Voiceprint feature extraction method based on multi-microphone fusion

Abstract

The invention discloses a voiceprint feature extraction method based on multi-microphone fusion, and the method comprises the steps: synchronously collecting the sound signal data of a plurality of microphones, extracting the voiceprint feature vector xi of the plurality of microphones, calculating the fault classification vector ci of each microphone according to a pre-trained classification model, obtaining a fusion classification vector t according to the fault classification vector ci, and extracting the voiceprint feature vector xi of each microphone according to the fusion classification vector t. The weight wi of each microphone is calculated according to the fault classification vector ci and the fusion classification vector t, and the fused voiceprint feature x is obtained according to the calculated weight wi. According to the method, the accurate signal feature is obtained by adopting the mode that the multiple microphones are fused to extract the feature, the accuracy and reliability of fault recognition are improved, and the fault recognition efficiency is improved. According to the method, abnormal microphones can be found, timely replacement and maintenance are facilitated, fusion voiceprint feature extraction of the multiple microphones is achieved, and suspected faulty microphones can be found by observing weight values.

Description

Technical Field

[0001] The present invention relates to the technical field of voiceprint monitoring fault identification, and in particular to a voiceprint feature extraction method based on multi-microphone fusion. Background Art

[0002] Substation equipment, such as transformers, are responsible for voltage conversion, power distribution, and interconnection between power grids of different voltage levels. Their safe operation is related to the stability of the power system. Therefore, transformer fault monitoring is very important. Common transformer faults include partial discharge, DC bias, loose core, overload, etc.

[0003] Traditional online fault monitoring relies on on-site inspections by operation and maintenance personnel. The inspectors use their ears to judge the fault type based on experience. This method has high requirements for inspectors and it is difficult to detect faults in time. Voiceprint monitoring technology monitors the sound emitted by the transformer and analyzes it in real time. It uses an artificial intelligence classification model trained with historical data to predict faults, improves the accuracy of fault identification, and realizes non-stop detection of equipment. Voiceprint monitoring uses a microphone to collect sound signals and extracts multiple features such as time domain, frequency domain, and time-frequency domain from the signals. These features are input into machine learning or deep learning classification models to determine the fault type of substation equipment.

[0004] Since a single microphone has the risk of performance degradation or even damage, it is impossible to correctly extract the voiceprint signal features, resulting in misjudgment. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] Therefore, the purpose of the present invention is to provide a voiceprint feature extraction method based on multi-microphone fusion, which obtains accurate signal characteristics by fusing multiple microphones to extract features, improves the accuracy and reliability of fault identification, and is also conducive to discovering abnormal microphones for timely replacement and repair.

[0007] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0008] A voiceprint feature extraction method based on multi-microphone fusion, the specific steps are as follows:

[0009] S1. Synchronously collect the sound signal data of multiple microphones and extract the voiceprint feature vector x of the multiple microphones i ;

[0010] S2. Calculate the fault classification vector c of each microphone based on the pre-trained classification model i ;

[0011] S3, according to the fault classification vector c i Obtain the fusion classification vector t, and according to the fault classification vector c i And the fusion classification vector t calculates the weight w of each microphone i ;

[0012] S4, according to the calculation weight w i Get the fused voiceprint feature x.

[0013] As a preferred solution of the voiceprint feature extraction method based on multi-microphone fusion described in the present invention, in step S1, the specific steps of synchronously collecting multi-channel microphone sound signal data are as follows: synchronously collect a signal x of length N through n microphones i (j), where 1≤i≤n, 1≤j≤N, x i (j) represents the signal amplitude.

[0014] As a preferred solution of the voiceprint feature extraction method based on multi-microphone fusion described in the present invention, in step S1, the voiceprint feature vector x of the multi-microphones is extracted. i The specific steps are as follows:

[0015] Based on the acquired signal x of length N i (j), calculate the m-dimensional voiceprint feature vector x of microphone i i ;

[0016] x i =[x i1 ,x i2 ,...x im ],1≤i≤n,

[0017] Among them, x iq is the eigenvalue extracted from the signal collected by microphone i, and 1≤q≤m. These eigenvalues ​​are the signal x i (j) time domain, frequency domain or time-frequency domain characteristics.

[0018] As a preferred solution of the voiceprint feature extraction method based on multi-microphone fusion described in the present invention, the fault classification vector c of each microphone is calculated according to the pre-trained classification model. i The specific steps are as follows:

[0019] The voiceprint feature vector x i Input into a pre-trained classification model based on machine learning or deep learning, where the classification model is trained with a large number of samples;

[0020] Voiceprint feature vector x i After inputting the classification model, the fault classification vector c of microphone i is output i =[c i1 ,c i2 ,...c ip ];

[0021] Among them, 1≤i≤n, p represents the number of categories that the model can output, c ik represents the probability that the classification model predicts that it is the kth type of fault, and 1≤k≤p, that is,

[0022] As a preferred solution of the voiceprint feature extraction method based on multi-microphone fusion described in the present invention, in step S3, according to the fault classification vector c i The specific steps to obtain the fused classification vector t are as follows:

[0023] First, according to each c i Get the number of classification results 1≤i≤n, that is, c ik Take the largest k value, and you can get n r i value;

[0024] Then, from these n r i The value with the largest number of views among the values ​​is taken as the number of fusion classification results f.

[0025] Finally, determine the fusion classification vector t = [t1, t2, ... t p ], where t f =1, the rest t j =0,j≠f.

[0026] As a preferred solution of the voiceprint feature extraction method based on multi-microphone fusion described in the present invention, in step S3, according to the fault classification vector c i And the fusion classification vector t calculates the weight w of each microphone i The specific steps are as follows:

[0027] According to the calculation formula: Calculate the weight w i ;

[0028] Observe these n weights w i ,If a certain weight value is significantly different from other weight values, it is inferred that the microphone corresponding to the weight value may be abnormal and needs to be inspected, repaired or replaced.

[0029] As a preferred solution of the voiceprint feature extraction method based on multi-microphone fusion described in the present invention, in step S4, according to the calculation weight w i The specific calculation formula for obtaining the fused voiceprint feature x is:

[0030]

[0031] Compared with the prior art, the present invention has the following beneficial effects: the present invention adopts a method of fusing and extracting features from multiple microphones to obtain accurate signal features, improve the accuracy and reliability of fault identification, and is also conducive to discovering abnormal microphones for timely replacement and repair. It can realize the fusion voiceprint feature extraction of multiple microphones, and by observing the weight value, the suspected faulty microphone can be discovered. Through the weighted method, the adverse effect of the abnormal microphone on feature extraction can be suppressed. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below in combination with the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0033] Figure 1 The present invention is a flow chart of a voiceprint feature extraction method based on multi-microphone fusion. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0035] Secondly, the present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0036] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0037] The present invention provides a voiceprint feature extraction method based on multi-microphone fusion, which obtains accurate signal features by fusing and extracting features through multiple microphones, improves the accuracy and reliability of fault identification, and is also conducive to discovering abnormal microphones for timely replacement and repair.

[0038] like Figure 1As shown in the figure, the voiceprint feature extraction method based on multi-microphone fusion has the following specific steps:

[0039] Step S1, assuming there are n microphones, synchronously collect a signal x of length N i (j), where 1≤i≤n, 1≤j≤N, x i (j) represents the signal amplitude.

[0040] Step S2, based on the signal x of length N acquired in step S1 i (j), calculate the m-dimensional voiceprint feature vector x of microphone i i =[x i1 ,x i2 ,...x im ],1≤i≤n, where x iq ,1≤q≤m is the eigenvalue extracted from the signal collected by microphone i. These eigenvalues ​​can be the signal x i The time domain, frequency domain or time-frequency domain characteristics of (j) are calculated using existing technologies, such as calculating signal energy, zero-crossing rate, spectrum, wavelet coefficients, etc.

[0041] Step S3: The m-dimensional voiceprint feature vector x obtained by microphone i in step S2 is i , is input into a pre-trained classification model based on machine learning or deep learning (neural network). The pre-trained model is generated using existing technology and obtained through a large number of sample training. i After inputting the classification model, the fault classification vector c of microphone i can be output i =[c i1 ,c i2 ,...c ip ],1≤i≤n, where p represents the number of categories that the model can output, c ik , 1≤k≤p represents the probability that the classification model predicts that it is the kth type of fault, that is,

[0042] Step S4: based on the fault classification vector c of the n microphones obtained in step S3 i =[c i1 ,c i2 ,...c ip ],1≤i≤n, calculate the fusion classification vector t=[t1,t2,...t p ], the specific calculation method is: first, according to each c i Get the number of classification results 1≤i≤n, that is, c ik Take the largest k value, and you can get n r i value; then, from these n r iThe value with the largest number of values ​​is taken as the number of fusion classification results f. For example, suppose there are 7 microphones and their corresponding classification results are {1, 2, 3, 3, 2, 3, 3}, of which there are 4 3s, 2 2s, and 1 1. Since 3 has the largest number, the number of fusion classification results f is 3. Finally, determine the fusion classification vector t = [t1, t2, ...t p ], where t f =1, the rest t j =0,j≠f.

[0043] Step S5, based on the fault classification vector c of the n microphones in step S3 i =[c i1 ,c i2 ,...c ip ], 1≤i≤n, and the fused classification vector t=[t1,t2,...t p ], calculate the weight w of each microphone i , the calculation formula is: Observe these n weights w i If a certain weight value is significantly different from other weight values, it can be inferred that the microphone corresponding to the weight value may be abnormal and should be inspected, repaired or replaced. Because when all microphones are normal, the weights of each microphone w i The difference is not big.

[0044] Step S6: m-dimensional feature vector x of each microphone i obtained in step S2 i =[x i1 ,x i2 ,...x im ], 1≤i≤n and the n weights w obtained in step S5 i , calculate the characteristic vector x of the fusion of n microphones, the calculation formula is:

[0045] Through the above steps, the fusion voiceprint feature extraction of multiple microphones can be realized. By observing the weight value w i , the suspected faulty microphone can be found; through the weighted method, the adverse effect of abnormal microphone on feature extraction can be suppressed.

[0046] Although the present invention has been described above with reference to the embodiments, various modifications may be made thereto and parts thereof may be replaced by equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the various features in the embodiments disclosed in the present invention may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A voiceprint feature extraction method based on multi-microphone fusion, characterized in that: The specific steps are as follows: S1. Synchronously collect the sound signal data of multiple microphones and extract the voiceprint feature vector x of the multiple microphones i ; S2. Calculate the fault classification vector c of each microphone according to the pre-trained classification model i ; S3, according to the fault classification vector c i Obtain the fusion classification vector t, and according to the fault classification vector c i And the fusion classification vector t calculates the weight w of each microphone i ; S4, according to the calculation weight w i Get the fused voiceprint feature x.

2. The voiceprint feature extraction method based on multi-microphone fusion according to claim 1 is characterized in that: In step S1, the specific steps of synchronously collecting the sound signal data of multiple microphones are as follows: synchronously collect a signal x of length N through n microphones i (j), where 1≤i≤n, 1≤j≤N, x i (j) represents the signal amplitude.

3. The voiceprint feature extraction method based on multi-microphone fusion according to claim 2 is characterized in that: In step S1, the voiceprint feature vector x of the multi-channel microphone is extracted i The specific steps are as follows: Based on the acquired signal x of length N i (j), calculate the m-dimensional voiceprint feature vector x of microphone i i ; x i =[x i1 ,x i2 ,...x im ],1≤i≤n, Among them, x iq is the eigenvalue extracted from the signal collected by microphone i, and 1≤q≤m. These eigenvalues ​​are the i (j) time domain, frequency domain or time-frequency domain characteristics.

4. The voiceprint feature extraction method based on multi-microphone fusion according to claim 1 is characterized in that: Calculate the fault classification vector c of each microphone based on the pre-trained classification model i The specific steps are as follows: The voiceprint feature vector x i Input into a pre-trained classification model based on machine learning or deep learning, where the classification model is trained with a large number of samples; Voiceprint feature vector x i After inputting the classification model, the fault classification vector c of microphone i is output i =[c i1 ,c i2 ,...c ip ]; Among them, 1≤i≤n, p represents the number of categories that the model can output, c ik represents the probability that the classification model predicts that it is the kth type of fault, and 1≤k≤p, that is, 5. The voiceprint feature extraction method based on multi-microphone fusion according to claim 1 is characterized in that: In step S3, according to the fault classification vector c i The specific steps to obtain the fused classification vector t are as follows: First, according to each c i Get the number of classification results 1≤i≤n, that is, c ik Take the largest k value, and you can get n r i value; Then, from these n r i The value with the largest number of views among the values ​​is taken as the number of fusion classification results f; Finally, determine the fusion classification vector t = [t1, t2, ... t p ], where t f =1, the rest t j =0,j≠f.

6. The voiceprint feature extraction method based on multi-microphone fusion according to claim 1 is characterized in that: In step S3, according to the fault classification vector c i And the fusion classification vector t calculates the weight w of each microphone i The specific steps are as follows: According to the calculation formula: Calculate the weight w i ; Observe these n weights w i ,If a certain weight value is significantly different from other weight values, it is inferred that the microphone corresponding to the weight value may be abnormal and needs to be inspected, repaired or replaced.

7. The voiceprint feature extraction method based on multi-microphone fusion according to claim 1 is characterized in that: In step S4, the weight w is calculated according to the i The specific calculation formula for obtaining the fused voiceprint feature x is:

Images