SYSTEM AND METHOD OF PATTERN RECOGNITION IN VERY HIGH DIMENSIONAL SPACE

• US 20080183471A1
• Filed: 03/28/2008
• Published: 07/31/2008
• Est. Priority Date: 11/02/2000
• Status: Active Grant

• Current Assignee: AT&T Corporation (AT&T, Inc.)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Atal, Bishnu Saroop
• Granted Patent: US 7,869,997 B2
• Time in Patent Office: Days
• US Class Current: 704/254
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

A system and method of recognizing speech comprises an audio receiving element and a computer server. The audio receiving element and the computer server perform the process steps of the method. The method involves training a stored set of phonemes by converting them into n-dimensional space, where n is a relatively large number. Once the stored phonemes are converted, they are transformed using single value decomposition to conform the data generally into a hypersphere. The received phonemes from the audio-receiving element are also converted into n-dimensional space and transformed using single value decomposition to conform the data into a hypersphere. The method compares the transformed received phoneme to each transformed stored phoneme by comparing a first distance from a center of the hypersphere to a point associated with the transformed received phoneme and a second distance from the center of the hypersphere to a point associated with the respective transformed stored phoneme.

Claims

1. A computer-readable medium storing instructions for controlling a computing device to process phonemes for use in recognizing a received phoneme having an associated received-signal vector using a stored plurality of phoneme classes, each of the plurality of phoneme classes comprising class phonemes, the instructions comprising, for each class phoneme:

generating an expanded stored-phoneme vector from the class phoneme; and

transforming the expanded stored-phoneme vector into an orthogonal form associated with a hypersphere having a center and a radius, wherein a received phoneme may be recognized by generating an expanded received-signal vector into an orthogonal form for analysis in the hypersphere.

2. The computer-readable medium of claim 1, wherein the step of generating the expanded stored-phoneme vector from the class phoneme further comprises:

determining a phoneme vector as a time-frequency representation of the class phoneme;

dividing the phoneme vector into phoneme segments;

assigning each phoneme segment into a plurality of phoneme parameters; and

expanding each phoneme segment and plurality of phoneme parameters into an expanded stored-phoneme vector with expanded vector parameters.

3. The computer-readable medium of claim 1, wherein the step of transforming the expanded stored-phoneme vector into an orthogonal form further comprises:

setting [x₁ x₂ . . . x_m]=[u₁ u₂ . . . u_m] Λ

V^t, where x_k is a k^th acoustic vector for a corresponding stored phoneme, u_k is the corresponding orthogonal vector and Λ and

V are diagonal and unitary matrices, respectively.

4. The computer-readable medium of claim 1, wherein the instructions further comprise transforming the expanded received signal vector which is associated with the received phoneme into an orthogonal form using a singular-value decomposition to conform the expanded received-single signal vector into the hypersphere.

5. A method of recognizing a received phoneme, the method comprising:

generating an expanded received-signal vector from a received-signal vector associated with a received phoneme;

transforming the expanded received-signal vector into an orthogonal form associated with a hypersphere having a center and a radius;

determining a first distance associated with the orthogonal form of the expanded received-signal vector and a second distance associated respectfully with each orthogonal form of expanded stored phoneme vectors; and

recognizing the received phoneme according to a comparison of the first distance with the second distance.

6. The method of claim 5, wherein generating the expanded received-signal vector further comprises:

processing the received-signal vector as a time-frequency representation of a received signal associated with the received phoneme;

dividing the received-signal vector into received-signal segments;

assigning each received-signal segment a plurality of received-signal parameters; and

expanding each received-signal segment and plurality of received-signal parameters into the expanded received-signal vector.

7. The method of claim 5, wherein transforming the expanded received-signal vector into an orthogonal form further comprises:

setting [y_k]=[z_k] Λ

V^t, where y_k is a k^th acoustic vector for a corresponding received phoneme, z_k is the corresponding orthogonal vector and Λ and

V are diagonal and unitary matrices, respectively.

8. The method of claim 5, wherein transforming the expanded stored phoneme vector into an orthogonal form uses singular-value decomposition and wherein transforming the expanded received-signal vector into an orthogonal form using singular-value decomposition further conforms the stored-phoneme vector into the hypersphere.

9. The method of claim 8, wherein determining a distance associated with the orthogonal form of the expanded received-signal vector and each orthogonal form of the expanded stored-phoneme vectors further comprises:

comparing a distance from the center of the hypersphere of the orthogonal form of the expanded received-signal vector with a distance from the center of the hypersphere for each orthogonal form of the expanded stored-phoneme vector.

10. The method of claim 9, wherein determining a distance associated with the orthogonal form of the expanded received-signal vector and each orthogonal form of the expanded stored-phoneme vectors further comprises:

determining a difference between the distance from the center of the hypersphere of the orthogonal form of the expanded received-signal vector and the distance from the center of the hypersphere for each orthogonal form of the expanded stored-phoneme vectors, wherein the expanded stored-phoneme vectors associated with m-shortest differences between the distance from the center of the hypersphere of the orthogonal form of the expanded received-signal vector and the distance from the center of the hypersphere for each orthogonal form of the expanded stored-phoneme vectors are recognized as most likely to be associated with the received phoneme.

11. A computing device for recognizing a received phoneme, the computing device comprising:

a module configured to generate an expanded stored-phoneme vector from each of a plurality of class phonemes;

a module configured to transform the expanded stored-phoneme vector into an orthogonal form associated with a hypersphere having a center and a radius; and

a module configured to recognize a received phoneme by generating an expanded received-signal vector associated with the received phoneme into an orthogonal form for analysis in the hypersphere.

12. The computing device of claim 11, wherein the module configured to generate the expanded stored phoneme vector from the class phoneme further:

determines the phoneme vector as a time-frequency representation of the class phoneme;

divides the phoneme vector into phoneme segments;

assigns each phoneme segment into a plurality of phoneme parameters; and

expands each phoneme segment and plurality of phoneme parameters into an expanded stored phoneme vector with expanded vector parameters.

13. The computing device of claim 11, wherein the module configured to transform the expanded stored-phoneme vector into an orthogonal form further:

sets [x₁ x₂ . . . x_m]=[u₁ u₂ . . . u_m] Λ

V^t, where x_k is a k^th acoustic vector for a corresponding stored phoneme, u_k is the corresponding orthogonal vector and Λ and

V are diagonal and unitary matrices, respectively.

14. The computing device of claim 11, further comprising a module configured to transform the expanded received signal vector which is associated with the received phoneme into an orthogonal form using a singular-value decomposition to conform the expanded received-single signal vector into the hypersphere.

15. A computing device for recognizing a received phoneme, the computing device comprising:

a module configured to generate an expanded received-signal vector from a received signal vector associated with a received phoneme;

a module configured to transform the expanded received-signal vector into an orthogonal form associated with a hypersphere having a center and a radius;

a module configured to determine a first distance associated with the orthogonal form of the expanded received-signal vector and a second distance associated with each orthogonal form of expanded stored phoneme vectors; and

a module configured to recognize the received phoneme according to a comparison of the first distance with the second distance.

16. The computing device of claim 15, wherein the module configured to generate the expanded received-signal vector further:

processes the received-signal vector as a time-frequency representation of a received signal associated with the received phoneme;

divides the received-signal vector into received-signal segments;

assigns each received-signal segment into a plurality of received-signal parameters; and

expands each received-signal segment and plurality of received-signal parameters into the expanded received-signal vector.

17. The computing device of claim 15, wherein the module configured to transform the expanded received-signal vector into an orthogonal form further:

sets [y_k]=[z_k] Λ

V^t, where y_k is a k^th acoustic vector for a corresponding received phoneme, z_k is the corresponding orthogonal vector and Λ and

V are diagonal and unitary matrices, respectively.

18. The computing device of claim 15, wherein the module configured to transform the expanded stored phoneme vector into an orthogonal form uses singular-value decomposition and wherein transforming the expanded received-signal vector into an orthogonal form using singular-value decomposition further conforms the stored-phoneme vector into the hypersphere.

19. The computing device of claim 15, wherein the module configured to determine a distance associated with the orthogonal form of the expanded received-signal vector and each orthogonal form of the expanded stored-phoneme vectors further:

compares a distance from the center of the hypersphere of the orthogonal form of the expanded received-signal vector with a distance from the center of the hypersphere for each orthogonal form of the expanded stored-phoneme vector.

20. The computing device of claim 15, wherein the module configured to determine a distance associated with the orthogonal form of the expanded received-signal vector and each orthogonal form of the expanded stored-phoneme vectors further:

determines a difference between the distance from the center of the hypersphere of the orthogonal form of the expanded received-signal vector and the distance from the center of the hypersphere for each orthogonal form of the expanded stored-phoneme vectors, wherein the expanded stored-phoneme vectors associated with m-shortest differences between the distance from the center of the hypersphere of the orthogonal form of the expanded received-signal vector and the distance from the center of the hypersphere for each orthogonal form of the expanded stored-phoneme vectors are recognized as most likely to be associated with the received phoneme.