Granted Patent B2

US 12,658,182 · App. 18/706,313 · Granted Jun 16, 2026

Inventors: Saisai Zou (Beijing, CN); Lei Jia (Beijing, CN); Haifeng Wang (Beijing, CN)

Assignee: Beijing Baidu Netcom Science Technology Co., Ltd.

G10L15/16G10L13/02G10L15/02G10L2015/027G10L2015/088

Abstract

Claims (68)

1 . A speech wake-up method, comprising:

performing a word recognition on a speech to be recognized from a global aspect to obtain a wake-up word recognition result;

performing a syllable recognition on the speech to be recognized from a local aspect to obtain a wake-up syllable recognition result, in response to determining that the wake-up word recognition result represents that the speech to be recognized contains a predetermined wake-up word; and

determining that the speech to be recognized is a correct wake-up speech, in response to determining that the wake-up syllable recognition result represents that the speech to be recognized contains a predetermined syllable,

wherein the performing the word recognition on the speech to be recognized from the global aspect to obtain the wake-up word recognition result comprises:

performing a convolution on the speech to be recognized to obtain a first-level feature vector sequence, wherein the speech to be recognized comprises a speech frame sequence, and the first-level feature vector sequence corresponds to the speech frame sequence;

performing a gate recurrent operation on the first-level feature vector sequence to obtain a second-level feature vector sequence, and

performing a classification on the second-level feature vector sequence to obtain the wake-up word recognition result,

wherein the performing the gate recurrent operation on the first-level feature vector sequence to obtain the second-level feature vector sequence comprises repeatedly performing an operation comprising:

determining an update gate of a current moment and a candidate hidden layer information of the current moment based on an output vector of a previous moment and an input vector of the current moment, wherein the input vector of the current moment is a first-level feature vector at the current moment in the first-level feature vector sequence;

determining a hidden layer information of the current moment based on the candidate hidden layer information of the current moment, a hidden layer information of the previous moment, and the update gate of the current moment; and

determining an output vector of the current moment based on the hidden layer information of the current moment and a predetermined parameter, wherein the output vector of the current moment is a second-level feature vector at the current moment in the second-level feature vector sequence,

wherein the predetermined parameter is a projection parameter, the predetermined parameter is determined based on a threshold of the number of lightweight parameters, the threshold of the number of lightweight parameters refer to a parameter setting benchmark, so that a size of the predetermined parameter is less than or equal to the threshold of the number of lightweight parameters.

2 . The method according to claim 1 , wherein the performing a syllable recognition on the speech to be recognized from a local aspect to obtain a wake-up syllable recognition result in response to determining that the wake-up word recognition result represents that the speech to be recognized contains a predetermined wake-up word comprises:

performing a syllable feature extraction on the speech to be recognized to obtain a syllable feature matrix; and

performing a classification on the syllable feature matrix to obtain the wake-up syllable recognition result.

3 . The method according to claim 2 , wherein the performing a syllable feature extraction on the speech to be recognized to obtain a syllable feature matrix comprises:

performing a feature extraction on the speech to be recognized to obtain a feature matrix;

performing a dimension reduction on the feature matrix to obtain a dimension-reduced feature matrix; and

performing a multi-stage speech-enhanced encoding on the dimension-reduced feature matrix to obtain the syllable feature matrix.

4 . The method according to claim 1 , further comprising:

determining that the speech to be recognized is an incorrect wake-up speech, in response to determining that the wake-up word recognition result represents that the speech to be recognized does not contain the predetermined wake-up word.

5 . The method according to claim 1 , further comprising:

displaying a target object on a display interface in response to determining that the speech to be recognized is a correct wake-up speech; and

outputting a feedback speech.

6 . An electronic device, comprising:

at least one processor; and

a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to:

perform a word recognition on a speech to be recognized from a global aspect to obtain a wake-up word recognition result;

perform a syllable recognition on the speech to be recognized from a local aspect to obtain a wake-up syllable recognition result, in response to determining that the wake-up word recognition result represents that the speech to be recognized contains a predetermined wake-up word; and

determine that the speech to be recognized is a correct wake-up speech, in response to determining that the wake-up syllable recognition result represents that the speech to be recognized contains a predetermined syllable,

wherein the at least one processor is further configured to:

perform a convolution on the speech to be recognized to obtain a first-level feature vector sequence, wherein the speech to be recognized comprises a speech frame sequence, and the first-level feature vector sequence corresponds to the speech frame sequence;

perform a gate recurrent operation on the first-level feature vector sequence to obtain a second-level feature vector sequence; and

perform a classification on the second-level feature vector sequence to obtain the wake-up word recognition result,

wherein the at least one processor is further configured to repeatedly perform an operation comprising:

determining an update gate of a current moment and a candidate hidden layer information of the current moment based on an output vector of a previous moment and an input vector of the current moment, wherein the input vector of the current moment is a first-level feature vector at the current moment in the first-level feature vector sequence;

determining a hidden layer information of the current moment based on the candidate hidden layer information of the current moment, a hidden layer information of the previous moment, and the update gate of the current moment; and

determining an output vector of the current moment based on the hidden layer information of the current moment and a predetermined parameter, wherein the output vector of the current moment is a second-level feature vector at the current moment in the second-level feature vector sequence,

wherein the predetermined parameter is a projection parameter, the predetermined parameter is determined based on a threshold of the number of lightweight parameters, the threshold of the number of lightweight parameters refer to a parameter setting benchmark, so that a size of the predetermined parameter is less than or equal to the threshold of the number of lightweight parameters.

7 . A non-transitory computer-readable storage medium having computer instructions therein, wherein the computer instructions are configured to cause a computer to:

perform a word recognition on a speech to be recognized from a global aspect to obtain a wake-up word recognition result;

perform a syllable recognition on the speech to be recognized from a local aspect to obtain a wake-up syllable recognition result, in response to determining that the wake-up word recognition result represents that the speech to be recognized contains a predetermined wake-up word; and

determine that the speech to be recognized is a correct wake-up speech, in response to determining that the wake-up syllable recognition result represents that the speech to be recognized contains a predetermined syllable,

wherein the computer instructions are further configured to cause the computer to:

perform a convolution on the speech to be recognized to obtain a first-level feature vector sequence, wherein the speech to be recognized comprises a speech frame sequence, and the first-level feature vector sequence corresponds to the speech frame sequence;

perform a gate recurrent operation on the first-level feature vector sequence to obtain a second-level feature vector sequence; and

perform a classification on the second-level feature vector sequence to obtain the wake-up word recognition result,

wherein the computer instructions are further configured to cause the computer to repeatedly perform an operation comprising:

determining an update gate of a current moment and a candidate hidden layer information of the current moment based on an output vector of a previous moment and an input vector of the current moment, wherein the input vector of the current moment is a first-level feature vector at the current moment in the first-level feature vector sequence;

determining a hidden layer information of the current moment based on the candidate hidden layer information of the current moment, a hidden layer information of the previous moment, and the update gate of the current moment; and

determining an output vector of the current moment based on the hidden layer information of the current moment and a predetermined parameter, wherein the output vector of the current moment is a second-level feature vector at the current moment in the second-level feature vector sequence,

wherein the predetermined parameter is a projection parameter, the predetermined parameter is determined based on a threshold of the number of lightweight parameters, the threshold of the number of lightweight parameters refer to a parameter setting benchmark, so that a size of the predetermined parameter is less than or equal to the threshold of the number of lightweight parameters.

8 . The electronic device according to claim 6 , wherein the at least one processor is further configured to:

perform a syllable feature extraction on the speech to be recognized to obtain a syllable feature matrix; and

perform a classification on the syllable feature matrix to obtain the wake-up syllable recognition result.

9 . The electronic device according to claim 8 , wherein the at least one processor is further configured to:

perform a feature extraction on the speech to be recognized to obtain a feature matrix;

perform a dimension reduction on the feature matrix to obtain a dimension-reduced feature matrix; and

perform a multi-stage speech-enhanced encoding on the dimension-reduced feature matrix to obtain the syllable feature matrix.

10 . The electronic device according to claim 6 , wherein the at least one processor is further configured to:

determine that the speech to be recognized is an incorrect wake-up speech, in response to determining that the wake-up word recognition result represents that the speech to be recognized does not contain the predetermined wake-up word.

11 . The electronic device according to claim 6 , wherein the at least one processor is further configured to:

display a target object on a display interface in response to determining that the speech to be recognized is a correct wake-up speech; and

output a feedback speech.

12 . The non-transitory computer-readable storage medium according to claim 7 , wherein the computer instructions are further configured to cause the computer to:

perform a syllable feature extraction on the speech to be recognized to obtain a syllable feature matrix; and

perform a classification on the syllable feature matrix to obtain the wake-up syllable recognition result.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (39)