power spectra;

spectrum extracting means for extracting from the obtained power spectra each partial power spectrum within a frequency range defined as;

space="preserve" listing-type="equation">fP×

n-fP/2<

fA<

fP×

n+fP/2

where fP is the detected pitch frequency and n is 0 or a positive integer; and

inverse Fourier transforming means for performing inverse Fourier transformation of each partial power spectrum to obtain the spectrum envelope.

29. The apparatus according to claim 28, wherein said pitch frequency detecting means includes cepstrum analysis means for performing cepstrum analysis of the power spectra transformed by said Fourier transforming means, to detect the pitch frequency.

30. The apparatus according to claim 29, wherein said spectrum envelope extracting means further comprises smoothing means for smoothing the spectrum envelope from said inverse Fourier transforming means.