CN102982351A - Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network - Google Patents
Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network Download PDFInfo
- Publication number
- CN102982351A CN102982351A CN2012104593469A CN201210459346A CN102982351A CN 102982351 A CN102982351 A CN 102982351A CN 2012104593469 A CN2012104593469 A CN 2012104593469A CN 201210459346 A CN201210459346 A CN 201210459346A CN 102982351 A CN102982351 A CN 102982351A
- Authority
- CN
- China
- Prior art keywords
- neural network
- porcelain insulator
- data
- test data
- vibrational
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The invention discloses a porcelain insulator vibrational acoustics test data sorting technique based on a back propagation (BP) neural network. The process comprises the steps of designing a BP neural network, taking at least one typical data of each class from the three classes of porcelain insulator vibrational acoustics test data after the noise reduction, generating feature vector as the input vector and conducting training to the BP neural network and then obtaining the trained neural network, and the test input vector is generated by the porcelain insulator vibrational acoustics test data, meanwhile conducting fast fourier transformation (FFT) to the porcelain insulator vibrational acoustics test data to generate a spectrogram and identify the spectrogram according to the test data file name, and then inputting the test input vector in the trained neural network, and therefore the neural network output is a certain of insulator defect form class codes; storing the generated spectrogram in the folder corresponding to the defect form code. The porcelain insulator vibrational acoustics test data sorting technique based on the BP neural network can classify the porcelain insulator vibrational acoustics test data rapidly without manual intervention and is high in classification effectiveness.
Description
Technical field
The present invention relates to a kind of porcelain insulator vibroacoustics based on the BP neural network and detect data classification method, can carry out the artificial intelligence automatic classification to the data result of acoustic vibration method detection porcelain insulator.
Background technology
Porcelain insulator needs regularly to detect, and whether it has damage with interpretation, and the testing result of porcelain insulator is divided three kinds: upper end defectiveness, lower end defectiveness, zero defect.When insulator is carried out the detection of vibroacoustics method, the testing result data that store are the audio files of WAV form, need to could interpretation with the supporting software of checkout equipment its power spectral density plot being read out, this software at every turn multipotency reads 6 data simultaneously, reading all needs to click the mouse at every turn carries out multi-pass operations, after the complete data of interpretation are closed, also need again function software when thinking again interpretation, be not easy to simultaneously many data and cumbersome to the power spectral density plot access of data.For the detection of large-scale substation, once just produce hundreds of data, the interpretation of manually finishing data is wasted time and energy.
Summary of the invention
The technical problem to be solved in the present invention provides a kind of porcelain insulator vibroacoustics based on the BP neural network that improves classification effectiveness and detects data classification method.
For solving the problems of the technologies described above, the technical solution used in the present invention is: a kind of porcelain insulator vibroacoustics based on the BP neural network detects data classification method, and it may further comprise the steps:
1) adopts the MATLAB language, design a BP nerve network system;
2) for detecting data through the porcelain insulator vibroacoustics that is divided into 3 classes behind the noise reduction, every class is taken to few 1 typical data, adopts Mel cepstrum coefficient method generating feature vector, as input vector, obtain training sample, with defects of insulator formal classification code as output;
3) utilize step 2) in the training sample that obtains the BP neural network in the step 1) is trained, the BP neural network after obtaining training, the input layer number of this BP neural network is 24, the output layer nodes is 3;
4) the porcelain insulator vibroacoustics is detected data and generate the test input vector, simultaneously this porcelain insulator vibroacoustics being detected data carries out generating spectrogram after the FFT conversion and labelling to detect data file name, then with the BP neural network after the test input vector input training, its neural network output is some in the defects of insulator formal classification code;
5) spectrogram that generates in the step 4) is left in the file corresponding with BP neural network output vector; After being disposed, described spectrogram is divided into 3 classes, and namely porcelain insulator vibroacoustics detection data are divided into three classes.
Above-mentioned steps 2) noise-reduction method in adopts the wavelet decomposition noise reduction.
Above-mentioned Sort Code is divided into three classes, is designated as respectively [0 0 1], [0 1 0], [1 0 0].
Above-mentioned steps 2) in, when every class extracts a plurality of typical data (greater than 1), a plurality of typical datas is adopted Mel frequency cepstral coefficient method generating feature value; Then extract the Partial Feature vector of each typical data, these proper vectors that extract are formed one group of complete proper vector as input vector, the BP neural network is trained.
Principle of work of the present invention is as follows:
The present invention is directed to existing low, the manually-operated shortcoming of sorting technique efficient, a kind of porcelain insulator vibroacoustics testing result sorting technique based on the BP neural network is provided, utilize the good nonlinear function approximation capability of BP neural network and self-organized learning function, automatically finish the classification problem that porcelain insulator vibroacoustics in enormous quantities detects data.Porcelain insulator vibroacoustics method is detected the data file of acquisition as the input variable of BP neural network, behind network training, the relation between whether constructed nerve network system can be automatically damages the acoustics corresponding information of porcelain insulator and its inside is carried out nonlinear analysis and classification.
Artificial neural network ANN is the novel information disposal system of the brain functions such as 26S Proteasome Structure and Function, cranial nerve structure and thinking processing problem of imitation brain cell.It has complicated dynamics, parallel processing mechanism, and study, association and memory function also have self-organization, the adaptive ability of height.Have powerful mode identificating ability, by the great amount of samples study to reflection input feature vector amount, can classify and identify any complex state or process.Counterpropagation network (Back Propagation Net-work, be called for short the Bp network) be the most representative in the present artificial neural network pattern, use to get the most a kind of model, have self study, self-organization, self-adaptation and very strong non-linear mapping capability, can approach the arbitrary continuation function with arbitrary accuracy, therefore in the classification to the porcelain insulator testing result, can adopt the BP neural network to carry out automatic classification.
Because the data file that insulator is detected is sound signal, therefore to the classification of insulator testing result based on the identification basis to sound signal.In the audio identification field, the quality of feature extraction directly has influence on the result of identification.It is linear prediction cepstrum coefficient parameter (LPCC) and Mel frequency cepstral coefficient method (MFCC) that existing sound recognition system adopts more sound characteristic.The LPCC parameter can well embody people's sound channel feature, and calculated amount is little, but noise robustness is poor.The MFCC coefficient has been considered the auditory properties of people's ear, frequency spectrum is converted into non-linear frequency spectrum based on the Mel frequency, then be transformed on the cepstrum domain, use a string triangle filter of arranging at the low frequency region juxtaposition, catch the spectrum information of sound, therefore need to adopt the MFCC method that voice signal is carried out pre-service.
Neural Network Toolbox among the MATLAB (Neural Network Toolbox is called for short NNbox), the condition of providing convenience for setting up the artificial neural network system.The Neural Network Toolbox function is very perfect, various MATLAB functions are provided, the functions such as foundation, training and emulation that comprise neural network, and various improvement training algorithm functions, the user can carry out design and the emulation of neural network easily, also can suitably revise on the basis of MATLAB source file, form the kit of oneself to satisfy actual needs, therefore process the nerve network system that detects data and adopt the MATLAB programming.
The beneficial effect that adopts technique scheme to produce is: adopt the present invention to detect data to porcelain insulator vibroacoustics method fast and classify, and need not manual intervention, reduce the artificial subjective factor of classification results, the efficient of classification is high.
Description of drawings
Fig. 1 is the workflow diagram of neural network;
Fig. 2 is the synoptic diagram that concerns of Mel frequency and linear frequency;
Fig. 3 is the leaching process synoptic diagram of MFCC parameter;
Fig. 4 is Mel frequency filter group synoptic diagram.
Embodiment
Below the present invention is further detailed explanation.
Referring to accompanying drawing 1, detect typical case that upper end defectiveness, lower end defectiveness, flawless porcelain insulator produce with the acoustic vibration method and detect data and behind the wavelet decomposition noise reduction nerve network system is trained, the expectation neural network can detect other data and be divided into three such classes.
Network system judges whether also have non-classified data under the current directory, if having then process, if not then finish this categorizing system.
If when under the current directory unfiled data being arranged, at first unfiled data are carried out the wavelet decomposition noise reduction.Remove the noise signal of impact classification in the data, then data are carried out the conversion of Mel cepstrum coefficient.Then the data after the conversion are classified.
Thinking of the present invention is to utilize artificial intelligence neural networks to make up an intelligent classification system model, utilizes this disaggregated model that the porcelain insulator vibroacoustics is detected data and classifies, and the below is elaborated to the process of setting up of this disaggregated model:
1) generation of input vector
In the audio identification field, the quality of feature extraction directly has influence on the result of identification.Adopting in the present invention Mel cepstrum coefficient method is that the MFCC method is carried out pre-service to voice signal, and the size of pressing critical bandwidth from the low frequency to the high frequency in this section frequency band to one group of bandpass filter of rare arrangement, is carried out filtering to input signal by close.With the signal essential characteristic of signal energy of each bandpass filter output, to just can be used as the input feature vector of voice after the further processing of this feature process.Because this feature does not rely on the character of signal, input signal is not done any hypothesis and restriction, utilized again the achievement in research of auditory model, therefore, this parameter has preferably robustness, and still has preferably recognition performance when signal to noise ratio (S/N ratio) reduces.
MFCC is the cepstrum parameter that extracts in Mel scale frequency territory, and the Mel scale has been described the nonlinear characteristic of people's ear frequency, the available following formula approximate representation of relation of it and frequency:
In the formula
Be frequency, unit is Hz.Fig. 2 has shown the relation of Mel frequency and linear frequency:
The method of asking the Mel cepstrum coefficient is when time-domain signal is done/the frequency conversion after, use the triangular filter group that distributes according to the Mel scale to do convolution to its logarithm energy spectrum, again the output vector of bank of filters is done discrete cosine transform (DCT), the front N dimensional vector that obtains like this is called MFCC.
The leaching process of Mel cepstrum coefficient is as shown in Figure 3:
A. divide frame, windowing to certain insulator vibroacoustics detection data of input, then make discrete Fourier transform (DFT), obtain spectrum distribution information.If detect the DFT of data be:
In the formula
Be the detection data of input,
Counting of expression Fourier transform.
B. ask again spectrum amplitude square, obtain energy spectrum.
C. with the triangle filter group of energy spectrum by one group of Mel yardstick.
Define a bank of filters (number of wave filter is close with the number of critical band) that M wave filter arranged, the wave filter of employing is triangular filter, and centre frequency is
, the present invention gets
Each
Between the interval dwindle along with reducing of m value, along with
The value increase and broadening, as shown in Figure 4;
The frequency response of triangular filter is defined as:
Wherein
D. the logarithm energy that calculates each bank of filters output is:
E. obtain the MFCC coefficient through discrete cosine transform (DCT):
MFCC coefficient exponent number is got 12-16 usually, and the present invention chooses 16 rank cepstrum coefficients as proper vector.
The selection of output vector
Insulator in use defective appears at the two ends flange usually, and defective seldom appears in center section.Therefore, in the porcelain insulator vibroacoustics detects, according to defective locations, usually testing result is divided three classes: upper end defectiveness, lower end defectiveness, zero defect.With this three class result coding, be respectively 3 vectors such as [0 0 1], [0 1 0], [1 0 0], as Output rusults.
) training of BP neural network
The BP algorithm is the basic skills of the training of neural network training, least-squares algorithm during its basic thought.Its adopts gradient search technology, is minimum in the hope of the error mean square root of the real output value that makes network and desired output.x
1, x
2... x
nBe BP neural network input variable, d
1, d
2... d
nBe BP neural network desired output, W
IjBe the weights of input layer and hidden layer, W
JkBe the weights of hidden layer and output layer, the input node is n, and hidden layer node can be made as L=2n, and output node is m, and the concrete steps of its training are as follows:
A. BP neural network initialization: according to input vector, input layer is connected weights W with hidden layer
IjAnd hidden layer threshold value a, output layer threshold value b, given learning rate and neuron excitation function;
B. calculating hidden layer output calculates: according to input vector, input layer is connected weights Wij and hidden layer threshold value a with hidden layer, calculates hidden layer output H:
In the formula: L is the hidden layer node number; F is the hidden layer excitation function, and the present invention selects function to be:
C. output layer output is calculated: according to hidden layer output H, connect weights Wjk and threshold value b, calculate BP neural network classification output U:
D. error is calculated: according to network class output U and desired output d, and computational grid error in classification e:
E. right value update: according to the network class error e, upgrade network connection weights W
IjAnd W
Jk:
F. threshold value is upgraded: according to the network class error e, upgrade network node threshold value a, b:
G. whether the evaluation algorithm iteration finishes, if do not introduce, returns step b.
Because the present invention generate 500 group of 24 dimensional feature vector, so the input layer of the BP neural network among the present invention is made as 24 when insulator vibroacoustics detection data are carried out the conversion of Mel cepstrum coefficient.The insulator vibroacoustics detects data and is divided into 3 classes, so the output layer node is 3.When training, 500 group of 24 dimensional feature vector inputted as a batch data.In order to improve classification accuracy, 24 dimensional feature vectors that can extract a plurality of typical datas generations in certain class form together 500 group of 24 dimensional feature vector and carry out network training.
Embodiment
In order to verify validity of the present invention, carried out following experiment: with each 10 of Hebei South Power Network transformer station 3 classes in 2012 totally 30 porcelain insulator vibroacousticss detect data through the filtering noise reduction after and the data of correctly classifying as the training sample of disaggregated model, remaining detects data as test sample book through 128 porcelain insulator vibroacousticss of also correctly classifying behind the noise reduction take this transformer station, in 128 data, the defective data number in upper end is 9, several 23 of the defective data in lower end, flawless data have 96.Training and set up the porcelain insulator vibroacoustics and detect data classification model under the MATLAB environment at first utilizes Mel cepstrum coefficient method that 30 training samples are carried out proper vector and extracts.May further comprise the steps:
Step 1: certain the insulator vibroacoustics detection data to input are divided frame, windowing, then make discrete Fourier transform (DFT), obtain spectrum distribution information;
Step 2: ask again spectrum amplitude square, obtain energy spectrum;
Step 3: with the triangle filter group of energy spectrum by one group of Mel yardstick;
Step 4: the logarithm energy that calculates each bank of filters output;
Step 5: obtain 500 group of 24 dimension MFCC coefficient characteristics vector through discrete cosine transform (DCT).
Respectively get one in proper vector three classes that step 5 is obtained and as input value neural network is trained, BP neural network initial learn rate is made as 0.1, and the input layer number is 24, and the hidden layer node number is 30, output layer node numerical digit 3.After training is finished 128 test sample books are carried out class test, class test is carried out 3 times, averages as final classification accuracy result.Classification results sees Table the capable data of 1 PT.
Extract all kinds of in in each proper vector that is obtained by step 5 500 groups any 50 groups, it is 1 group that the proper vector that the porcelain insulator vibroacoustics of similar defective is detected data is put, form 500 group of 24 new dimensional feature vector, as sample space neural network is trained.BP neural network initial learn rate is made as 0.1, and the input layer number is 24, and the hidden layer node number is 30, output layer node numerical digit 3.After training is finished 128 test sample books are carried out class test, class test is carried out 3 times, averages as final classification accuracy result.Classification results sees Table the capable data of 1JQ.
Table 1 neural network classification accuracy
As can be seen from Table 1, many training samples so that classification accuracy be significantly improved.
Claims (4)
1. the porcelain insulator vibroacoustics based on the BP neural network detects data classification method, it is characterized in that: may further comprise the steps:
1) adopts the MATLAB language, design a BP nerve network system;
2) for detecting data through the porcelain insulator vibroacoustics that is divided into 3 classes behind the noise reduction, every class is taken to few 1 typical data, adopts Mel cepstrum coefficient method generating feature vector, as input vector, obtain training sample, with defects of insulator formal classification code as output;
3) utilize step 2) in the training sample that obtains the BP neural network in the step 1) is trained, the BP neural network after obtaining training, the input layer number of this BP neural network is 24, the output layer nodes is 3;
4) the porcelain insulator vibroacoustics is detected data and generate the test input vector, simultaneously this porcelain insulator vibroacoustics being detected data carries out generating spectrogram after the FFT conversion and labelling to detect data file name, then with the BP neural network after the test input vector input training, its neural network output is some in the defects of insulator formal classification code;
5) spectrogram that generates in the step 4) is left in the file corresponding with BP neural network output vector; After being disposed, described spectrogram is divided into 3 classes, and namely porcelain insulator vibroacoustics detection data are divided into three classes.
2. the porcelain insulator vibroacoustics based on the BP neural network according to claim 1 detects data classification method, it is characterized in that: the described noise-reduction method described step 2) adopts the wavelet decomposition noise reduction.
3. the porcelain insulator vibroacoustics based on the BP neural network according to claim 1 detects data classification method, and it is characterized in that: described Sort Code is divided into three classes, is designated as respectively [0 0 1], [0 1 0], [1 0 0].
4. the porcelain insulator vibroacoustics based on the BP neural network according to claim 1 detects data classification method, it is characterized in that: described step 2) when every class extracts a plurality of typical data, a plurality of typical datas are adopted Mel cepstrum coefficient method generating feature value; Then extract the Partial Feature vector of each typical data, these proper vectors that extract are formed one group of complete proper vector as input vector, the BP neural network is trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012104593469A CN102982351A (en) | 2012-11-15 | 2012-11-15 | Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012104593469A CN102982351A (en) | 2012-11-15 | 2012-11-15 | Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102982351A true CN102982351A (en) | 2013-03-20 |
Family
ID=47856333
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012104593469A Pending CN102982351A (en) | 2012-11-15 | 2012-11-15 | Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102982351A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103245524A (en) * | 2013-05-24 | 2013-08-14 | 南京大学 | Acoustic fault diagnosis method based on neural network |
| CN105956658A (en) * | 2016-04-29 | 2016-09-21 | 北京比特大陆科技有限公司 | Data processing method, data processing device and chip |
| CN109522445A (en) * | 2018-11-15 | 2019-03-26 | 辽宁工程技术大学 | A kind of audio classification search method merging CNNs and phase algorithm |
| CN113361557A (en) * | 2020-12-21 | 2021-09-07 | 南京仁智网络科技有限公司 | Training method of neural network for underground coal mine fire extinguishing control based on vibration data |
| CN113642714A (en) * | 2021-08-27 | 2021-11-12 | 国网湖南省电力有限公司 | Insulator pollution discharge state identification method and system based on small sample learning |
| CN113657021A (en) * | 2021-07-15 | 2021-11-16 | 交通运输部水运科学研究所 | Marine measurement period evaluation method based on BP neural network |
| CN113673135A (en) * | 2021-09-02 | 2021-11-19 | 河南工业大学 | Local resonance type acoustic metamaterial band gap adjusting and controlling method and system and storable medium |
| CN113739905A (en) * | 2020-05-27 | 2021-12-03 | 现代摩比斯株式会社 | Apparatus and method for locating noise occurring in steering system |
| CN114253308A (en) * | 2020-09-21 | 2022-03-29 | 陕西环保产业研究院有限公司 | Active control method and device for vibration of space frame structure |
| US20220318553A1 (en) * | 2021-03-31 | 2022-10-06 | Qualcomm Incorporated | Adaptive use of video models for holistic video understanding |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101995436A (en) * | 2010-10-26 | 2011-03-30 | 江苏大学 | Neural network based drawing part crack monitoring method |
-
2012
- 2012-11-15 CN CN2012104593469A patent/CN102982351A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101995436A (en) * | 2010-10-26 | 2011-03-30 | 江苏大学 | Neural network based drawing part crack monitoring method |
Non-Patent Citations (4)
| Title |
|---|
| 张万里等: ""MEL频率倒谱系数提取及其在声纹识别中的作用"", 《贵州大学学报》, vol. 22, no. 2, 31 May 2005 (2005-05-31) * |
| 郝晓军等: ""基于振动声学原理的瓷支柱绝缘子检测技术"", 《中国计量协会冶金分会2012年会论文集》, 20 September 2012 (2012-09-20) * |
| 陈嫄妗等: ""神经网络及BP网络工作原理"", 《郑州牧业工程高等专科学校学报》, vol. 21, no. 4, 30 November 2001 (2001-11-30) * |
| 马崇等: ""振动声学法在瓷支柱绝缘子检测中的应用"", 《河北电力技术》, vol. 31, no. 1, 29 February 2012 (2012-02-29) * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103245524A (en) * | 2013-05-24 | 2013-08-14 | 南京大学 | Acoustic fault diagnosis method based on neural network |
| CN105956658A (en) * | 2016-04-29 | 2016-09-21 | 北京比特大陆科技有限公司 | Data processing method, data processing device and chip |
| CN109522445A (en) * | 2018-11-15 | 2019-03-26 | 辽宁工程技术大学 | A kind of audio classification search method merging CNNs and phase algorithm |
| CN113739905A (en) * | 2020-05-27 | 2021-12-03 | 现代摩比斯株式会社 | Apparatus and method for locating noise occurring in steering system |
| CN114253308B (en) * | 2020-09-21 | 2022-08-30 | 陕西环保产业研究院有限公司 | Active control method and equipment for vibration of space frame structure |
| CN114253308A (en) * | 2020-09-21 | 2022-03-29 | 陕西环保产业研究院有限公司 | Active control method and device for vibration of space frame structure |
| CN113361557A (en) * | 2020-12-21 | 2021-09-07 | 南京仁智网络科技有限公司 | Training method of neural network for underground coal mine fire extinguishing control based on vibration data |
| US20220318553A1 (en) * | 2021-03-31 | 2022-10-06 | Qualcomm Incorporated | Adaptive use of video models for holistic video understanding |
| US11842540B2 (en) * | 2021-03-31 | 2023-12-12 | Qualcomm Incorporated | Adaptive use of video models for holistic video understanding |
| CN113657021A (en) * | 2021-07-15 | 2021-11-16 | 交通运输部水运科学研究所 | Marine measurement period evaluation method based on BP neural network |
| CN113642714A (en) * | 2021-08-27 | 2021-11-12 | 国网湖南省电力有限公司 | Insulator pollution discharge state identification method and system based on small sample learning |
| CN113642714B (en) * | 2021-08-27 | 2024-02-09 | 国网湖南省电力有限公司 | Insulator pollution discharge state identification method and system based on small sample learning |
| CN113673135A (en) * | 2021-09-02 | 2021-11-19 | 河南工业大学 | Local resonance type acoustic metamaterial band gap adjusting and controlling method and system and storable medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102982351A (en) | Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network | |
| CN109065030B (en) | Convolutional neural network-based environmental sound identification method and system | |
| Lee et al. | Sample-level deep convolutional neural networks for music auto-tagging using raw waveforms | |
| CN108899051B (en) | Speech emotion recognition model and recognition method based on joint feature representation | |
| CN107610707B (en) | A kind of method for recognizing sound-groove and device | |
| CN110428842A (en) | Speech model training method, device, equipment and computer readable storage medium | |
| CN108630209B (en) | Marine organism identification method based on feature fusion and deep confidence network | |
| CN109256118B (en) | End-to-end Chinese dialect identification system and method based on generative auditory model | |
| CN101923855A (en) | Test-irrelevant voice print identifying system | |
| CN109949823A (en) | A kind of interior abnormal sound recognition methods based on DWPT-MFCC and GMM | |
| CN102982803A (en) | Isolated word speech recognition method based on HRSF and improved DTW algorithm | |
| CN104978507A (en) | Intelligent well logging evaluation expert system identity authentication method based on voiceprint recognition | |
| CN108198561A (en) | A kind of pirate recordings speech detection method based on convolutional neural networks | |
| Mesgarani et al. | Speech discrimination based on multiscale spectro-temporal modulations | |
| CN113327626A (en) | Voice noise reduction method, device, equipment and storage medium | |
| CN112786057B (en) | Voiceprint recognition method and device, electronic equipment and storage medium | |
| Chen et al. | An audio scene classification framework with embedded filters and a DCT-based temporal module | |
| CN115762533A (en) | Bird song classification and identification method and device | |
| KS et al. | Comparative performance analysis for speech digit recognition based on MFCC and vector quantization | |
| CN113252323B (en) | Breaker mechanical fault identification method and system based on human ear hearing characteristics | |
| CN112329819A (en) | Underwater target identification method based on multi-network fusion | |
| Kaminski et al. | Automatic speaker recognition using a unique personal feature vector and Gaussian Mixture Models | |
| CN104102834A (en) | Method for identifying sound recording locations | |
| CN112735477B (en) | Voice emotion analysis method and device | |
| CN114141256A (en) | Voiceprint feature extraction model construction method and system based on wavelet neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130320 |