Method of extracting features in a voice recognition system

• US 6,182,036 B1
• Filed: 02/23/1999
• Issued: 01/30/2001
• Est. Priority Date: 02/23/1999
• Status: Expired due to Term

• Current Assignee: Google Technology Holdings LLC (Alphabet Inc.)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Poppert, Daniel Charles
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Storm, Donald L.
• Application Number:
  US09/256,280
• Time in Patent Office: 707 Days
• Field of Search: 704/243, 704/206, 704/201
• US Class Current: 704/243
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit

Abstract

A method of extracting features for a voice recognition system of a wireless communication device. The wireless communication device includes a transmitter and a receiver coupled to a call processor, a microphone for inputting audible sounds, and an analog-to-digital converter coupled to the microphone to output digital signals. The call processor includes a coefficient generator coupled to the analog-to-digital converter and generating representative coefficients, a differentiator coupled to the coefficient generator to generate delta coefficients, and an extractor outputting a portion of the representative coefficients and the delta coefficients as a feature vector for use in voice recognition, wherein every other representative coefficient is used in the feature vector.

Claims

1. A method of extracting features for a voice recognition system comprising the steps of:

inputting a digitized signal;

generating representative coefficients from the digitized signal;

generating differential coefficients from the representative coefficients; and

selectively using a portion of the representative coefficients and the differential coefficients to construct a feature vector, wherein every other representative coefficient is used in the feature vector.

2. The method as defined in claim 1, wherein the step of generating representative coefficients includese step of generating cepstral coefficients.

3. The method as defined in claim 2, wherein the step of generating differential coefficients includes the step of generating delta cepstral coefficients.

4. The method as defined in claim 3, wherein the cepstral coefficients are output first and the delta-cepstral coefficients are output second.

5. The method as defined in claim 2, wherein the cepstral coefficients are mel frequency cepstral coefficients.

6. The method as defined in claim 1, wherein every other differential coefficient is used in the feature vector.

7. A voice recognition system, comprising:

a microphone for inputting audible sounds;

an analog-to-digital converter coupled to the microphone to output digital signals;

a coefficient generator coupled to the analog-to-digital converter and generating representative coefficients;

a differentiator coupled to the coefficient generator to generate delta coefficients;

and an extractor selectively, outputting a portion of the representative coefficients and the delta coefficients as a feature vector for use in voice recognition, wherein every other representative coefficient is used in the feature vector.

8. A wireless communication device, comprising:

a call processor performing call processing;

a transmitter and a receiver, coupled to the call processor;

a microphone for inputting audible sounds;

an analog-to-digital converter coupled to the microphone to output digital signials;

a coefficient generator, positioned in the call processor and coupled to the analog-to-digital converter, generating representative coefficients;

an extractor, positioned in the call processor, selectively outputting a portion of the representative coefficients and the delta coefficients as a feature vector for use in voice recognition, wherein every other representative coefficient is used in the feature vector.