Method and apparatus for speaker recognition using lattice-ladder filters

• US 6,256,609 B1
• Filed: 08/05/1998
• Issued: 07/03/2001
• Est. Priority Date: 05/09/1997
• Status: Expired due to Term

• Current Assignee: Washington University In St Louis
• Original Assignee: Washington University In St Louis
• Inventors: Byrnes, Christopher I., Lindquist, Anders
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): ZINTEL, HAROLD ALBERT
• Application Number:
  US09/117,721
• Time in Patent Office: 1,063 Days
• Field of Search: 704/246, 704/247, 704/250, 704/219
• US Class Current: 704/246
• 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 speech analysis and synthesis, including speaker recognition, 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. A method and apparatus for speaker identification, and speaker verification with a smart card are disclosed implementing the lattice-ladder notch (LLN) filter methodology.

Claims

1. An enrollment device for enrolling a person into a speaker recognition system, said device including an LLN filter analyzer for converting said person'"'"'s voiced speech into a set of LLN filter parameters, and a storage medium for storing said LLN filter parameters for later retrieval and comparison with voiced speech of any person seeking to access said speaker recognition system.

2. The enrollment device of claim 1 wherein the LLN filter analyzer comprises a converter for converting each frame of said person'"'"'s voiced speech into a set of filter zeros and covariances, and a set of gains which match said filter zeros and covariances.

3. The enrollment device of claim 2 wherein the LLN filter analyzer further comprises a calculator for calculating a set of filter parameters from the filter zeros and filter covariances.

4. The enrollment device of claim 1 wherein said person'"'"'s voiced speech is broken down into a plurality of frames having a predetermined order.

5. The enrollment device of claim 1 wherein said storage medium comprises a smart card.

6. The enrollment device of claim 1 wherein said storage medium comprises a data base.

8. The speaker recognition device of claim 1 wherein said LLN filter analyzer further comprises a frame identifier for identifying an appropriate set of LLN filter parameters from amongst a set of stored LLN filter parameters for comparison.

9. The speaker recognition device of claim 1 wherein said previously determined set of LLN filter parameters are stored in a storage medium.

10. The speaker recognition device of claim 9 wherein said storage medium is physically dissociated from said speaker recognition device and further comprising an input device for accepting the dissociated storage medium and accessing the LLN filter parameters stored therein.

11. The speaker recognition device of claim 10 wherein said storage medium comprises a smart card.

12. The speaker recognition device of claim 9 wherein said storage medium comprises a memory device coupled to said comparator.

13. The speaker recognition device of claim 12 wherein said storage medium comprises a memory device in which is stored a plurality of LLN filter parameters associated with a plurality of persons.

7. A speaker recognition device comprising an LLN filter analyzer for analyzing a person'"'"'s voiced speech and converting it into a set of LLN filter parameters, and a comparator for comparing at least one parameter from said voiced speech LLN filter parameters with at least one parameter from a previously determined set of LLN filter parameters to make a speaker verification or identification decision.

14. The speaker recognition device of claim 7 wherein said person'"'"'s voiced speech is broken down into a plurality of frames, and wherein the converter converts each frame of voiced speech into LLN filter parameters including at least A(z) and B(z).

15. The speaker recognition device of claim 14 wherein the previously determined LLN filter parameters include at least a frame of speech and its corresponding set of A(z) and B(z).

16. The speaker recognition device of claim 15 wherein said frames are substantially arranged in a predetermined order.

17. A speaker recognition device comprising a comparator for comparing at least one of a set of LLN filter parameters for a person'"'"'s voiced speech with at least one of a stored set of previously determined voiced speech LLN filter parameters, and an analyzer for determining whether said LLN filter parameters have sufficiently matching patterns so as to be considered to be the same.

18. The speaker recognition device of claim 17 wherein said LLN filter parameters are correlated with individual frames of speech.

19. The speaker recognition device of claim 18 wherein LLN filter parameters are associated into a predetermined order so that the speaker is required to speak substantially an identical series of frames of speech as has been previously used to create the previously determined set of voiced speech LLN filter parameters in order to achieve a sufficiently matching pattern so as to be considered the same.

20. A method for recognizing a speaker comprising the steps of converting voiced speech into a set of LLN filter parameters, and comparing at least one parameter of said LLN filter parameters with at least one parameter of a previously stored set of LLN filter parameters.

21. The method of claim 20 further comprising the step of pattern matching said LLN filter parameters to determine if there is a match so as to verify the identity of the speaker.

22. The method of claim 21 further comprising the step of individually inputting a plurality of sets of stored LLN filter parameters prior to the comparison thereof with the voiced speech LLN filter parameters.

23. The method of claim 22 further comprising the step of accessing a data base in order to obtain the stored LLN filter parameters prior to comparison with the voiced speech LLN filter parameters.

24. The method of claim 23 further comprising the steps of associating the stored LLN filter parameters with a particular frame of speech in a particular order and using the order of frames to help determine which frames of voiced speech are to be matched with which stored frames.

25. A method for enrolling a person into a group of permitted enrollees of a speaker recognition system, said method comprising the steps of converting said persons voiced speech into a set of LLN filter parameters, and storing said set of LLN filter parameters into a storage medium for later retrieval and comparison.

26. The method claim 25 wherein said stored set of LLN filter parameters are comprised of coefficients of a zero polynomial and a polynomial, and further comprising the step of associating said coefficients of said stored set of LLN filter parameters with a particular frame of voiced speech.

27. The method of claim 26 further comprising the step of correlating the frames into a particular order for later comparison in that order.