System and method for acoustic signature extraction, detection, discrimination, and localization

• US 8,271,200 B2
• Filed: 03/22/2006
• Issued: 09/18/2012
• Est. Priority Date: 12/31/2003
• Status: Active Grant

• Current Assignee: Reality Analytics, Inc. (Renesas Electronics Corporation)
• Original Assignee: Jeffrey M Sieracki
• Inventors: Sieracki, Jeffrey M.
• Primary Examiner(s): Moran, Marjorie
• Assistant Examiner(s): Skibinsky, Anna
• Application Number:
  US11/387,034
• Publication Number:
  US 20060241916A1
• Time in Patent Office: 2,372 Days
• Field of Search: None
• US Class Current: 702/19
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 25/00 Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 muting semiconductor-based amplifiers when some special characteristics of a signal are sensed by a speech detector, e.g. sensing when no signal is present, H03G3/34

Abstract

A system and method whereby acoustic signals can be classified and identified as to nature and location of the original signal. The system and method determine from an arbitrary set of signals a signature or other characterizing feature and distinguish signals associated with a plurality of conditions by means of dictionaries comprising atoms of signals.

Claims

1. An apparatus for extracting, detecting, classifying, discriminating, localizing, and processing acoustic signals comprising:

a) at least one transducer device operable to collect a plurality of signals;

b) instructions for decomposing the signals into matched structure books wherein decomposing comprises performing simultaneous sparse approximations of the signals;

each said structure book representation defining a multi-dimensional combination of coefficients and atoms; and

c) a computer memory for storing said structure-book.

2. The apparatus of claim 1 wherein the at least one transducer device capable of collecting a plurality of signals includes one or more acoustic sensors.

3. The apparatus of claim 1 wherein the processing device executes a Greedy Adaptive Discrimination (GAD) decomposition algorithm operating with a dictionary of prototype atoms.

4. The apparatus of claim 3 wherein the dictionary comprises a set of Gabor atoms together with the Fourier and delta function bases.

5. The apparatus of claim 3 wherein the dictionary includes a random set of atoms.

6. The apparatus of claim 3 wherein each atom selected for the structure book representations by the GAD decomposition is caused to be orthogonal from all previously chosen atoms.

7. A method of signature extraction comprising the steps of:

a. establishing two or more analysis sets, each set comprising at least one signal;

b. decomposing the sets of signals into matched structure books wherein decomposing comprises performing simultaneous sparse approximations of the signals;

each said structure book representation defining a multi-dimensional combination of coefficients and atoms; and

c. actuating the processing device to produce a parametric mean structure book representative of a signature for each group of signals.

8. The method of claim 7 wherein a separate simultaneous sparse approximation is formed for each analysis set.

9. The method of claim 7 wherein a single simultaneous sparse approximation is formed for the aggregate collection of signals from all analysis sets.

10. A method of signal detection and classification comprising the steps of:

a. collecting a set of signals;

b. decomposing the sets of signals into matched structure books wherein decomposing comprises performing simultaneous sparse approximations of the signals;

each said structure book defining a multi-dimensional combination of coefficients and atoms; and

c. comparing the resulting structure books to structure books of known signatures.

11. The method of claim 10 wherein a single simultaneous sparse approximation is formed for the aggregate collection of signals from all analysis sets.

12. The method of claim 10 wherein step c is accomplished by examining the values of the coefficients of the atoms in the resulting structure books to typical values for similar atoms of one or more known signatures.

13. The method of claim 10 wherein the simultaneous sparse approximation is accomplished with respect to a dictionary comprising one or more known signature waveforms.

14. The method of claim 10 wherein the simultaneous sparse approximation is accomplished with respect to a dictionary comprising the aggregate collection of atoms present in previously extracted structure books of known signatures of interest.

15. The method of claim 14 wherein the dictionary additionally includes the aggregate collection of all atoms selected during the sparse approximation that produced the previously extracted structure books of known signatures of interest, whether or not said atoms are present in said structure books of signatures of interest.

16. The method of claim 10 wherein a plurality of signals are collected and at least one of the signals is time-shifted relative to at least one other signal before the simultaneous sparse approximation operation.

17. The method of claim 10 wherein:

a. the collected signal or signals are sparsely approximated in aggregate with at least one previously established signal; and

b. the structure books of the previously established signal or signals are compared to those of the newly collected signal or signals in order to classify the new signal or signals as either similar or different from the previous group.

18. The method of claim 10 wherein a plurality of signals are collected at different locations and, subsequent to detection, time and phase information from the resulting structure books at different locations are used to extract directional information about the signals.

19. The method of claim 10 wherein:

a. a plurality of signals are collected at different locations;

b. each signal is analyzed independently to detect signatures;

c. detected signatures are reconstructed; and

d. each channel is searched for the reconstructed signature to determine time of arrival information.