Hsin Huei Tseng

US Patent:

7769583, Aug 3, 2010

Filed:

May 13, 2006

Appl. No.:

11/383199

Inventors:

Upendra V. Chaudhari - Briarcliff Manor NY, US
Hsin I. Tseng - Los Angeles CA, US
Deepak S. Turaga - Elmsford NY, US
Olivier Verscheure - Frameries, BE

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G10L 19/00
G10L 19/02
G10L 17/00
G10L 21/04
G10L 19/12
H04M 1/64

US Classification:

704230, 704229, 704246, 704247, 704248, 704249, 704250, 704500, 704503, 704200, 704222, 379 8801, 379 8802

Abstract:

A system, method and computer program product for classification of an analog electrical signal using statistical models of training data. A technique is described to quantize the analog electrical signal in a manner which maximizes the compression of the signal while simultaneously minimizing the diminution in the ability to classify the compressed signal. These goals are achieved by utilizing a quantizer designed to minimize the loss in a power of the log-likelihood ratio. A further technique is described to enhance the quantization process by optimally allocating a number of bits for each dimension of the quantized feature vector subject to a maximum number of bits available across all dimensions.

US Patent:

8271278, Sep 18, 2012

Filed:

Apr 3, 2010

Appl. No.:

12/753859

Inventors:

Upendra V. Chaudhari - Briarcliff Manor NY, US
Hsin I. Tseng - Los Angeles CA, US
Deepak S. Turaga - Elmsford NY, US
Olivier Verscheure - Frameries, BE

Assignee:

International Business Machines Corporation - Armonk NY

International Classification:

G10L 19/00
G10L 19/02
G10L 17/00
G10L 21/04
G10L 11/00
G10L 19/12
H04M 1/64

US Classification:

704230, 704229, 704246, 704247, 704248, 704249, 704250, 704500, 704503, 704200, 704222, 379 8801, 379 8802

Abstract:

A system, method and computer program product for classification of an analog electrical signal using statistical models of training data. A technique is described to quantize the analog electrical signal in a manner which maximizes the compression of the signal while simultaneously minimizing the diminution in the ability to classify the compressed signal. These goals are achieved by utilizing a quantizer designed to minimize the loss in a power of the log-likelihood ratio. A further technique is described to enhance the quantization process by optimally allocating a number of bits for each dimension of the quantized feature vector subject to a maximum number of bits available across all dimensions.