Method and apparatus for speech analysis and synthesis using lattice ladder notch filters

• US 5,940,791 A
• Filed: 05/09/1997
• Issued: 08/17/1999
• Est. Priority Date: 05/09/1997
• Status: Expired due to Fees

• Current Assignee: Washington University In St Louis
• Original Assignee: Washington University In St Louis
• Inventors: Lindquist, Anders, Byrnes, Christopher I.
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): ZINTEL, HAROLD ALBERT
• Application Number:
  US08/854,150
• Time in Patent Office: 830 Days
• Field of Search: 704/258, 704/500, 704/219
• US Class Current: 704/219
• CPC Class Codes: G10L 17/02 Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
• CPC Class Codes: G10L 19/02 using spectral analysis, e.g. transform vocoders or subband vocoders
• CPC Class Codes: G10L 25/48 specially adapted for particular use

Abstract

A method and apparatus for analyzing and synthesizing speech includes a programmable lattice-ladder notch filter which may be programmed to exhibit both filter poles and filter zeros and thereby exhibit a power spectral density with a better fit to that of a speech frame such that, when energized by a selected signal sample, a more accurate regeneration of speech is achieved. The filter parameters may be reliably and systematically determined as a single solution to a mathematical analysis given a set of gain parameters matching the observed covariance data and having a prescribed set of transmission zeros. These transmission zeros may either be preselected as a design specification, or recovered from analysis of the speech data. A speech frame may be analyzed and provide a set of parameters which may be transmitted to a remote location where a synthesizer may accurately reproduce the speech frame.

Claims

1. A speech analyzer for speech segmented into a plurality of speech frames comprising:

means for determining filter zeros and covariances from a speech frame, and means for calculating a set of gains for an LLN filter which match said filter zeros and covariances so that as said LLN filter is excited by a selected signal said speech frame is regenerated.

2. The speech analyzer of claim 1 wherein said calculating means includes means for computing a set of filter parameters from the filter zeros and filter covariances.

3. The speech analyzer of claim 2 further comprising a signal selector for determining said excitation signal.

4. The speech analyzer of claim 3 further comprising means for determining whether said speech frame is voiced or unvoiced.

5. The speech analyzer of claim 2 wherein said calculating means includes means for optimizing an objective function which incorporates the filter zeros and filter covariances.

6. The speech analyzer of claim 5 further comprising means for determining an initial condition for said set of gains for said LLN filter.

7. The speech analyzer of claim 6 wherein said optimizing means includes means for using only algorithms which are defined solely in terms of arithmetic operations.

8. The speech analyzer of claim 2 further comprising a speech synthesizer in communication with said speech analyzer, said speech synthesizer including a programmable LLN filter, and an excitation generator connected to said LLN filter for generating said selected signal so that as said programmable filter is configured in accordance with said calculated set of gains, said speech frame is regenerated by said selected signal.

9. The speech analyzer of claim 8 wherein said speech synthesizer is remote from said speech analyzer and further comprising a telecommunications data link therebetween.

10. The speech analyzer of claim 8 wherein said excitation generator includes means for producing signals corresponding to either voiced or unvoiced speech.

11. The speech analyzer of claim 8 wherein said excitation generator includes a look up table having a plurality of entries and a corresponding code for each of said table entries so that said speech analyzer need only determine and specify a look up table entry to thereby specify a desired signal.

12. The speech analyzer of claim 11 wherein said speech analyzer further comprises means for determining a look up table entry corresponding to a signal which most closely produces the speech frame when used to excite the LLN filter.

13. The speech analyzer of claim 1 further comprising a signal selector for determining said excitation signal.

14. The speech analyzer of claim 13 further comprising means for determining whether said speech frame is voiced or unvoiced.

15. The speech analyzer of claim 1 wherein said calculating means includes means for optimizing an. objective function which incorporates the filter zeros and filter covariances.

16. The speech analyzer of claim 1 further comprising means for determining an initial condition for said set of gains for said LLN filter.

17. The speech analyzer of claim 1 wherein said optimizing means includes means for using only algorithms which are defined solely in terms of arithmetic operations.

18. A speech synthesizer having a programmable LLN filter, and an excitation generator for generating a pre-selected signal for exciting said LLN filter to thereby regenerate a desired frame of speech.

19. The speech synthesizer of claim 18 further comprising a telemetry link for receiving a set of filter parameters for programming said LLN filter and a code indicative of a desired excitation signal.

20. The speech synthesizer of claim 18 further comprising means for programming said LLN filter in response to said received filter parameters and wherein said excitation generator is responsive to said code.

21. The speech synthesizer of claim 20 further comprising a D/A converter for converting the output of said LLN filter to an analog waveform.

22. A method for analyzing a frame of speech comprising the steps of:

determining a set of filter zeros and covariances from said speech frame, and calculating therefrom a set of filter parameters comprising a set of gains for an LLN filter which match said filter zeros and covariances so that as said LLN filter is excited by a preselected signal said speech frame is regenerated.

23. The method of claim 22 wherein the calculating step includes the step of computing a set of filter poles from the filter zeros and covariances.

24. The method of claim 23 wherein the calculating step further includes the step of optimizing an objective function which incorporates the filter zeros and covariances.

25. The method of claim 24 further comprising the step of determining an initial condition for said set of gains for said LLN filter.

26. The method of claim 25 wherein the optimizing step includes the step of using only algorithms which are defined solely in terms of arithmetic operations.

27. The method of claim 23 further comprising the step of determining a desired signal for exciting said LLN filter to regenerate said speech frame.

28. The method of claim 22 further comprising the step of transmitting the filter parameters and a code corresponding to the desired excitation signal to a speech synthesizer for regeneration of said speech frame.

29. A method for regenerating a frame of speech comprising the steps of:

programming a programmable LLN filter, generating a pre-selected signal for exciting said LLN filter, and exciting said LLN filter with said generated signal to thereby regenerate said frame of speech.

30. The method of claim 29 further comprising the steps of receiving a set of filter parameters for programming said LLN filter, programming said LLN filter in accordance therewith, and receiving a code indicative of said generated signal.

31. The method of claim 30 further comprising the step of converting the output of said LLN filter to an analog waveform.

32. A speech analyzer including a circuit for detecting a frame of speech, and a translator of converting said speech frame into a set of LLN filter parameters, said LLN filter parameters being suitable for reproduction of said speech frame through excitation of an LLN filter configured with said filter parameters.