CN105118516A - Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) - Google Patents
Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) Download PDFInfo
- Publication number
- CN105118516A CN105118516A CN201510637964.1A CN201510637964A CN105118516A CN 105118516 A CN105118516 A CN 105118516A CN 201510637964 A CN201510637964 A CN 201510637964A CN 105118516 A CN105118516 A CN 105118516A
- Authority
- CN
- China
- Prior art keywords
- engineering machinery
- lpcc
- sound
- voice signal
- linear prediction
- 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
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides an identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC), and the method comprises the following steps: collecting background noise signals in a real environment on a construction spot; placing the engineering machinery in an ideal environment, and starting the engineering machinery; collecting sound signals sent by the engineering machinery by a sound colleting device; performing endpoint detection, windowing and framing for the sound signals, wherein the endpoint detection is realized by a zero-crossing rate, and the sound signal framing is carried out by adopting Hamming window; extracting the LPCC from every framed signal; establishing an engineering machinery sound feature fingerprint database by adopting a support vector machine; extracting the LPCC from the real-time sound signals, then performing matching between the support vector machine and the feature fingerprint database so as to realize classification.
Description
Technical field
The present invention relates to voice recognition technology field, specifically, relate to a kind of engineering machinery recognition methods based on sound linear prediction residue error.
Background technology
Due to reasons such as land used anxiety, appearance of the city construction, city transmission of electricity major part adopts the form of cable laying, and compared with overground cable, underground cable transmission of electricity has that floor area is little, antijamming capability strong, be conducive to the appearance of the city advantage such as to beautify.But owing to there is a large amount of nonstandard road construction engineer operation, engineering machinery is destroyed to make underground cable suffer, the stability affecting electric power system also can bring huge inconvenience to people's lives and commercial production simultaneously.In road construction process, the destroyed engineering machinery of underground cable is very easily caused to mainly contain mechanical percussion hammer, cutting machine, hand-held electric pick etc.The voice signal produced when operation by gathering project machinery is also carried out Intelligent Recognition and can effectively give the alarm before cable is destroyed, thus auxiliary power company takes measures on customs clearance to prevent cable to be destroyed in time.
Underground cable is often subject to the destruction of engineering machinery, and does not have the special recognition methods for engineering machinery at present, therefore, the present invention proposes a kind of method carrying out recognitive engineering machinery by extracting engineering machinery sound characteristic.
Summary of the invention
For above-mentioned deficiency of the prior art, the invention provides a kind of engineering machinery recognition methods based on sound linear prediction residue error.
The technical solution used in the present invention is:
Based on an engineering machinery recognition methods for sound linear prediction residue error, comprise the steps:
The engineering machinery LPCC coefficient extracted under utilizing ecotopia builds engineering machinery sound characteristic fingerprint base;
The real-time voice signal of engineering machinery extracts LPCC coefficient and sets up;
According to the sound source belonging to it, demarcation of class mark is carried out to the LPCC coefficient of the every frame signal of engineering machinery;
Utilize support vector machine learning algorithm, adopt the LPCC engineering machinery sound characteristic fingerprint base of the every frame signal of engineering machinery after demarcating to carry out model training study, set up disaggregated model;
Real-time for engineering machinery voice signal is extracted the LPCC feature of every frame signal, be input in the disaggregated model that support vector machine learning algorithm trained, according to the classifier functions built, calculate the output valve of this frame measured signal, namely the fitting function that aforementioned training data obtains is adopted, the function calculating measured signal LPCC feature exports, and in the vector obtained, the position of greastest element is the type of engineering apparatus.
The engineering machinery LPCC coefficient extracted under utilizing ecotopia builds engineering machinery sound characteristic fingerprint base and comprises:
Engineering machinery voice signal extracts LPCC coefficient and sets up;
Background noise is extracted LPCC coefficient and is set up;
Engineering machinery voice signal extracts the foundation of LPCC coefficient and comprises:
Under engineering machinery is placed in ecotopia, star up engineering machinery;
The voice signal that voice collection device gathering project machinery sends;
The pre-service of engineering machinery voice signal;
Framing is carried out to engineering machinery voice signal;
LPCC coefficient is extracted to the every frame engineering machinery voice signal after framing.
The pre-service of engineering machinery voice signal comprises: engineering machinery voice signal carries out end-point detection, windowing, framing.
Engineering machinery voice signal end-point detection is realized by zero-crossing rate.
Hamming window is adopted to carry out framing to engineering machinery voice signal.
Background noise is extracted the foundation of LPCC coefficient and is comprised:
Background noise is gathered at the construction field (site) under true environment;
Background noise pre-service;
Framing is carried out to background noise;
LPCC coefficient is extracted to the every frame background noise after framing.
Background noise pre-service comprises: background noise carries out end-point detection, windowing, framing.
Background noise end-point detection is realized by zero-crossing rate.
Hamming window is adopted to carry out framing to background noise.
The real-time voice signal of engineering machinery extracts the foundation of LPCC coefficient and comprises:
Voice collection device gathering project machinery real-time audio signal;
The real-time voice signal pre-service of engineering machinery;
Framing is carried out to the real-time voice signal of engineering machinery;
LPCC coefficient is extracted to the real-time voice signal of the every frame engineering machinery after framing.
The pre-service of engineering machinery real-time voice signal comprises: the real-time voice signal of engineering machinery carries out end-point detection, windowing, framing.
The real-time voice signal end-point detection of engineering machinery is realized by zero-crossing rate.
Hamming window is adopted to carry out framing to the real-time voice signal of engineering machinery.
The voice signal sent when described background noise is the sound of yellow sand car, the voice of people and site works machine work.
Described engineering machinery comprises: electric pick, cutting machine, excavator and mechanical percussion hammer.
The described distance gathered apart from device and engineering machinery, comprise closely, in, three kinds of distances far away.
Described gather apart from device and electric pick, cutting machine, excavator be closely 10 meters; Middle distance is 30 meters; Be 50 meters at a distance.
Described collection is closely 50 meters apart from device and mechanical percussion hammer; Middle distance is 100 meters; Be 150 meters at a distance.
The every frame of the voice signal that engineering machinery sends chooses 1024 sampled points, and it is 512 sampled points that frame moves.
Described voice collection device is cross acoustic array.
The time interval that described cross acoustic array gathers sound is 0.5 second.
The extracting method of described LPCC feature is:
First try to achieve the LPC value of every frame signal, and then try to achieve its LPCC value;
Adopt Levinson-Durbin Algorithm for Solving LPC value;
LPC is linear predictor coefficient, the value of the present or following voice signal sampled point is predicted by the value of the voice signal sampled point in past, namely the value s (n) of voice signal of n-th can be predicted by the linear combination of the value of front p point, as shown in Equation 1:
s(n)≈a
1s(n-1)+a
2s(n-2)+…+a
ps(n-p)(1)
Wherein a
1, a
2... a
pbe linear predictor coefficient (LPC), p is LPC exponent number, gets 16 here
Levinson-Durbin algorithm, by making mean square deviation between the sampled value of actual audio signal and linear prediction sampled value (formula (2)) minimum, can obtain p rank linear predictor coefficient a
1, a
2... a
p.
After linear predictor coefficient obtains, sonification system transfer function model can be obtained:
LPCC is LPC expression in the cepstral domain, and cepstrum coefficient utilizes Homomorphic processing exactly, takes the logarithm after asking transform to voice signal, then transform of negating just can obtain cepstrum coefficient.After acquisition linear predictor coefficient, obtain system model, if shock response is c (n), is defined can be obtained by cepstrum:
(3) are substituted into (4), and to z
-1ask partial derivative, obtain (5):
Make (5) the right and left corresponding coefficient equal, obtain (6):
Wherein, c
1, c
2... c
nbe LPCC value, n is LPCC exponent number, gets n=18 here.
Described engineering machinery sound characteristic fingerprint base method for building up is as follows:
To the 18 rank LPCC values that every frame signal extracts, add row as class mark above, label ' 0 ' represents hand-held electric pick, ' 1 ' and represents noise, ' 2 ' and represent cutting machine, ' 3 ' and represent mechanical percussion hammer, and ' 4 ' represents excavator.Thus form the proper vector on 19 rank;
With libsvm as sorter, choose Radial basis kernel function (RBF) as sorter kernel function; RBF has two parameters: penalty factor c and parameter gamma, can choose numerical value by the grid search function opti_svm_coeff of libsvm; Thus make the RBF kernel function chosen can effectively by data-mapping to higher dimensional space, and convert the engineering instrument signal of all kinds of demarcation to different high dimensional signal by class mark in higher dimensional space;
Training process uses Svmtrain function, comprises four parameters: proper vector, by the above-mentioned labelled LPCC value extracted; Kernel function type, selects RBF kernel function; RBF kernel functional parameter c and gamma, uses grid data service to determine; Can obtain the variable of a model by name after calling svmtrain, this variable save training gained model information, being got off by this variable save identifies for next step;
The svmtest function of the LPCC value libsvm every frame signal obtained carries out intelligent classification;
Svmtest has three parameters, first is class mark, is used for testing discrimination, here get the first row of LPCC value, second is proper vector, namely stores the variable of LPCC value, 3rd is Matching Model, is exactly the variable model that above-mentioned steps training process obtains;
Call rreturn value that svmtest obtains to classify exactly acquired results, i.e. class mark, thus the device type producing this sound can be determined.
The beneficial effect of hinge structure of the present invention:
The present invention can improve recognition effect and reduce alert rate by mistake, effectively can identify hand-held electric pick, cutting machine, mechanical percussion hammer, excavator four kind equipment.The many employing vibrations of the anti-outside destroy of existing underground cable or fiber-optic signal detect, the present invention adopts sound, detecting distance is farther, to hand-held electric pick, cutting machine, the effective detecting distance of excavator can reach 60m, the effective detecting distance of mechanical percussion hammer can reach 180m, and discrimination can reach more than 80%.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet according to the engineering machinery recognition methods based on sound linear prediction residue error of the present invention;
Fig. 2 is that described engineering machinery sound characteristic fingerprint base is set up and intelligent classification process flow diagram;
Embodiment
Referring to drawings and Examples, the present invention will be described in detail:
Accompanying drawing 1 is known, and a kind of engineering machinery recognition methods based on sound linear prediction residue error, comprises the steps:
The engineering machinery LPCC coefficient extracted under utilizing ecotopia builds engineering machinery sound characteristic fingerprint base;
The real-time voice signal of engineering machinery extracts LPCC coefficient and sets up;
According to the sound source belonging to it, demarcation of class mark is carried out to the LPCC coefficient of the every frame signal of engineering machinery;
Utilize support vector machine learning algorithm, adopt the LPCC engineering machinery sound characteristic fingerprint base of the every frame signal of engineering machinery after demarcating to carry out model training study, set up disaggregated model.
(using function svmtrain); Engineering instrument signal by all kinds of demarcation converts different high dimensional signals to by class mark, Data classification is converted to Function Fitting problem.Specific as follows, convert the five class training datas demarcated (" 0 " represents hand-held electric pick, " 1 " expression noise signal, " 2 " expression cutting machine, " 3 " expression mechanical percussion hammer, and " 4 " represent excavator) to five dimensional vectors, wherein
" 0 " represents that hand-held electric pick converts to (1-1-1-1-1)
" 1 " represents that noise signal converts to (-11-1-1-1)
" 2 " represent that cutting machine converts to (-1-11-1-1)
" 3 " represent that mechanical percussion hammer converts to (-1-1-11-1)
" 4 " represent that excavator converts to (-1-1-1-11)
The structure of such sorter just converts Function Fitting to, and function output is five dimensional vectors.
Real-time for engineering machinery voice signal is extracted the LPCC feature of every frame signal, be input in the disaggregated model that support vector machine learning algorithm trained (i.e. the rreturn value of svmtrain function), according to the classifier functions built, calculate the output valve of this frame measured signal, namely the fitting function that aforementioned training data obtains is adopted, the function calculating measured signal LPCC feature exports (output is five dimensional vectors), and in the vector obtained, the position of greastest element is the type of engineering apparatus.
The engineering machinery LPCC coefficient extracted under utilizing ecotopia builds engineering machinery sound characteristic fingerprint base and comprises:
Engineering machinery voice signal extracts LPCC coefficient and sets up;
Background noise is extracted LPCC coefficient and is set up;
Engineering machinery voice signal extracts the foundation of LPCC coefficient and comprises:
Under engineering machinery is placed in ecotopia, star up engineering machinery;
The voice signal that voice collection device gathering project machinery sends;
The pre-service of engineering machinery voice signal;
Framing is carried out to engineering machinery voice signal;
LPCC coefficient is extracted to the every frame engineering machinery voice signal after framing.
The pre-service of engineering machinery voice signal comprises: engineering machinery voice signal carries out end-point detection, windowing, framing.
Engineering machinery voice signal end-point detection is realized by zero-crossing rate.
Hamming window is adopted to carry out framing to engineering machinery voice signal.
Background noise is extracted the foundation of LPCC coefficient and is comprised:
Background noise is gathered at the construction field (site) under true environment;
Background noise pre-service;
Framing is carried out to background noise;
LPCC coefficient is extracted to the every frame background noise after framing.
Background noise pre-service comprises: background noise carries out end-point detection, windowing, framing.
Background noise end-point detection is realized by zero-crossing rate.
Hamming window is adopted to carry out framing to background noise.
The real-time voice signal of engineering machinery extracts the foundation of LPCC coefficient and comprises:
Voice collection device gathering project machinery real-time audio signal;
The real-time voice signal pre-service of engineering machinery;
Framing is carried out to the real-time voice signal of engineering machinery;
LPCC coefficient is extracted to the real-time voice signal of the every frame engineering machinery after framing.
The pre-service of engineering machinery real-time voice signal comprises: the real-time voice signal of engineering machinery carries out end-point detection, windowing, framing.
The real-time voice signal end-point detection of engineering machinery is realized by zero-crossing rate.
Hamming window is adopted to carry out framing to the real-time voice signal of engineering machinery.
The voice signal sent when described background noise is the sound of yellow sand car, the voice of people and site works machine work.
Described noise signal, gets yellow sand car sound, each 1000 frames of building site ground unrest, because these two kinds of sound source frequencies of occurrences are high, large on recognizer impact.
Described engineering machinery comprises: electric pick, cutting machine, excavator and mechanical percussion hammer.
The described distance gathered apart from device and engineering machinery, comprise closely, in, three kinds of distances far away.
Described gather apart from device and electric pick, cutting machine, excavator be closely 10 meters; Middle distance is 30 meters; Be 50 meters at a distance.
Described collection is closely 50 meters apart from device and mechanical percussion hammer; Middle distance is 100 meters; Be 150 meters at a distance.
The every frame of the voice signal that engineering machinery sends chooses 1024 sampled points, and it is 512 sampled points that frame moves.
Described voice collection device is cross acoustic array.
The time interval that described cross acoustic array gathers sound is 0.5 second.
The extracting method of described LPCC feature is:
First try to achieve LPC (linear predictor coefficient) value of every frame signal, and then try to achieve its LPCC value;
Adopt Levinson-Durbin Algorithm for Solving LPC value;
LPC is linear predictor coefficient, the value of the present or following voice signal sampled point is predicted by the value of the voice signal sampled point in past, namely the value s (n) of voice signal of n-th can be predicted by the linear combination of the value of front p point, as shown in Equation 1:
s(n)≈a
1s(n-1)+a
2s(n-2)+…+a
ps(n-p)(1)
Wherein a
1, a
2... a
pbe linear predictor coefficient (LPC), p is LPC exponent number, gets 16 here
Levinson-Durbin algorithm, by making mean square deviation between the sampled value of actual audio signal and linear prediction sampled value (formula (2)) minimum, can obtain p rank linear predictor coefficient a
1, a
2... a
p.
After linear predictor coefficient obtains, sonification system transfer function model can be obtained:
LPCC is LPC expression in the cepstral domain, and cepstrum coefficient utilizes Homomorphic processing exactly, takes the logarithm after asking transform to voice signal, then transform of negating just can obtain cepstrum coefficient.After acquisition linear predictor coefficient, obtain system model, if shock response is c (n), is defined can be obtained by cepstrum:
(3) are substituted into (4), and to z
-1ask partial derivative, obtain (5):
Make (5) the right and left corresponding coefficient equal, obtain (6):
Wherein, c
1, c
2... c
nbe LPCC value, n is LPCC exponent number, gets n=18 here.
Described engineering machinery sound characteristic fingerprint base method for building up is as follows:
To the 18 rank LPCC values that every frame signal extracts, add row as class mark above, label ' 0 ' represents hand-held electric pick, ' 1 ' and represents noise, ' 2 ' and represent cutting machine, ' 3 ' and represent mechanical percussion hammer, and ' 4 ' represents excavator.Thus form the proper vector on 19 rank;
With libsvm as sorter, choose Radial basis kernel function (RBF) as sorter kernel function; RBF has two parameters: penalty factor c and parameter gamma, can select optimum value by the grid search function opti_svm_coeff of libsvm; Thus make the RBF kernel function chosen can effectively by data-mapping to higher dimensional space, and convert the engineering instrument signal of all kinds of demarcation to different high dimensional signal by class mark in higher dimensional space, Data classification be converted to Function Fitting problem.
Specific as follows, convert the five class training datas demarcated (" 0 " represents hand-held electric pick, " 1 " expression noise signal, " 2 " expression cutting machine, " 3 " expression mechanical percussion hammer, and " 4 " represent excavator) to five dimensional vectors, wherein
" 0 " represents that hand-held electric pick converts to (1-1-1-1-1)
" 1 " represents that noise signal converts to (-11-1-1-1)
" 2 " represent that cutting machine converts to (-1-11-1-1)
" 3 " represent that mechanical percussion hammer converts to (-1-1-11-1)
" 4 " represent that excavator converts to (-1-1-1-11)
The structure of such sorter just converts Function Fitting to, and function output is five dimensional vectors.
Training process uses Svmtrain function, comprises four parameters: proper vector, by the above-mentioned labelled LPCC value extracted; Kernel function type, selects RBF kernel function; RBF kernel functional parameter c and gamma, uses grid data service to determine; Can obtain the variable of a model by name after calling svmtrain, this variable save training gained model information, being got off by this variable save identifies for next step;
The svmtest function of the LPCC value libsvm every frame signal obtained carries out intelligent classification;
Svmtest has three parameters, first is class mark, is used for testing discrimination, here get the first row of LPCC value, second is proper vector, namely stores the variable of LPCC value, 3rd is Matching Model, is exactly the variable model that above-mentioned steps training process obtains;
Call rreturn value that svmtest obtains to classify exactly acquired results, i.e. class mark, thus the device type producing this sound can be determined.
Engineering machinery sound characteristic fingerprint base is set up.Owing to there is no the special sound characteristic fingerprint base for engineering machinery at present, therefore need to set up an engineering machinery sound characteristic fingerprint base.Concrete grammar is as follows:
To the 18 rank LPCC values that pickaxe, cutting machine, mechanical percussion hammer, excavator, the every frame signal of noise in technical scheme extract, wherein to four kinds of engineering machinery sound, respectively get 1000 frames near, in, different distance signal 18 rank far away LPCC value, for noise signal, get yellow sand car sound, each 1000 frames of building site ground unrest.
Row are added as class mark above, thus form the vector on 19 rank, shown by great many of experiments, when each distance gets 1000 frame signal, can show the feature of voice signal under this distance, add data again and there will be redundancy, promote unhelpful to discrimination, therefore to the voice signal under four kinds of engineering machinery different distance, respectively get 1000 frames near, in, different distance signal 18 rank far away LPCC value, for noise signal, get yellow sand car sound, each 1000 frames of building site ground unrest, try to achieve its 18 rank LPCC value, these 18 exponent numbers are arranged as class mark according to front increase by, hand-held electric pick is represented with label " 0 ", " 1 " represents noise signal, " 2 " represent cutting machine, " 3 " represent mechanical percussion hammer, " 4 " represent excavator.This just forms the matrix that 12000 row 19 arrange, and wherein first be classified as class mark, the second to ten nine are classified as corresponding LPCC feature.
Proper vector 3*4*1000 being amounted to 12000 group of 19 rank uses SVM to train, wherein 3 represent near, in, three kinds of distances far away; 4 represent four kinds of engineering machinery; 1000 is 1000 frame signals that each distance is chosen.
Use the svmtrain function of svmlib, svmtrain has four parameters: proper vector, the vector that 12000 row 19 obtained after namely extracting feature arrange; Kernel function type, chooses RBF kernel function; The value of the parameter c of RBF function and gamma, penalty factor c and parameter gamma selects optimal value by the grid search function opti_svm_coeff of libsvm; Returning of svmtrain remains training gained model information, the sorter obtained after training exactly, i.e. the sound characteristic fingerprint base of four kinds of engineering machinery.
SVM adopts libsvm, and it can realize many classification features, and wherein function svmtrain is used for training, and svmtest function is used for identifying.Svm has four kinds of conventional kernel functions: linear kernel (LinearKernel), polynomial kernel (PolynomialKernel), Radial basis kernel function (RadialBasisFunction), Sigmoid core (SigmoidKernel), apply these four kinds of different IPs functions to be trained to power and to be respectively: 82.72%, 62.49%, 88.25%, 81.91%, Radial basis kernel function effect is best, therefore, Radial basis kernel function is chosen here as sorter kernel function.The value of its penalty factor c and parameter gamma selects optimal value by the grid search function opti_svm_coeff of libsvm.Svmtrain function return value saves training gained model information, this value is preserved and identifies for next step.
The identification of engineering machinery sound.The svmtest function of the LPCC value libsvm every frame signal obtained carries out intelligent classification, svmtest has three parameters, first is class mark, be used for testing discrimination, second is proper vector, namely store the variable of LPCC value, the 3rd is Matching Model, is exactly the rreturn value of above-mentioned steps training process svmtrain function.Call rreturn value that svmtest obtains to classify exactly acquired results, i.e. class mark, thus the device type producing this sound can be determined.
SVM adopts libsvm, and it can realize many classification features, and wherein function svmtrain is used for training, and svmtest function is used for identifying.Svm has four kinds of conventional kernel functions: linear kernel (LinearKernel), polynomial kernel (PolynomialKernel), Radial basis kernel function (RadialBasisFunction), Sigmoid core (SigmoidKernel), apply these four kinds of different IPs functions to be trained to power and to be respectively: 82.72%, 62.49%, 88.25%, 81.91%, Radial basis kernel function effect is best.
Choose Radial basis kernel function as sorter kernel function; The value of its penalty factor c and parameter gamma selects optimal value by the grid search function opti_svm_coeff of libsvm;
Can obtain the variable of a model by name after the training of Svmtrain function, this variable save training gained model information, gets off this variable save for next step identification.
The svmtest function of the LPCC value libsvm every frame signal obtained carries out intelligent classification;
Svmtest has three parameters, first is class mark, is used for testing discrimination, here get the first row of LPCC value, second is proper vector, namely stores the variable of LPCC value, 3rd is Matching Model, is exactly the variable model that above-mentioned steps training process obtains;
Call rreturn value that svmtest obtains to classify exactly acquired results, i.e. class mark, thus the device type producing this sound can be determined.
The present invention can improve recognition effect and reduce alert rate by mistake, effectively can identify hand-held electric pick, cutting machine, mechanical percussion hammer, excavator four kind equipment.The many employing vibrations of the anti-outside destroy of existing underground cable or fiber-optic signal detect, the present invention adopts sound, detecting distance is farther, to hand-held electric pick, cutting machine, the effective detecting distance of excavator can reach 60m, the effective detecting distance of mechanical percussion hammer can reach 180m, and discrimination can reach more than 80%.
The above is only preferred embodiment of the present invention, not does any pro forma restriction to structure of the present invention.Every above embodiment is done according to technical spirit of the present invention any simple modification, equivalent variations and modification, all belong within the scope of technical scheme of the present invention.
Claims (10)
1., based on an engineering machinery recognition methods for sound linear prediction residue error, comprise the steps:
The engineering machinery LPCC coefficient extracted under utilizing ecotopia builds engineering machinery sound characteristic fingerprint base;
The real-time voice signal of engineering machinery extracts LPCC coefficient and sets up;
According to the sound source belonging to it, demarcation of class mark is carried out to the LPCC coefficient of the every frame signal of engineering machinery;
Utilize support vector machine learning algorithm, adopt the LPCC engineering machinery sound characteristic fingerprint base of the every frame signal of engineering machinery after demarcating to carry out model training study, set up disaggregated model;
Real-time for engineering machinery voice signal is extracted the LPCC feature of every frame signal, be input in the disaggregated model that support vector machine learning algorithm trained, according to the classifier functions built, calculate the output valve of this frame measured signal, namely the fitting function that aforementioned training data obtains is adopted, the function calculating measured signal LPCC feature exports, and in the vector obtained, the position of greastest element is the type of engineering apparatus.
2., according to claim 1 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described utilize ecotopia under the engineering machinery LPCC coefficient that extracts build engineering machinery sound characteristic fingerprint base and comprise:
Engineering machinery voice signal extracts LPCC coefficient and sets up;
Background noise is extracted LPCC coefficient and is set up.
3. according to claim 2 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described engineering machinery voice signal extracts the foundation of LPCC coefficient and comprises:
Under engineering machinery is placed in ecotopia, star up engineering machinery;
The voice signal that voice collection device gathering project machinery sends;
The pre-service of engineering machinery voice signal;
Framing is carried out to engineering machinery voice signal;
LPCC coefficient is extracted to the every frame engineering machinery voice signal after framing.
4. according to claim 3 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: the pre-service of described engineering machinery voice signal comprises: engineering machinery voice signal carries out end-point detection, windowing, framing.
5. according to claim 4 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described engineering machinery voice signal end-point detection is realized by zero-crossing rate.
6. according to claim 4 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described employing Hamming window carries out framing to engineering machinery voice signal.
7. according to claim 2 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described background noise is extracted the foundation of LPCC coefficient and comprised:
Background noise is gathered at the construction field (site) under true environment;
Background noise pre-service;
Framing is carried out to background noise;
LPCC coefficient is extracted to the every frame background noise after framing.
8. according to claim 7 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described background noise pre-service comprises: background noise carries out end-point detection, windowing, framing.
9. according to claim 8 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: described background noise end-point detection is realized by zero-crossing rate.
10. according to claim 8 based on the engineering machinery recognition methods of sound linear prediction residue error, it is characterized in that: adopt Hamming window to carry out framing to background noise.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510637964.1A CN105118516A (en) | 2015-09-29 | 2015-09-29 | Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510637964.1A CN105118516A (en) | 2015-09-29 | 2015-09-29 | Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105118516A true CN105118516A (en) | 2015-12-02 |
Family
ID=54666479
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510637964.1A Pending CN105118516A (en) | 2015-09-29 | 2015-09-29 | Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105118516A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106100771A (en) * | 2016-06-16 | 2016-11-09 | 苏州科达科技股份有限公司 | A kind of two-way time delay detecting method and device |
| CN106297770A (en) * | 2016-08-04 | 2017-01-04 | 杭州电子科技大学 | The natural environment sound identification method extracted based on time-frequency domain statistical nature |
| CN108847252A (en) * | 2018-04-26 | 2018-11-20 | 杭州电子科技大学 | Acoustic feature extraction method based on acoustical signal sound spectrograph grain distribution |
| CN113689673A (en) * | 2021-08-18 | 2021-11-23 | 广东电网有限责任公司 | Cable monitoring protection method, device, system and medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101540048A (en) * | 2009-04-21 | 2009-09-23 | 北京航空航天大学 | Image quality evaluating method based on support vector machine |
| CN101599126A (en) * | 2009-04-22 | 2009-12-09 | 哈尔滨工业大学 | Utilize the support vector machine classifier of overall intercommunication weighting |
| CN102005135A (en) * | 2010-12-09 | 2011-04-06 | 上海海事大学 | Genetic algorithm-based support vector regression shipping traffic flow prediction method |
| CN102024180A (en) * | 2010-12-23 | 2011-04-20 | 浙江大学 | Support vector machine-based parameter-adaptive motion prediction method |
| CN202048980U (en) * | 2011-05-03 | 2011-11-23 | 北京中矿信电科技股份有限公司 | Audio-based mechanical failure detection device |
| CN103136587A (en) * | 2013-03-07 | 2013-06-05 | 武汉大学 | Power distribution network operating state classification recognition method based on support vector machine |
| CN103150578A (en) * | 2013-04-09 | 2013-06-12 | 山东师范大学 | Training method of SVM (Support Vector Machine) classifier based on semi-supervised learning |
| CN103226948A (en) * | 2013-04-22 | 2013-07-31 | 山东师范大学 | Audio scene recognition method based on acoustic events |
| CN103310789A (en) * | 2013-05-08 | 2013-09-18 | 北京大学深圳研究生院 | Sound event recognition method based on optimized parallel model combination |
| CN103456301A (en) * | 2012-05-28 | 2013-12-18 | 中兴通讯股份有限公司 | Ambient sound based scene recognition method and device and mobile terminal |
-
2015
- 2015-09-29 CN CN201510637964.1A patent/CN105118516A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101540048A (en) * | 2009-04-21 | 2009-09-23 | 北京航空航天大学 | Image quality evaluating method based on support vector machine |
| CN101599126A (en) * | 2009-04-22 | 2009-12-09 | 哈尔滨工业大学 | Utilize the support vector machine classifier of overall intercommunication weighting |
| CN102005135A (en) * | 2010-12-09 | 2011-04-06 | 上海海事大学 | Genetic algorithm-based support vector regression shipping traffic flow prediction method |
| CN102024180A (en) * | 2010-12-23 | 2011-04-20 | 浙江大学 | Support vector machine-based parameter-adaptive motion prediction method |
| CN202048980U (en) * | 2011-05-03 | 2011-11-23 | 北京中矿信电科技股份有限公司 | Audio-based mechanical failure detection device |
| CN103456301A (en) * | 2012-05-28 | 2013-12-18 | 中兴通讯股份有限公司 | Ambient sound based scene recognition method and device and mobile terminal |
| CN103136587A (en) * | 2013-03-07 | 2013-06-05 | 武汉大学 | Power distribution network operating state classification recognition method based on support vector machine |
| CN103150578A (en) * | 2013-04-09 | 2013-06-12 | 山东师范大学 | Training method of SVM (Support Vector Machine) classifier based on semi-supervised learning |
| CN103226948A (en) * | 2013-04-22 | 2013-07-31 | 山东师范大学 | Audio scene recognition method based on acoustic events |
| CN103310789A (en) * | 2013-05-08 | 2013-09-18 | 北京大学深圳研究生院 | Sound event recognition method based on optimized parallel model combination |
Non-Patent Citations (5)
| Title |
|---|
| PEIPEI SHEN ET AL: ""Automatic Speech Emotion Recognition Using Support Vector Machine"", 《2011 INTERNATIONAL CONFERENCE ON ELECTRONIC & MECHANICAL ENGINEERING AND INFORMATION TECHNOLOGY》 * |
| SANWEI YANG ET AL: ""Acoustics recognition of construction equipments based on LPCC features and SVM"", 《PROCEEDINGS OF THE 34TH CHINESE CONTROL CONFERENCE》 * |
| 唐发明 等: ""基于支持向量机的仓储害虫声音识别"", 《华中科技大学学报（自然科学版）》 * |
| 白静 张雪英: ""基于支持向量机的抗噪语音识别"", 《太原理工大学学报》 * |
| 蔡铁 朱杰: ""一种基于类融合向量的支持向量机及其在语音识别系统中的应用"", 《电路与系统》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106100771A (en) * | 2016-06-16 | 2016-11-09 | 苏州科达科技股份有限公司 | A kind of two-way time delay detecting method and device |
| CN106100771B (en) * | 2016-06-16 | 2018-07-03 | 苏州科达科技股份有限公司 | A kind of two-way time delay detecting method and device |
| CN106297770A (en) * | 2016-08-04 | 2017-01-04 | 杭州电子科技大学 | The natural environment sound identification method extracted based on time-frequency domain statistical nature |
| CN106297770B (en) * | 2016-08-04 | 2019-11-22 | 杭州电子科技大学 | The natural environment sound identification method extracted based on time-frequency domain statistical nature |
| CN108847252A (en) * | 2018-04-26 | 2018-11-20 | 杭州电子科技大学 | Acoustic feature extraction method based on acoustical signal sound spectrograph grain distribution |
| CN113689673A (en) * | 2021-08-18 | 2021-11-23 | 广东电网有限责任公司 | Cable monitoring protection method, device, system and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105022835B (en) | A kind of intelligent perception big data public safety recognition methods and system | |
| CN102027338B (en) | Signal judgment method, signal judgment apparatus, and signal judgment system | |
| CN103971678B (en) | Keyword spotting method and apparatus | |
| CN103310789B (en) | A kind of sound event recognition method of the parallel model combination based on improving | |
| CN107256635B (en) | Vehicle identification method based on distributed optical fiber sensing in intelligent traffic | |
| CN102522082B (en) | Recognizing and locating method for abnormal sound in public places | |
| Cao et al. | Excavation equipment classification based on improved MFCC features and ELM | |
| CN104916289A (en) | Quick acoustic event detection method under vehicle-driving noise environment | |
| CN106707099A (en) | Monitoring and locating method based on abnormal electricity consumption detection module | |
| CN105118516A (en) | Identification method of engineering machinery based on sound linear prediction cepstrum coefficients (LPCC) | |
| CN103968933A (en) | Fuzzy domain characteristics based optical fiber vibration signal identifying method | |
| CN101364408A (en) | Sound image combined monitoring method and system | |
| CN104167207A (en) | Equipment sound identification method based on transformer substation patrol inspection robot | |
| CN102163427A (en) | Method for detecting audio exceptional event based on environmental model | |
| CN105095624A (en) | Method for identifying optical fibre sensing vibration signal | |
| CN105640748A (en) | Vibration blind-guiding clothing | |
| CN105678343A (en) | Adaptive-weighted-group-sparse-representation-based diagnosis method for noise abnormity of hydroelectric generating set | |
| CN106992011A (en) | Engineering machinery sound identification method based on MF PLPCC features | |
| CN109000876A (en) | SNS optical fiber based on autocoder deep learning impacts recognition methods | |
| CN109473119A (en) | A kind of acoustic target event-monitoring method | |
| CN104008644A (en) | Urban road traffic noise measurement method based on gradient descent | |
| CN106202977A (en) | A kind of low frequency oscillation mode based on blind source separation algorithm analyzes method | |
| CN104103280A (en) | Dynamic time warping algorithm based voice activity detection method and device | |
| CN112985574A (en) | High-precision classification identification method for optical fiber distributed acoustic sensing signals based on model fusion | |
| CN103559289A (en) | Language-irrelevant keyword search method and system |
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 |
Application publication date: 20151202 |
|
| RJ01 | Rejection of invention patent application after publication |