Voice identification method and apparatus

Patent metadata

Field	Value
Publication number	US-9666186-B2
Application number	US-201314100884-A
Country	US
Kind code	B2
Filing date	Dec 9, 2013
Priority date	Jan 24, 2013
Publication date	May 30, 2017
Grant date	May 30, 2017

Abstract

Embodiments of the present invention provide a voice identification method, which includes: obtaining voice data; obtaining a confidence value according to the voice data; obtaining a noise scenario according to the voice data; obtaining a confidence threshold corresponding to the noise scenario; and if the confidence value is greater than or equal to the confidence threshold, processing the voice data. An apparatus is also provided. The method and apparatus that flexibly adjust the confidence threshold according to the noise scenario greatly improve a voice identification rate under a noise environment.

First claim

What is claimed is:

1. A mobile terminal, comprising:
a microphone, configured to obtain voice data;
a non-transitory medium, configured to store pre-established noise type models; and
one or more processors, configured to:
determine a confidence value in response to the voice data, the confidence value being designated to be used for validating speech recognition processing result of the voice data;
obtain a result value for each of the pre-established noise type models by inputting a frequency cepstrum coefficient of a noise in the voice data to each one of the pre-established noise type models;
select a first pre-established noise type model which has a maximum result value for the voice data;
determine a noise scenario associated with the first pre-established noise type model;
determine a confidence threshold corresponding to the noise scenario; and
perform an operation in accordance with the speech recognition processing result of the voice data in a case that the confidence value being greater than or equal to the confidence threshold;
wherein the noise scenario comprises a noise type and a noise magnitude, the noise magnitude comprises a signal-to-noise ratio and a noise level;
when the noise type is an on-board environment, when the noise level is smaller than −30 dB, and when the signal-to-noise ratio is smaller than 10 dB, the confidence threshold is a value between 35 and 50;
when the noise type is an on-board environment, when the noise level is greater than −30 dB and smaller than −40 dB, and when the signal-to-noise ratio is greater than 10 dB and smaller than 20 dB, the confidence threshold is a value between 40 and 55.

2. The mobile terminal according to claim 1 , wherein the pre-established noise type mode is established by:
obtaining noise data;
obtaining a frequency cepstrum coefficient of the noise data; and
processing the frequency cepstrum coefficient of the noise data according to an Expectation-maximization algorithm.

3. The mobile terminal according to claim 2 , wherein the pre-established noise type mode is a Gaussian mixture model.

4. The mobile terminal according to claim 1 , wherein the noise type corresponds to the first pre-established noise type model, and the one or more processors, further configured to:
obtain a feature parameter of the voice data;
perform voice activity detection based on the feature parameter of the voice data; and
obtain the noise magnitude based on a result of the voice activity detection.

5. The mobile terminal according to claim 1 , the one or more processors, further configured to:
prompt a user in a case that the confidence value is smaller than the confidence threshold.

6. A voice identification method performed by a mobile terminal, comprising:
obtaining voice data;
determining a confidence value in response to the voice data, the confidence value being designated to be used for validating speech recognition processing result of the voice data;
obtaining a result value for each of the pre-established noise type models by inputting a frequency cepstrum coefficient of a noise in the voice data to each one of the pre-established noise type models;
selecting a first pre-established noise type model which has a maximum result value for the voice data;
determining a noise scenario associated with the first pre-established noise type model;
determining a confidence threshold corresponding to the noise scenario; and
performing an operation in accordance with the speech recognition processing result of the voice data in a case that the confidence value being greater than or equal to the confidence threshold;
wherein the noise scenario comprises a noise type and a noise magnitude, the noise magnitude comprises a signal-to-noise ratio and a noise level;
when the noise type is an on-board environment, when the noise level is smaller than −30 dB, and when the signal-to-noise ratio is smaller than 10 dB, the confidence threshold is a value between 35 and 50;
when the noise type is an on-board environment, when the noise level is greater than −30 dB and smaller than −40 dB, and when the signal-to-noise ratio is greater than 10 dB and smaller than 20 dB, the confidence threshold is a value between 40 and 55.

7. The method according to claim 6 , wherein the pre-established noise type mode is established by:
obtaining noise data;
obtaining a frequency cepstrum coefficient of the noise data; and
processing the frequency cepstrum coefficient of the noise data according to an Expectation-maximization algorithm.

8. The method according to claim 7 , wherein the pre-established noise type mode is a Gaussian mixture model.

9. The method according to claim 6 , wherein the noise type corresponds to the first pre-established noise type model, and the method further comprising:
obtaining a feature parameter of the voice data;
performing voice activity detection based on the feature parameter of the voice data; and
obtaining a noise magnitude based on a result of the voice activity detection.

10. The method according to claim 6 , further comprising:
prompting a user in a case that the confidence value is smaller than the confidence threshold.

Assignees

• Huawei Device Co Ltd

Inventors

• Jiang Hongrui
• Wang Xiyong
• Liang Junbin
• Zheng Weijun
• Zhou Junyang

Classifications

• G10L15/08

  Speech classification or search · CPC title

• G10L15/20 Primary

  Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence) · CPC title

• G10L25/84

  for discriminating voice from noise · CPC title

• G10L25/24

  the extracted parameters being the cepstrum · CPC title