System for detecting speech interval and recognizing continuous speech in a noisy environment through real-time recognition of call commands

• US 8,930,196 B2
• Filed: 08/22/2012
• Issued: 01/06/2015
• Est. Priority Date: 05/28/2008
• Status: Active Grant

• Current Assignee: Kaifi, LLC (Intellectual Discovery Co., Ltd.)
• Original Assignee: Koreapowervoice Company Limited
• Inventors: Jung, Heui-Suck, Chin, Se-Hoon, Roh, Tae-Young
• Primary Examiner(s): JACKSON, JAKIEDA R
• Application Number:
  US13/591,479
• Publication Number:
  US 20120316879A1
• Time in Patent Office: 867 Days
• Field of Search: 704/275
• US Class Current: 704/275
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence

Abstract

A continuous speech recognition system to recognize continuous speech smoothly in a noisy environment. The system selects call commands, configures a minimum recognition network in token, which consists of the call commands and mute intervals including noises, recognizes the inputted speech continuously in real time, analyzes the reliability of speech recognition continuously and recognizes the continuous speech from a speaker. When a speaker delivers a call command, the system for detecting the speech interval and recognizing continuous speech in a noisy environment through the real-time recognition of call commands measures the reliability of the speech after recognizing the call command, and recognizes the speech from the speaker by transferring the speech interval following the call command to a continuous speech-recognition engine at the moment when the system recognizes the call command.

Claims

1. A method for recognizing a speech of a speaker by a device, the method comprising:

receiving, by a preset call command recognition unit, an input speech spoken by the speaker;

receiving, by the preset call command recognition unit, a subsequent input speech spoken, by the speaker, subsequently to the input speech;

comparing, by the preset call command recognition unit, the input speech with a preset call command;

determining, by the preset call command recognition unit, whether the input speech is recognized as the preset call command according to a result of the comparison; and

based on determining that the input speech is recognized as the preset call command, activating a continuous speech recognition unit configured to determine whether the subsequent input speech is recognized as an actual command, in order to reduce computational load on the device; and

wherein the preset call command recognition unit comprises a token passing unit configured to assume the input speech to be the preset call command according to the result of the comparison using a minimum recognition network composed of a silence interval accompanied by noise and the preset call command.

2. The method of claim 1,

wherein determining whether the input speech is recognized as the preset call command further comprises:

computing, by a confidence detection unit, a confidence rate of the input speech; and

recognizing, by the confidence detection unit, the input speech as the preset call command based on the computed confidence rate.

3. The method of claim 2,

wherein recognizing the input speech comprises:

recognizing the input speech as the preset call command if the computed confidence rate is higher than a threshold value.

4. The method of claim 2,

wherein recognizing the input speech comprises:

determining whether or not the input speech is an authentic call command based on the computed confidence rate; and

recognizing the input speech as the preset call command if it is determined that the input speech is the authentic call command.

5. The method of claim 1,

wherein determining the input speech is recognized as the preset call command further comprises:

computing a log likelihood ratio of the input speech; and

assuming the input speech to be the preset call command based on the computed log likelihood ratio.

6. The method of claim 5,

wherein the computed log likelihood ratio is a natural logarithm of a probability on the input speech for a phoneme model of the preset call command, and the computed confidence rate is another log likelihood ratio.

7. The method of claim 6,

wherein said another log likelihood ratio is computed by subtracting, from the log likelihood ratio, a natural logarithm of a probability on the input speech for an anti-phoneme model of the preset call command.

8. The method of claim 1, further comprising:

determining, using a rule check unit, whether or not the recognized subsequent input speech corresponds to a target sentence.

9. An apparatus for recognizing a speech of a speaker, the apparatus comprising:

a processor,

a preset call command recognition unit using the processor; and

a continuous speech recognition unit using the processor,

wherein the preset call command recognition unit is configured to receive an input speech spoken by the speaker;

receive a subsequent input speech spoken, by the speaker, subsequently to the input speech;

compare the input speech with a preset call command; and

recognize the input speech as the preset call command according to a result of the comparison,

wherein the continuous speech recognition unit is configured to recognize the subsequent input speech as an actual command if the input speech is recognized as the preset call command by the preset call command recognition unit, and

wherein the preset call command recognition unit comprises a token passing unit configured to assume the input speech to be the preset call command according to the result of the comparison using a minimum recognition network composed of a silence interval accompanied by noise and the preset call command.

10. The apparatus of claim 9,

wherein the preset call command recognition unit further comprises:

an adaptive filter configured to receive the input speech and the subsequent input speech; and

a confidence detection unit configured to compute a confidence rate of the assumed input speech and recognize the assumed input speech as the preset call command based on the computed confidence rate.

11. The apparatus of claim 10,

wherein the confidence detection unit is further configured to recognize the assumed input speech as the preset call command if the computed confidence rate is higher than a threshold value.

12. The apparatus of claim 10,

wherein the confidence detection unit is further configured to determine whether or not the assumed input speech is an authentic call command based on the computed confidence rate and recognize the assumed input speech as the preset call command if it is determined that the assumed input speech is the authentic call command.

13. The apparatus of claim 10,

wherein the token passing unit is further configured to compute a log likelihood ratio of the input speech and assume the input speech to be the preset call command based on the computed log likelihood ratio.

14. The apparatus of claim 13,

wherein the computed log likelihood ratio is a natural logarithm of a probability on the input speech for a phoneme model of the preset call command, and the computed confidence rate is another log likelihood ratio.

15. The apparatus of claim 14,

wherein said another log likelihood ratio is computed by subtracting, from the log likelihood ratio, a natural logarithm of a probability on the input speech for an anti-phoneme model of the preset call command.

16. The apparatus of claim 9,

wherein the continuous speech recognition unit comprises:

a keyword spotting unit configured to recognize the subsequent input speech as the actual command if the input speech is recognized as the preset call command; and

a rule check unit configured to determine whether the recognized subsequent input speech corresponds to a target sentence.

17. A method for recognizing a speech of a speaker by a single device, the method comprising:

receiving, by a preset call command recognition unit of the single device, an input speech spoken by the speaker;

receiving, by the preset call command recognition unit, a subsequent input speech spoken, by the speaker, subsequently to the input speech;

comparing, by the preset call command recognition unit, the input speech with a preset call command;

determining, by the preset call command recognition unit of the single device, whether the input speech is recognized as the preset call command according to a result of the comparison; and

only if the input speech is recognized as the preset call command, activating a continuous speech recognition unit of the single device that is configured to perform continuous speech recognition to determine whether the subsequent input speech is recognized as an actual command, in order to reduce overall computational load on the single device; and

wherein the preset call command recognition unit comprises a token passing unit configured to assume the input speech to be the preset call command according to the result of the comparison using a minimum recognition network composed of a silence interval accompanied by noise and the preset call command.

18. The method of claim 17, wherein the single device is a portable device or an embedded system.

19. The method of claim 17, wherein the single device,

wherein determining whether the input speech is recognized as the preset call command further comprises:

computing, by a confidence detection unit, a confidence rate of the input speech; and

recognizing, by the confidence detection unit, the input speech as the preset call command based on the computed confidence rate.

20. The method of claim 17,

wherein determining whether the input speech is recognized as the preset call command further comprises:

computing a log likelihood ratio of the input speech; and

assuming the input speech to be the preset call command based on the computed log likelihood ratio.