Control method based on voice recognition

Abstract

The invention discloses a control method based on voice recognition, and belongs to the technical field of voice recognition, and the method comprises the following steps: S1, collecting the control voice of a user, and carrying out the framing processing of the control voice; s2, generating a global audio matrix, a first combination matrix and a second combination matrix according to a plurality of frames of control voice segments obtained through framing processing, and constructing a control voice objective function according to the global audio matrix, the first combination matrix and the second combination matrix; s3, processing the control voice by using the control voice target function to obtain an optimal control voice; and S4, generating a control password according to the optimal control voice. The control voice objective function is generated, quality processing of the control voice is completed, the accuracy of converting the control voice into the control password is improved, home control is more visual and easy to use, and popularization of smart home is facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of speech recognition, and in particular relates to a control method based on speech recognition. Background Art

[0002] With the continuous advancement of science and technology, the development of intelligence has also been continuously improving. Among them, the smart home system is closely related to people's lives, providing people with a more comfortable, convenient and safer living environment, and therefore has received widespread attention. One of the advantages of smart homes is that they can be controlled by voice. However, in some application scenarios, the use of smart homes will generate noise. For example, gas stoves and range hoods will generate noise that is sufficient to affect the user's voice after being turned on and used; due to the noise, users cannot accurately control the smart home by speaking, affecting the control effect of the smart home. In addition, the method of controlling smart homes through voice recognition in the prior art has the problem of poor control effect. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a control method based on speech recognition.

[0004] The technical solution of the present invention is: a control method based on speech recognition comprises the following steps:

[0005] S1. Collect the user's control voice and perform frame processing on the control voice;

[0006] S2, generating a global audio matrix, a first combination matrix and a second combination matrix according to the plurality of frame control speech segments obtained by the frame processing, and constructing a control speech target function according to the global audio matrix, the first combination matrix and the second combination matrix;

[0007] S3, processing the control speech using the control speech objective function to obtain the optimal control speech;

[0008] S4. Generate a control command based on the optimal control voice.

[0009] In S4, deep learning and other methods can be used to convert the optimal control voice into text, i.e., control commands.

[0010] Furthermore, S2 includes the following sub-steps:

[0011] S21, constructing a frequency spectrum matrix of each frame of the control speech segment according to the linear prediction cepstral coefficients of each frame of the control speech segment;

[0012] S22, obtaining a global audio matrix according to the spectrum matrix of all frame control speech segments;

[0013] S23, traversing the user's control voice to generate a first combination matrix and a second combination matrix;

[0014] S24. Construct a control speech target function according to the global audio matrix, the first combination matrix and the second combination matrix.

[0015] The beneficial effect of the above further scheme is: in the present invention, the linear prediction cepstral coefficient is used as the characteristic parameter of the audio signal, and the global audio matrix can be generated by the linear prediction cepstral coefficient of each frame of the control speech segment. Since the global audio matrix takes into account the characteristic parameters of all frames, it can effectively represent the spectral characteristics of the control speech and improve the accuracy and robustness of speech recognition. In addition, the control speech is traversed, and the corresponding first combination matrix and second combination matrix are regenerated through parameters such as frame shift and short-time zero-crossing rate. These two matrices can analyze the energy and length of the signal and participate in the construction of the objective function. The control speech is processed by combining the control speech objective function with a high-pass filter to effectively remove invalid noise and make the speech clearer.

[0016] Further, in S21, the spectrum matrix p of the i-th frame control speech segment i The expression is:

[0017] ; In the formula, d i represents the linear prediction cepstral coefficient of the control speech segment of the i-th frame, e represents the exponent, and N represents the signal-to-noise ratio of the control speech.

[0018] Furthermore, in S22, the expression of the global audio matrix P is:

[0019] ; In the formula, Q ave represents the mean of the largest singular values ​​of all spectral matrices, λ ave Represents the mean of the largest eigenvalues ​​of all spectral matrices.

[0020] Further, S23 includes the following sub-steps:

[0021] S231, calculating a characteristic frame coefficient of each frame of the control speech segment according to the frame shift of two adjacent frames of the control speech segment and the frame length of each frame of the control speech segment;

[0022] S232, arranging the characteristic frame coefficients of all the frame speech segments into a frame coefficient sequence in time order;

[0023] S233, calculating all differential values ​​of the frame coefficient sequence, taking all positive differential values ​​as the first coefficient sequence, and taking all negative differential values ​​as the second coefficient sequence;

[0024] S234, generating a first combination matrix according to the first coefficient sequence and the second coefficient sequence;

[0025] S235, randomly dividing the user's control voice into a first control voice segment and a second control voice segment;

[0026] S236: Generate a second combination matrix according to the first control voice segment and the second control voice segment.

[0027] The beneficial effect of the above further scheme is: in the present invention, by analyzing the frame shift of adjacent frames, the frame coefficient series between all frames in the control speech is determined, and two coefficient sequences are generated by the positive and negative conditions of all differential values ​​of the frame coefficient series to obtain the first combination matrix. The short-time zero-crossing rate refers to the number of times the signal waveform passes through the zero value (or a certain threshold value) in the speech signal. It reflects the frequency characteristics of the signal in the time period. The part with a higher short-time zero-crossing rate usually corresponds to the unvoiced segment or high-frequency noise in the speech signal, thereby generating the second combination matrix.

[0028] Further, in S231, the characteristic frame coefficient M of the i-th frame control speech segment i The calculation formula is:

[0029] Where, L i Indicates the frame length of the i-th frame control speech segment, e represents the index, l i-1,i represents the frame shift between the i-1th frame control speech segment and the i-th frame control speech segment, l i,i+1 Indicates the frame shift between the control speech segment of the i-th frame and the control speech segment of the i+1-th frame.

[0030] Further, in S234, the expression of the first combination matrix A is:

[0031] ; In the formula, a1 represents the sum of all elements in the first coefficient sequence, and a2 represents the sum of all elements in the second coefficient sequence.

[0032] Further, in S236, the expression of the second combination matrix B is:

[0033] ; Wherein, b1 represents the short-time zero-crossing rate of the first control speech segment, and b2 represents the short-time zero-crossing rate of the second control speech segment.

[0034] Further, in S24, the expression of the control speech target function Z is:

[0035] ; Where A represents the first combination matrix, B represents the second combination matrix, P represents the global audio matrix, Tr(·) represents the trace operation of the matrix, ||·|| F Represents the F-norm operation.

[0036] Furthermore, in S3, the method for obtaining the optimal control speech is: using the control speech target function value as the quality factor of the high-pass filter, and using the high-pass filter to process the control speech to obtain the optimal control speech.

[0037] The beneficial effects of the present invention are as follows: the present invention discloses a control method based on speech recognition, performs feature analysis on the control speech emitted by the user, extracts feature parameters of each frame of the control speech segment to generate a global audio matrix, a first combination matrix and a second combination matrix, generates a control speech target function, completes quality processing of the control speech, improves the accuracy of converting the control speech into control commands, makes home control more intuitive and easy to use, and contributes to the popularization of smart homes. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Flow chart of a control method based on speech recognition. DETAILED DESCRIPTION

[0039] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0040] like Figure 1 As shown, the present invention provides a control method based on speech recognition, comprising the following steps:

[0041] S1. Collect the user's control voice and perform frame processing on the control voice;

[0042] S2, generating a global audio matrix, a first combination matrix and a second combination matrix according to the plurality of frame control speech segments obtained by the frame processing, and constructing a control speech target function according to the global audio matrix, the first combination matrix and the second combination matrix;

[0043] S3, processing the control speech using the control speech objective function to obtain the optimal control speech;

[0044] S4. Generate a control command based on the optimal control voice.

[0045] In S4, deep learning and other methods can be used to convert the optimal control voice into text, i.e., control commands.

[0046] In this embodiment of the present invention, S2 includes the following sub-steps:

[0047] S21, constructing a frequency spectrum matrix of each frame of the control speech segment according to the linear prediction cepstral coefficients of each frame of the control speech segment;

[0048] S22, obtaining a global audio matrix according to the spectrum matrix of all frame control speech segments;

[0049] S23, traversing the user's control voice to generate a first combination matrix and a second combination matrix;

[0050] S24. Construct a control speech target function according to the global audio matrix, the first combination matrix and the second combination matrix.

[0051] In the present invention, the linear prediction cepstral coefficient is used as the characteristic parameter of the audio signal. The global audio matrix can be generated by the linear prediction cepstral coefficient of each frame of the control speech segment. Since the global audio matrix takes into account the characteristic parameters of all frames, it can effectively represent the spectral characteristics of the control speech and improve the accuracy and robustness of speech recognition. In addition, the control speech is traversed, and the corresponding first combination matrix and second combination matrix are regenerated through parameters such as frame shift and short-time zero-crossing rate. These two matrices can analyze the energy and length of the signal and participate in the construction of the objective function. The control speech is processed by combining the control speech objective function with a high-pass filter to effectively remove invalid noise and make the speech clearer.

[0052] In the embodiment of the present invention, in S21, the spectrum matrix p of the i-th frame control speech segment is i The expression is:

[0053] ; In the formula, d i represents the linear prediction cepstral coefficient of the control speech segment of the i-th frame, e represents the exponent, and N represents the signal-to-noise ratio of the control speech.

[0054] In the embodiment of the present invention, in S22, the expression of the global audio matrix P is:

[0055] ; In the formula, Q ave represents the mean of the largest singular values ​​of all spectral matrices, λ ave Represents the mean of the largest eigenvalues ​​of all spectral matrices.

[0056] In the embodiment of the present invention, S23 includes the following sub-steps:

[0057] S231, calculating a characteristic frame coefficient of each frame of the control speech segment according to the frame shift of two adjacent frames of the control speech segment and the frame length of each frame of the control speech segment;

[0058] S232, arranging the characteristic frame coefficients of all the frame speech segments into a frame coefficient sequence in time order;

[0059] S233, calculating all differential values ​​of the frame coefficient sequence, taking all positive differential values ​​as the first coefficient sequence, and taking all negative differential values ​​as the second coefficient sequence;

[0060] S234, generating a first combination matrix according to the first coefficient sequence and the second coefficient sequence;

[0061] S235, randomly dividing the user's control voice into a first control voice segment and a second control voice segment;

[0062] S236: Generate a second combination matrix according to the first control voice segment and the second control voice segment.

[0063] In the present invention, the frame coefficient series between all frames in the control speech are determined by analyzing the frame shift of adjacent frames, and two coefficient sequences are generated by the positive and negative conditions of all differential values ​​of the frame coefficient series to obtain the first combination matrix. The short-time zero-crossing rate refers to the number of times the signal waveform passes through the zero value (or a certain threshold value) in the speech signal. It reflects the frequency characteristics of the signal in the time period. The part with a higher short-time zero-crossing rate usually corresponds to the unvoiced segment or high-frequency noise in the speech signal, thereby generating the second combination matrix.

[0064] In the embodiment of the present invention, in S231, the characteristic frame coefficient M of the i-th frame control speech segment is i The calculation formula is:

[0065] Where, L i Indicates the frame length of the i-th frame control speech segment, e represents the index, l i-1,i represents the frame shift between the i-1th frame control speech segment and the i-th frame control speech segment, l i,i+1 Indicates the frame shift between the control speech segment of the i-th frame and the control speech segment of the i+1-th frame.

[0066] In the embodiment of the present invention, in S234, the expression of the first combination matrix A is:

[0067] ; In the formula, a1 represents the sum of all elements in the first coefficient sequence, and a2 represents the sum of all elements in the second coefficient sequence.

[0068] In the embodiment of the present invention, in S236, the expression of the second combination matrix B is:

[0069] ; Wherein, b1 represents the short-time zero-crossing rate of the first control speech segment, and b2 represents the short-time zero-crossing rate of the second control speech segment.

[0070] In the embodiment of the present invention, in S24, the expression of the control speech target function Z is:

[0071] ; Where A represents the first combination matrix, B represents the second combination matrix, P represents the global audio matrix, Tr(·) represents the trace operation of the matrix, ||·|| F Represents the F-norm operation.

[0072] In the embodiment of the present invention, in S3, the method for obtaining the optimal control speech is: using the control speech target function value as the quality factor of the high-pass filter, and using the high-pass filter to process the control speech to obtain the optimal control speech.

[0073] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A control method based on speech recognition, characterized in that: The following steps are involved: S1. Collect the user's control voice and perform frame processing on the control voice; S2, generating a global audio matrix, a first combination matrix and a second combination matrix according to the plurality of frame control speech segments obtained by the frame processing, and constructing a control speech target function according to the global audio matrix, the first combination matrix and the second combination matrix; S3, processing the control speech using the control speech objective function to obtain the optimal control speech; S4. Generate a control command based on the optimal control voice.

2. The control method based on speech recognition according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21, constructing a frequency spectrum matrix of each frame of the control speech segment according to the linear prediction cepstral coefficients of each frame of the control speech segment; S22, obtaining a global audio matrix according to the spectrum matrix of all frame control speech segments; S23, traversing the user's control voice to generate a first combination matrix and a second combination matrix; S24. Construct a control speech target function according to the global audio matrix, the first combination matrix and the second combination matrix.

3. The control method based on speech recognition according to claim 2, characterized in that: In S21, the spectrum matrix p of the i-th frame control speech segment i The expression is: ; In the formula, d i represents the linear prediction cepstral coefficient of the control speech segment of the i-th frame, e represents the exponent, and N represents the signal-to-noise ratio of the control speech.

4. The control method based on speech recognition according to claim 2, characterized in that: In S22, the expression of the global audio matrix P is: Where Q ave represents the mean of the largest singular values ​​of all spectral matrices, λ ave Represents the mean of the largest eigenvalues ​​of all spectral matrices.

5. The control method based on speech recognition according to claim 2, characterized in that: The S23 comprises the following sub-steps: S231, calculating a characteristic frame coefficient of each frame of the control speech segment according to the frame shift of two adjacent frames of the control speech segment and the frame length of each frame of the control speech segment; S232, arranging the characteristic frame coefficients of all the frame speech segments into a frame coefficient sequence in time order; S233, calculating all differential values ​​of the frame coefficient sequence, taking all positive differential values ​​as the first coefficient sequence, and taking all negative differential values ​​as the second coefficient sequence; S234, generating a first combination matrix according to the first coefficient sequence and the second coefficient sequence; S235, randomly dividing the user's control voice into a first control voice segment and a second control voice segment; S236: Generate a second combination matrix according to the first control voice segment and the second control voice segment.

6. The control method based on speech recognition according to claim 5, characterized in that: In S231, the characteristic frame coefficient M of the i-th frame control speech segment i The calculation formula is: Where, L i Indicates the frame length of the i-th frame control speech segment, e represents the index, l i-1,i represents the frame shift between the i-1th frame control speech segment and the i-th frame control speech segment, l i,i+1 Indicates the frame shift between the control speech segment of the i-th frame and the control speech segment of the i+1-th frame.

7. The control method based on speech recognition according to claim 5, characterized in that: In S234, the expression of the first combination matrix A is: ; In the formula, a1 represents the sum of all elements in the first coefficient sequence, and a2 represents the sum of all elements in the second coefficient sequence.

8. The control method based on speech recognition according to claim 5, characterized in that: In S236, the expression of the second combination matrix B is: ; Wherein, b1 represents the short-time zero-crossing rate of the first control speech segment, and b2 represents the short-time zero-crossing rate of the second control speech segment.

9. The control method based on speech recognition according to claim 2, characterized in that: In S24, the expression of the control speech target function Z is: ; Where A represents the first combination matrix, B represents the second combination matrix, P represents the global audio matrix, Tr(·) represents the trace operation of the matrix, ||·|| F Represents the F-norm operation.

10. The control method based on speech recognition according to claim 1, characterized in that: In S3, the method for obtaining the optimal control speech is: using the control speech target function value as the quality factor of the high-pass filter, and using the high-pass filter to process the control speech to obtain the optimal control speech.

Images