System and method for taxonomically distinguishing unconstrained signal data segments

• US 9,691,395 B1
• Filed: 12/28/2012
• Issued: 06/27/2017
• Est. Priority Date: 12/31/2011
• Status: Active Grant

• Current Assignee: Reality Analytics, Inc. (Renesas Electronics Corporation)
• Original Assignee: Reality Analytics, Inc. (Renesas Electronics Corporation)
• Inventors: Sieracki, Jeffrey M.
• Primary Examiner(s): Sirjani, Fariba
• Application Number:
  US13/729,828
• Time in Patent Office: 1,642 Days
• CPC Class Codes: G06F 18/2136 based on sparsity criteria, e.g. with an overcomplete basis
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence
• 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 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
• CPC Class Codes: G10L 2021/02087 the noise being separate speech, e.g. cocktail party
• CPC Class Codes: G10L 21/0272 Voice signal separating

Abstract

A system and method are provided for taxonomically distinguishing grouped segments of signal data captured in unconstrained manner for a plurality of sources. The system comprises a vector unit constructing for each of the grouped signal data segments at least one vector of predetermined form. A sparse decomposition unit selectively executes in at least a training system mode a simultaneous sparse approximation upon a joint corpus of vectors for a plurality of signal segments of distinct sources. The sparse decomposition unit adaptively generates at least one sparse decomposition for each vector with respect to a representative set of decomposition atoms. A discriminant reduction unit executes during the training system mode to derive an optimal combination of atoms from the representative set. A classification unit executes in a classification system mode to discover for an input signal segment a degree of correlation relative to each of the distinct sources.

Claims

1. A system for taxonomically distinguishing grouped segments of signals captured in unconstrained manner by a transducer for a plurality of sources, the system comprising:

(a) a vector construction unit constructing at least one vector of predetermined form for each of the grouped signal segments;

(b) a training unit coupled to said vector construction unit, said training unit including;

a decomposition portion executing an adaptive sparse transformation upon a joint corpus of vectors for a plurality of signal segments of distinct sources, said decomposition portion generating for each said vector in said joint corpus at least one adaptive decomposition defined on a sparse transformation plane as a coefficient weighted sum of a representative set of decomposition atoms, and,

a discriminant reduction portion coupled to said decomposition portion, said discriminant reduction portion being executable to mutually associate decomposition atoms of the representative set in m-wise manner for determining a cooperative strength thereof in distinguishing one distinct source from another, within a multi-dimensional plane, and thereby derive from said representative set at least one optimal combination of atoms for cooperatively distinguishing signals attributable to different ones of the distinct sources, wherein m is greater than or equal to 2; and

,

(c) a classification unit coupled to said vector construction unit, said classification unit including;

a projection portion projecting a spectral vector of an input signal segment onto said sparse transformation plane to generate an adaptive decomposition therefor as a coefficient weighted sum of said representative set of decomposition atoms, and,

a classification decision portion coupled to said projection portion, said classification decision portion being executable to discover for said adaptive decomposition of said input signal segment a degree of similarity relative to each of the distinct sources according to the optimal combination, and to thereby determine one of the distinct sources to have generated the input signal segment according to the degree of similarity.

2. The system as recited in claim 1, wherein said adaptive sparse transformation includes a simultaneous sparse approximation executing a greedy adaptive decomposition (GAD) process referencing a Gabor dictionary.

3. The system as recited in claim 1, wherein said vector construction unit applies a Short-Time-Fourier-Transform (STFT) process upon each captured segment of signal, said vector construction unit generating a spectral vector defined in a time-frequency plane for each said segment.

4. The system as recited in claim 1, wherein captured segments of signals include spatially-captured terrestrial data of a source terrain, and the distinct sources include regions of distinct terrain characteristics.

5. The system as recited in claim 1, wherein captured segments of signals include photographically captured anatomic image data of a source organism, and the distinct sources include regions of distinct species of organisms.

6. The system as recited in claim 1, wherein said discriminant reduction portion includes a Support Vector Machine (SVM) part programmably implemented therein, said SVM part mutually k-wise comparing the distinct sources in adaptive decomposition to selectively determine one of said at least one optimal combination of atoms for each said mutual k-wise comparison.

7. The system as recited in claim 6, wherein a plurality of sub-segments are delineated within each said segment;

and, said adaptive decomposition and cepstral projection portions generate over each said sub-segment a parametric mean of said adaptive decompositions, each said sub-segment parametric mean being defined in terms of said representative set of decomposition atoms.

8. The system as recited in claim 7, wherein said SVM part determines for each said k-wise comparison of sources a m-dimensional decision subspace defined by a corresponding set of m optimal atoms;

and, said classification decision portion executes a non-parametric voting process iteratively mapping corresponding portions of said input signal segment adaptive decomposition to each said decision subspace.

9. The system as recited in claim 6, wherein:

said SVM portion executes pair-wise comparisons of distinct sources, said SVM portion determining for each said pair-wise comparison of sources a two-dimensional decision subspace defined by a corresponding pair of optimal atoms; and

,

said classification unit executes a non-parametric voting process iteratively mapping corresponding portions of said input signal segment adaptive decomposition to each said decision subspace.

10. The system as recited in claim 1, wherein at least one said signal segment is attributable to a known distinct source prior to the adaptive sparse transformation of the decomposition portion, said decomposition and discriminant reduction portions thereby executing to identify a distinct class corresponding to the known distinct source.

11. The system as recited in claim 1, wherein none of said signal segments is attributable to a known distinct source prior to the adaptive sparse transformation of the decomposition portion, said decomposition and discriminant reduction portions thereby executing to cluster together similar ones of said segments.

12. The system as recited in claim 1, wherein said vector construction unit includes a transformation portion executing spectrographic transformation upon each said captured segment of signal received thereby, said vector construction unit generating the spectral vector for each said segment.

13. The system as recited in claim 12, wherein:

said spectrographic transformation includes a Short-Time-Fourier-Transform (STFT) process, and said spectral vectors are defined in a time-frequency domain; and

,

said adaptive decompositions are each defined in a cepstral-frequency domain as a coefficient weighted sum of said representative set of atoms.

14. The system as recited in claim 1, wherein:

at least one of the vector construction unit, training unit, or classification unit is implemented as part of a mobile communication device, and

the mobile communication device captures part or all of the grouped segments of signal data.

15. A method for taxonomically distinguishing grouped segments of signals captured in unconstrained manner by a transducer for a plurality of sources, the method comprising:

constructing for each of the grouped signal segments at least one vector of predetermined form;

selectively executing in a processor an adaptive sparse approximation to generate an adaptive decomposition of each said vector, said adaptive sparse approximation in a training system mode executing upon a joint corpus of vectors for a plurality of signal segments of distinct sources, generating at least one adaptive decomposition for each said vector defined on a sparse transformation plane as a coefficient weighted sum of a representative set of decomposition atoms;

executing discriminant reduction in a processor during the training system mode to mutually associate decomposition atoms of the representative set in m-wise manner for determining a cooperative strength thereof in distinguishing one distinct source from another, within a multi-dimensional subspace, and thereby derive from said representative set at least one optimal combination of atoms for cooperatively distinguishing signals attributable to different ones of the distinct sources, wherein m is greater than or equal to 2;

projecting a spectral vector of an input signal segment onto said sparse transformation plane to generate an adaptive decomposition therefor as a coefficient weighted sum of said representative set of decomposition atoms; and

,

executing classification upon said adaptive decomposition of an input signal segment during a classification system mode, said classification including executing a processor to discover a degree of similarity for said input signal segment relative to each of the distinct sources according to the optimal combination, and to thereby determine one of the distinct sources to have generated the input signal segment according to the degree of similarity.

16. The method as recited in claim 15, wherein said discriminant reduction includes carrying out a Support Vector Machine (SVM) process mutually k-wise comparing the distinct sources in adaptive decomposition to selectively determine one of said at least one optimal combination of atoms for each said k-wise comparison.

17. The method as recited in claim 16, wherein:

said SVM portion executes pair-wise comparisons of distinct sources, said SVM process determining for each said pair-wise comparison of sources a two-dimensional decision subspace defined by a corresponding pair of optimal atoms; and

,

said classification includes a non-parametric voting process iteratively mapping corresponding portions of said input signal segment adaptive decomposition to each said decision subspace.

18. The method as recited in claim 17, wherein at least one said signal segment is attributable to a known distinct source prior to initiation of the training system mode, said adaptive sparse approximation and discriminant reduction thereby executing in the training system mode to identify a distinct class corresponding to the known distinct source.

19. The method as recited in claim 17, wherein none of said signal segments is attributable to a known distinct source prior to initiation of the training system mode, said adaptive sparse approximation and discriminant reduction thereby executing in the training system mode to cluster together similar ones of said segments.

20. The method as recited in claim 17, wherein a plurality of sub-segments are delineated within each said segment;

and, a parametric mean of said adaptive decompositions over each said sub-segment is generated, each said sub-segment parametric mean being defined in terms of said representative set of decomposition atoms.

21. The method as recited in claim 20, wherein said adaptive sparse approximation and parametric mean are carried out according to a greedy adaptive decomposition (GAD) process.

22. The method as recited in claim 16, wherein said segments of signals include spatially-captured terrestrial data of a source terrain, and the distinct sources include regions of distinct terrain characteristics.

23. The method as recited in claim 16, wherein said segments of signals include photographically captured anatomic image data of a source organism, and the distinct sources include regions of distinct species of organisms.

24. The method as recited in claim 16 at least partially implemented as an application for a mobile communication device for execution thereon to identify and classify said input signal captured thereby.