System and method for identification of a speaker by phonograms of spontaneous oral speech and by using formant equalization using one vowel phoneme type

• US 9,047,866 B2
• Filed: 03/23/2012
• Issued: 06/02/2015
• Est. Priority Date: 09/24/2009
• Status: Active Grant

• Current Assignee: Obschestvo S Ogranichennoi Otvetstvennost'Yu "Centr Rechevyh Tehnologij", Speech Technology Center Limited
• Original Assignee: Speech Technology Center Limited
• Inventors: Koval, Sergey Lvovich
• Primary Examiner(s): Kazeminezhad, Farzad
• Application Number:
  US13/429,260
• Publication Number:
  US 20120232899A1
• Time in Patent Office: 1,166 Days
• Field of Search: 704/246, 704/241, 704/238
• US Class Current: 1/1
• 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 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/14 Use of phonemic categorisation or speech recognition prior to speaker recognition or verification
• CPC Class Codes: G10L 17/20 Pattern transformations or operations aimed at increasing system robustness, e.g. against channel noise or different working conditions
• CPC Class Codes: G10L 25/15 the extracted parameters being formant information

Abstract

A system and method for identification of a speaker by phonograms of oral speech is disclosed. Similarity between a first phonogram of the speaker and a second, or sample, phonogram is evaluated by matching formant frequencies in referential utterances of a speech signal, where the utterances for comparison are selected from the first phonogram and the second phonogram. Referential utterances of speech signals are selected from the first phonogram and the second phonogram, where the referential utterances include formant paths of at least three formant frequencies; wherein the first two formants are within typical variability limits for one vowel phoneme type. The selected referential utterances including at least two identical formant frequencies are compared therebetween. Similarity of the compared referential utterances from matching other formant frequencies is evaluated, where similarity of the phonograms is determined from evaluation of similarity of all the compared referential utterances.

Claims

1. A method for identification of a speaker by phonograms of oral speech, the method comprising:

Selecting, via a computing device, referential utterances of speech signals from a first phonogram of the speaker and a second, or sample, phonogram, wherein the referential utterances comprise formant paths of at least three formant frequencies, and values of the first two formants of the formant paths are within given typical variability limits of formant frequency values for a specific type of vowel phoneme in a given language;

Comparing, via the computing device, out of the selected referential utterances those selected referential utterances comprising at least two formant frequencies having the same values;

and evaluating, via the computing device, similarity of the compared referential utterances on the basis of comparison of other formant frequencies out of the at least three formant frequencies, wherein similarity of the phonograms is determined from evaluation of similarity of all the compared referential utterances.

2. The method according to claim 1, wherein the formant frequencies in each of the selected referential utterance are calculated as average values for fixed time intervals in which the formant frequencies are relatively constant.

3. The method according to claim 1, wherein the selection from the phonograms for comparison is conducted in relation to at least two referential utterances of a speech signal related to sounds articulated as differently as possible with maximum and minimum frequency values for the first and the second formants in the given phonogram.

4. The method according to claim 1, wherein,

before calculating values of formant frequencies,

subjecting a power spectrum of the speech signal for each phonogram to inverse filtering,

wherein the time average for each frequency component of the power spectrum is calculated, at least for particular utterances of the phonogram, and then

a power spectrum of the phonogram for each frequency component of the spectrum is divided by its inverse mean value.

5. The method according to claim 1, wherein, before calculating values of formant frequencies,

subjecting a power spectrum of a speech signal for each phonogram to inverse filtering,

wherein the time average for each frequency component of the power spectrum is calculated, at least for individual utterances of the phonogram, and then

an original value logarithm of the spectra is taken, and an average value logarithm of the phonogram signal power spectrum for each frequency component is subtracted from its original value logarithm.

6. A system for identification of a speaker by phonograms of oral speech, the system comprising:

a computer memory configured to store digital audio signal files representative of a plurality of phonograms converted into digital form;

a computing device configured to;

select referential utterances of speech signals from a first phonogram of the speaker and a second, or sample, phonogram, wherein the referential utterances comprise formant paths of at least three formant frequencies, and values of the first two formants of the formant paths are within given typical variability limits of formant frequency values for a specific type of vowel phoneme in a given language;

compare out of the selected referential utterances those selected referential utterances comprising at least two formant frequencies values having the same values; and

evaluate similarity of the compared referential utterances on the basis of comparison of other formant frequencies out of the at least three formant frequencies, wherein similarity of the phonograms is determined from evaluation of similarity of all the compared referential utterances.

7. The system according to claim 6, wherein the formant frequencies in each of the selected referential utterance are calculated as average values for fixed time intervals in which the formant frequencies are relatively constant.

8. The system according to claim 6, wherein the selection from the phonograms for comparison is conducted in relation to at least two referential utterances of a speech signal related to sounds articulated as differently as possible with maximum and minimum frequency values for the first and the second formants in the given phonogram.

9. The system according to claim 6, wherein the computing device additionally subjects a power spectrum of the speech signal for each phonogram to inverse filtering before calculating values of formant frequencies, wherein the time average for each frequency component of the power spectrum is calculated, at least for particular utterances of the phonogram, and then a power spectrum of the phonogram for each frequency component of the spectrum is divided by its inverse mean value.

10. The system according to claim 6, wherein the computing device additionally subjects a power spectrum of a speech signal for each phonogram to inverse filtering before calculating values of formant frequencies,

wherein the time average for each frequency component of the power spectrum is calculated, at least for individual utterances of the phonogram, and then

an original value logarithm of the spectra is taken, and an average value logarithm of the phonogram signal power spectrum for each frequency component is subtracted from its original value logarithm.

11. A system for identification of a speaker by phonograms of oral speech, the system comprising:

means for selecting referential utterances of speech signals from a first phonogram of the speaker and a second, or sample, phonogram, wherein the referential utterances comprise formant paths of at least three formant frequencies, and values of the first two formants of the formant paths are within the given typical variability limits of formant frequency values for a specific type of vowel phoneme in a given language;

means for comparing out of the selected referential utterances those selected referential utterances comprising at least two formant frequencies having the same values;

and means for evaluating similarity of the compared referential utterances on the basis of comparison of other formant frequencies out of the at least at least three formant frequencies, wherein similarity of the phonograms is determined from evaluation of similarity of all the compared referential utterances.

12. The system according to claim 11, wherein the formant frequencies in each of the selected referential utterance are calculated as average values for fixed time intervals in which the formant frequencies are relatively constant.

13. The system according to claim 11, wherein the selecting from the phonograms for comparing is conducted in relation to at least two referential utterances of a speech signal related to sounds articulated as differently as possible with maximum and minimum frequency values for the first and the second formants in the given phonogram.

14. The system according to claim 11, additionally comprising

means for subjecting a power spectrum of the speech signal for each phonogram to inverse filtering before calculating values of formant frequencies,

wherein the time average for each frequency component of the power spectrum is calculated, at least for particular utterances of the phonogram, and then a power spectrum of the phonogram for each frequency component of the spectrum is divided by its inverse mean value.

15. The system according to claim 11, additionally comprising

means for subjecting a power spectrum of a speech signal for each phonogram to inverse filtering before calculating values of formant frequencies,

wherein the time average for each frequency component of the power spectrum is calculated, at least for individual utterances of the phonogram, and then an original value logarithm of the spectra is taken, and an average value logarithm of the phonogram signal power spectrum for each frequency component is subtracted from its original value logarithm.