CN109920448A - A kind of identifying system and method for automatic driving vehicle traffic environment special type sound - Google Patents
A kind of identifying system and method for automatic driving vehicle traffic environment special type sound Download PDFInfo
- Publication number
- CN109920448A CN109920448A CN201910141320.1A CN201910141320A CN109920448A CN 109920448 A CN109920448 A CN 109920448A CN 201910141320 A CN201910141320 A CN 201910141320A CN 109920448 A CN109920448 A CN 109920448A
- Authority
- CN
- China
- Prior art keywords
- sound
- signal
- unit
- formula
- time
- 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
Abstract
The present invention provides the identifying systems and method of a kind of automatic driving vehicle traffic environment special type sound, including information acquisition unit, information process unit, acoustic recognition unit, storage unit and control unit, information acquisition unit obtains the real time environment voice signal in traffic environment, and is passed along information process unit；Information process unit carries out abnormal sound detection to the voice signal of acquisition, after the signal for detecting suspected target special type sound, carries out feature extraction to echo signal, and be transferred to acoustic recognition unit；Acoustic recognition unit identifies sound type according to the sound characteristic of acquisition, and result is conveyed to vehicle-mounted ECU, realizes the perception of traffic sounds environment.The present invention can be according to the extraordinary sound detected, and identifies extraordinary sound type, supplements the information blind spot of traditional camera and radar；More fully traffic information is provided for autonomous driving vehicle.
Description
Technical field
The invention belongs to automatic Pilot environment sensing field, especially a kind of automatic driving vehicle traffic environment special type sound
Identifying system and method.
Background technique
In recent years, with the development of artificial intelligence and autonomous driving vehicle, the safety and reliability of automatic Pilot is more next
More more paid close attention to.Automated driving system mainly includes three parts: perception, decision, control.Wherein context aware systems
It is to realize the basis of automatic Pilot, and Multi-sensor Fusion is to realize the inexorable trend of automatic Pilot environment sensing.Currently, existing
Onboard sensor mainly include laser radar, millimetre-wave radar, camera etc..But in city traffic road condition, due to height
The barriers such as building block, it is difficult to find the traffic incident that " do not see ".
Summary of the invention
For effective solution problem above, the invention proposes a kind of automatic driving vehicle traffic environment special type sound
Identifying system and method according to the extraordinary sound detected, and identify extraordinary sound type, provide more for autonomous driving vehicle
Comprehensive traffic information.The extraordinary sound of this patent meaning specifically includes that the energy such as police whistle sound, whistle sound, impact sound and shriek
Show the sound of traffic events.
To achieve the above object, the present invention is as follows using more specific technical solution:
A kind of identifying system of automatic driving vehicle traffic environment special type sound, which is characterized in that including information collection list
Member, information process unit, acoustic recognition unit, storage unit and control unit,
Voice signal is converted to digital signal for acquiring traffic environment voice signal by the information acquisition unit,
It is transferred to information process unit；
The input terminal of the information process unit is connected with information acquisition unit, output end is connected with acoustic recognition unit,
For handling the traffic environment audio digital signal transmitted, the abnormal sound whether having in ambient sound digital signal detected
Sound signal, and the characteristic parameter of abnormal sound segment is extracted, characteristic parameter is transferred to acoustic recognition unit；
The output end of acoustic recognition unit is connected with vehicle ECU, for identification the motion state of sound source and including which kind of out
Extraordinary voice signal, and recognition result is conveyed to vehicle ECU, so that vehicle makes corresponding measures to keep clear；The storage
Unit includes sound status memory block and sound characteristic parameter memory block, is respectively used to storage sample audio motion state and sample
Sound characteristic parameter；
The acoustic recognition unit, storage unit are connected with control unit respectively, and control unit can call storage unit
The data of middle storage are for training and updating identification model.
Further, the storage unit can be automatically updated from internet obtains sound characteristic parameter, also can be from shifting
Dynamic storage dish obtains, for updating identification model.
Further, the information acquisition unit is the microphone array that roof is arranged in.
The recognition methods of the identifying system of automatic driving vehicle traffic environment special type sound, which is characterized in that including as follows
Step:
S1 information acquisition unit obtains the ambient sound in road, and is transformed into digital signal and is conveyed to signal processing list
Member；
S2 signal processing unit carries out framing to the digital signal information stream of ambient sound, adding window pre-processes；Number is believed
Number carry out end-point detection, detect whether with abnormal sound signal, if there is abnormal sound signal, then utilize blind source separating side
Method isolates abnormal sound signal；
S3 sound processing unit extracts the time and frequency domain characteristics of abnormal sound signal, scramble characteristic of field；
The method that S4 acoustic recognition unit voice recognition process uses two-stage classification,
First order identification: machine sound and voice are distinguished, and identifies the motion state of sound source；
Which kind of second level identification: identify as extraordinary voice signal.
Further, n microphone array of roof is located in the S1, obtains airborne sound in traffic environment
Sound signal, when system is opened, microphone array starts to acquire airborne voice signal in real-time traffic environment.
Further, end-point detection uses auto-relativity function method to utilize formula that is, by after signal framing in the S2
(1) short-time autocorrelation function of every frame data is sought；The otherness of the auto-correlation function of noise and extraordinary sound is recycled to detect
Extraordinary voice signal out；
In formula: R is the auto-correlation function of signal；I is frame number, indicates the i-th frame；K is retardation；L is frame length；Y is signal
Amplitude.
Further, the method for the blind source separating is: the process of sound mix is interpreted as m source sound by mixing
System receives n signal by n microphone, and the hybrid matrix A of sound mix system is found out using its corresponding relationship, recycles
Algorithm finds out separation system matrix B, so as to find out the abnormal sound signal isolated；
During the blind source separating, m source sound Si(t) after mixing, n signal X is received by n microphonej
(t), Xj(t) and Si(t) shown in relationship such as formula (2),
In above formula: m is the number of source sound；N is the number of microphone；T is the time；Xj(t) it is received for microphone array
The voice signal arrived；VjIt (t) is the engine noise and road noise in environment；SiIt (t) is source voice signal；ajiFor sound mix system
The hybrid parameter of system；
The voice signal Y isolatedi(t) and Xi(t) shown in relationship such as formula (3):
In above formula: m is the number of source sound；N is the number of microphone；T is the time；Xj(t) it is received for microphone array
The voice signal arrived；Yi(t) the source voice signal to isolate；bijFor the separation parameter of sound separation system.
The hybrid parameter a of hybrid system is found out using fast independent component analysis (FastICA) algorithmjiWith separation system
Separation parameter bij, source voice signal Y is found out using the corresponding relationship of formula (2) and formula (3)i(t), to complete voice signal
Separation.
Further, the extracting method of the time and frequency domain characteristics parameter is first to find out the time-frequency spectrum of abnormal sound, is then used
The maximum power spectral densities of each time point track time frequency signal, seek the number of peaks of characteristic parameter unit time；Then
The signal obtained to tracking is smoothed, and finds out characteristic parameter time-frequency derivative；
The specific method is as follows for the extraction of scramble characteristic of field:
(1) abnormal sound signal is set as S (t), the extraordinary sound of every frame is obtained after filtering, preemphasis, framing, adding window
Sound signal S (t), to S (t)) it does discrete Fourier transform (DFT) and obtains frequency domain signal X (ω), time domain is converted into frequency domain；It asks
The X (ω) obtained is indicated are as follows:
In above formula: t is the time；ω is frequency；X (ω) is frequency-region signal；X (t) is time-domain signal after framing；N is in Fu
Leaf transformation siding-to-siding block length；
(2) square of X (ω), i.e. energy spectrum are asked；By the filter group that is made of M filter to energy spectral filter,
The centre frequency of m-th of filter is f (m), m=1,2 ..., M；M-th of triangular filter transmission function are as follows:
In above formula, Hm(ω) is triangular filter transmission function,
(3) the energy logarithm superposition for calculating m-th of filter output is denoted as S (m), and expression formula is as follows:
In formula: S (m) is the energy logarithm of superposition；M is filter group number；X (ω) is frequency-region signal；Hm(ω) is three
Angle filter transfer function；
(4) mel cepstrum coefficients (MFCC) is obtained through discrete cosine transform (DCT) again:
In above formula: C (n) is mel cepstrum coefficients；S (m) is the energy logarithm of superposition；M is filter group number；L is plum
That cepstrum coefficient dimension.
Further, the dimension of mel cepstrum coefficients takes L=12 herein.
Further, the number M of filter and the number of critical band are close in filter group, and M takes 22-26；Filter is adopted
Use triangular filter.
Further, the voice recognition process is specific as follows:
(1) first order identifies, according to the time-frequency characteristics of extraction: passing through band limits, the number of peaks of unit time and list
Zero-crossing rate in the time of position tentatively identifies mechanic sound and voice；Furthermore according to the peak value of unit time in the time-frequency domain of extraction
Zero-crossing rate in amount and unit time, identifies the motion state of sound, method particularly includes: the number of peaks and unit of unit time
Zero-crossing rate in time becomes larger explanation and is proximate to state, and gradually becoming smaller explanation is far from state, and two kinds are basically unchanged explanation
It is to maintain relative static conditions；
(2) second level identifies, according to the MFCC parameter training identification model for the standardized special sound that test obtains, then will
The extraordinary voice signal of preliminary classification is input in trained SVM identification model, completes voice recognition.
Specifically, SVM identification model algorithm is as follows:
Svm classifier model is established, characteristic parameter C (n) the training SVM of the sample special type sound stored in storage unit is passed through
The characteristic parameter C ' (n) of the abnormal sound obtained in real time is carried out Classification and Identification as input by classifier, and specific SVM is non-linear
Algorithm is as follows:
Training sampleX in formulai∈R1It is i-th of input pattern, di∈ {+1, -1 } is that its is corresponding
Output expectation.Pass through Nonlinear Mapping φ (x): R1→Rn, former space input vector is mapped to N-dimensional feature space, it is optimal super
Plane:W is hyperplane normal in formula, is adjustable weight vector, and b is biasing, determines phase
For the optimal location of origin.Problem above can be converted to when meeting certain condition, and be minimized | | W | | the problem of, it can table
It is shown as following form:
In formula, C is constant, influences decision accuracy, and ξ is slack variable.Above-mentioned optimization is solved using Langrange multiplier to ask
Topic, expression formula:
In formula, αi, βiFor non-negative Lagrange coefficient, enable:
It obtains:
For solving high-dimensional feature space dot product φ (xi)T·φ(xj), without knowing Nonlinear Mapping, always find one
A kernel function that can satisfy Mercer condition original space, so that: K (xi, xj)=φ (xi)T·φ(xj), therefore can again by
Problem description are as follows:
Selection of kernel function Gauss kernel function herein:
σ is bandwidth in formula, controls radial effect range.
After sample data training, obtains the classifier that optimal hyperlane and relevant parameter are constituted and divided according to the following formula
Class:
To complete the Classification and Identification of extraordinary sound.
The present invention can help autonomous driving vehicle to obtain more fully traffic route information, supplement traditional camera and thunder
The information blind spot reached；It can not only identify that sound type can also identify the motion state of sound source using time and frequency domain characteristics parameter,
Such as: identify opposing stationary police and the police whistle sound in still relative motion；The present invention can solve the extraordinary sound of small number
Sound overlapping, the present invention divides two stage recognition voice signal using the time-frequency feature and scramble spectrum signature of sound, to improve sound
Accuracy of identification；Furthermore two-stage classification of the invention, which can guarantee to reach under conditions of certain accuracy of identification, reduces characteristic parameter
The purpose of dimension, to improve recognition speed；The present invention can also identify the motion state of sound source.
Detailed description of the invention
Fig. 1 is the identifying system structural schematic diagram of automatic driving vehicle traffic environment special type sound；
Fig. 2 is extraordinary voice recognition flow chart
Fig. 3 is blind source separating schematic diagram
Specific embodiment
With reference to the accompanying drawing and specific embodiment the present invention will be described in detail, but these embodiments are not intended to limit
The present invention, structure that those skilled in the art are made according to these embodiments, method or transformation functionally are equal
It is included within the scope of the present invention.
Fig. 1 show the schematic diagram of the identifying system of automatic driving vehicle traffic environment special type sound of the present invention.Institute
The identifying system of automatic driving vehicle traffic environment special type sound is stated, mainly includes information acquisition unit, information process unit, sound
Sound recognition unit, storage unit and control unit.
The information acquisition unit is the microphone array that roof is arranged in, and for acquiring traffic environment voice signal, is wrapped
Include police whistle sound, whistle sound, shriek and other sound；And voice signal is converted to digital signal, it is transferred to information processing list
Member.
The input terminal of the information process unit is connected with information acquisition unit, output end is connected with acoustic recognition unit.
Information process unit is to receive real-time environmental sound signal, handles the traffic environment audio digital signal transmitted, is detected
The abnormal sound signal whether having in ambient sound digital signal, and the characteristic parameter of abnormal sound segment is extracted, feature is joined
Number is transferred to acoustic recognition unit.
Acoustic recognition unit goes out the motion state of sound source and for identification including which kind of extraordinary voice signal.Voice recognition list
The output end of member is connected with vehicle ECU, and recognition result is conveyed to vehicle ECU, provides traffic sounds environmental information, and then make
It obtains vehicle and makes corresponding measures to keep clear.
The storage unit includes sound status memory block and sound characteristic parameter memory block, is respectively used to storage sound fortune
Dynamic state and sound characteristic parameter；The acoustic recognition unit, storage unit are connected with control unit respectively, and control unit can
Call the data stored in storage unit for training and updating identification model.
Storage unit can receive the target special type sound for the standard that the experiment in internet or mobile hard disk obtains
Characteristic parameter, and be supplied to acoustic recognition unit improves accuracy of identification in order to update identification model with this, increases identification sound
The quantity of sound.
The course of work of the identifying system of automatic driving vehicle traffic environment special type sound of the present invention are as follows: starting vehicle
Unlocking vehicle control system, microphone array obtains the real time environment voice signal in traffic environment, and is passed along at information
Manage unit；Information process unit carries out extraordinary sound detection to the voice signal of acquisition, detects suspected target special type sound
After signal, feature extraction is carried out to echo signal, and be transferred to acoustic recognition unit；Acoustic recognition unit is according to the sound of acquisition
Feature identifies sound type, and result is conveyed to vehicle-mounted ECU, realizes the perception of traffic sounds environment.
Specifically, the recognition methods of the automatic driving vehicle traffic environment special type sound, as shown in Fig. 2, including following
Step:
The microphone array of S1 roof obtains the ambient sound in road, and is transformed into digital signal and is conveyed to signal processing
Unit.
S2 signal processing unit carries out framing, adding window to the digital signal information stream of ambient sound.Endpoint is carried out to signal
Detection, is detected whether with abnormal sound signal, if there is abnormal sound signal, is then isolated using the method for blind source separating different
Normal voice signal.
The end-point detection is by taking auto-relativity function method end-point detection as an example.After signal framing, asked using following formula (1)
Take the short-time autocorrelation function of every frame data.The otherness of the auto-correlation function of noise and extraordinary sound is recycled to detect special type
Voice signal.
In above formula (1): R is the auto-correlation function of signal；I is frame number, indicates the i-th frame；K is retardation；L is frame length；Y is
Signal amplitude.
The energy threshold E obtained using the energy size of extraordinary voice signal, duration information and test0With time threshold
Value T0Compare to have discriminated whether that extraordinary voice signal overlapping then utilizes the method for blind source separating if there is extraordinary sound is overlapped
Find out the extraordinary voice signal of overlapping.
The principle of blind source separating is interpreted as m source sound by mixed stocker as shown in figure (3), by the process of sound mix
System, receives n signal by n microphone, and the hybrid matrix A of sound mix system is found out using its corresponding relationship, recycles and calculates
Method finds out separation system matrix B, so as to find out the abnormal sound signal isolated.
During the blind source separating, m source sound Si(t) after mixing, n signal X is received by n microphonej
(t), Xj(t) and Si(t) shown in relationship such as formula (2),
In above formula (2): m is the number of source sound；N is the number of microphone；T is the time；Xj(t) it is connect for microphone array
The voice signal received；VjIt (t) is the engine noise and road noise in environment；SiIt (t) is source voice signal；ajiFor sound mix
The hybrid parameter of system；
The voice signal Y isolatedi(t) and Xj(t) shown in relationship such as formula (3):
In above formula (3): m is the number of source sound；N is the number of microphone；T is the time；Xj(t) it is connect for microphone array
The voice signal received；Yi(t) the source voice signal to isolate；bijFor the separation parameter of sound separation system；
The hybrid parameter a of hybrid system is found out using fast independent component analysis (FastICA) algorithmjiWith separation system
Separation parameter bij, source voice signal Y is found out using the corresponding relationship of formula (2) and formula (3)i(t), to complete voice signal
Separation.
For S3 after Signal separator, or when judging to be overlapped without extraordinary voice signal, sound processing unit extracts abnormal sound
The time and frequency domain characteristics of sound signal, scramble characteristic of field.
Wherein, the extraction of time and frequency domain characteristics parameter is the time-frequency spectrum for first finding out abnormal sound, then with each time point
Maximum power spectral densities track time frequency signal, seek the number of peaks of characteristic parameter unit time；Then tracking is obtained
Signal is smoothed, and finds out characteristic parameter time-frequency derivative.
By taking police whistle sound as an example, first find out the time-frequency spectrum of police whistle sound, then with the maximum power spectral densities of each time point come
Time frequency signal is tracked, seeks the characteristic parameter of time frequency signal: the number of peaks of unit time, band limits and in the unit time
The characteristic parameters such as zero-crossing rate.
That scramble field parameter extracts is mel cepstrum coefficients C (n), and the specific method is as follows:
(1) original object special type voice signal or the signal separated by blind source separating, i.e. abnormal sound signal,
Be set as S (n), the extraordinary voice signal S (t) of every frame obtained after filtering, preemphasis, framing, adding window, to S (t)) do it is discrete
Fourier transformation (DFT) obtains frequency domain signal X (ω), and time domain is converted into frequency domain；The X (ω) acquired is indicated are as follows:
In above formula (4): t is the time；ω is frequency；X (ω) is frequency-region signal；X (t) is time-domain signal after framing；N is Fu
In leaf transformation siding-to-siding block length.
(2) square of X (ω), i.e. energy spectrum are asked.It is filtered by using the filter group of M filter, filter
Number and critical band number it is close, for the filter used for triangular filter, the centre frequency of m-th of filter is f
(m), m=1,2 ..., M, M usually take 22-26.M-th of triangular filter transmission function are as follows:
In above formula (5), Hm(ω) is triangular filter transmission function,
(3) the energy logarithm superposition for calculating m-th of filter output is denoted as S (m), and expression formula is as follows:
In formula (6): S (m) is the energy logarithm of superposition；M is filter group number；X (ω) is frequency-region signal；Hm(ω) is
Triangular filter transmission function；
(4) mel cepstrum coefficients (MFCC) is obtained through discrete cosine transform (DCT) again:
In formula (7): C (n) is mel cepstrum coefficients；S (m) is the energy logarithm of superposition；M is filter group number；L is plum
That cepstrum coefficient dimension.
After obtaining MFCC, determines the dimension of mel cepstrum coefficients under certain precision, preferably take L=12.
S4 voice recognition.The method that voice recognition process uses two-stage classification, first order identification, distinguishes machine sound, voice
With the relative motion state of sound source；Second level identification, identifies specific voice signal.
(1) first order identifies, according to the time-frequency characteristics of extraction: when band limits, the number of peaks and unit of unit time
Interior zero-crossing rate can tentatively identify mechanic sound and voice；Furthermore according to the peak value of unit time in the time-frequency domain of extraction
Zero-crossing rate in amount and unit time, can identify the motion state of sound, such as, the number of peaks and unit of unit time
Zero-crossing rate in time becomes larger explanation and is proximate to state, and gradually becoming smaller explanation is two kinds of principle state, is basically unchanged explanation
It is to maintain relative static conditions.The category identification process of machine sound and life is without Hidden Markov or support vector machines etc.
Identification model is by realizing with the selection algorithm of threshold value comparison.
(2) second level identifies, according to the MFCC parameter training identification model for the standardized special sound that test obtains, then will
The extraordinary voice signal of preliminary classification is input in trained SVM identification model, completes voice recognition.Specific SVM identification
Model algorithm is as follows:
Svm classifier model is established, characteristic parameter C (n) the training SVM of the sample special type sound stored in storage unit is passed through
The characteristic parameter C ' (n) of the abnormal sound obtained in real time is carried out Classification and Identification as input by classifier, and specific SVM is non-linear
Algorithm is as follows:
Training sampleX in formulai∈R1It is i-th of input pattern, di∈ {+1, -1 } is that its is corresponding
Output expectation.Pass through Nonlinear Mapping φ (x): R1→Rn, former space input vector is mapped to N-dimensional feature space, it is optimal super
Plane:W is hyperplane normal in formula, is adjustable weight vector, and b is biasing, determines phase
For the optimal location of origin.Problem above can be converted to when meeting certain condition, and be minimized | | W | | the problem of, it can table
It is shown as following form:
In formula, C is constant, influences decision accuracy, and ξ is slack variable.Above-mentioned optimization is solved using Langrange multiplier to ask
Topic, expression formula:
In formula, αi, βiFor non-negative Lagrange coefficient, enable:
It obtains:
For solving high-dimensional feature space dot product φ (xi)T·φ(xj), without knowing Nonlinear Mapping, always find one
A kernel function that can satisfy Mercer condition original space, so that: K (xi, xj)=φ (xi)T·φ(xj), therefore can again by
Problem description are as follows:
Selection of kernel function Gauss kernel function herein:
σ is bandwidth in formula, controls radial effect range.
After sample data training, obtains the classifier that optimal hyperlane and relevant parameter are constituted and divided according to the following formula
Class:
To complete the Classification and Identification of extraordinary sound.
Recognition result is conveyed to vehicle ECU by control module, sound, image and radar signal are integrated by ECU, further
It makes decisions and controls.
The embodiment is a preferred embodiment of the present invention, but present invention is not limited to the embodiments described above, not
In the case where substantive content of the invention, any conspicuous improvement that those skilled in the art can make, replacement
Or modification all belongs to the scope of protection of the present invention.
Claims (10)
1. a kind of identifying system of automatic driving vehicle traffic environment special type sound, which is characterized in that including information acquisition unit,
Information process unit, acoustic recognition unit, storage unit and control unit,
Voice signal is converted to digital signal for acquiring traffic environment voice signal by the information acquisition unit, is transmitted
To information process unit；
The input terminal of the information process unit is connected with information acquisition unit, output end is connected with acoustic recognition unit, is used for
The traffic environment audio digital signal transmitted is handled, the abnormal sound message whether having in ambient sound digital signal is detected
Number, and the characteristic parameter of abnormal sound segment is extracted, characteristic parameter is transferred to acoustic recognition unit；
Acoustic recognition unit goes out the motion state of sound source and for identification including which kind of extraordinary voice signal；
The storage unit includes sound status memory block and sound characteristic parameter memory block, is respectively used to storage sound movement shape
State and sound characteristic parameter；
The acoustic recognition unit, storage unit are connected with control unit respectively, and control unit can be called in storage unit and be deposited
The data of storage are for training and updating identification model.
2. the identifying system of automatic driving vehicle traffic environment special type sound according to claim 1, which is characterized in that described deposits
Storage unit can be automatically updated from internet obtains sound characteristic parameter, can also obtain from mobile storage disc, for updating identification
Model.
3. the identifying system of automatic driving vehicle traffic environment special type sound according to claim 1, which is characterized in that the information
Acquisition unit is the microphone array that roof is arranged in.
4. the recognition methods of the identifying system of claim 1 automatic driving vehicle traffic environment special type sound, which is characterized in that packet
Include following steps:
S1 information acquisition unit obtains the ambient sound in road, and is transformed into digital signal and is conveyed to signal processing unit；
S2 signal processing unit carries out framing to the digital signal information stream of ambient sound, adding window pre-processes；To digital signal into
Row end-point detection is detected whether with abnormal sound signal, if there is abnormal sound signal, then utilizes the method for blind source separating point
Separate out abnormal sound signal；
S3 sound processing unit extracts the time and frequency domain characteristics of abnormal sound signal, scramble characteristic of field；
The method that S4 acoustic recognition unit voice recognition process uses two-stage classification,
First order identification: machine sound and voice are distinguished, and identifies the motion state of sound source；
Which kind of second level identification: identify as extraordinary voice signal.
5. the recognition methods of automatic driving vehicle traffic environment special type sound according to claim 3, which is characterized in that institute
End-point detection seeks every frame data in short-term using formula (1) that is, by after signal framing using auto-relativity function method in the S2 stated
Auto-correlation function；The otherness of the auto-correlation function of noise and extraordinary sound is recycled to detect extraordinary voice signal；
In formula (1): R is the auto-correlation function of signal；I is frame number, indicates the i-th frame；K is retardation；L is frame length；Y is signal width
Value.
6. the recognition methods of automatic driving vehicle traffic environment special type sound according to claim 3, which is characterized in that
The method of the blind source separating is: the process of sound mix is interpreted as m source sound by hybrid system, by n wheat
Gram wind receives n signal, and the hybrid matrix A of sound mix system is found out using its corresponding relationship, recycles algorithm, finds out separation
Sytem matrix B, so as to find out the abnormal sound signal isolated；
During the blind source separating, m source sound Si(t) after mixing, n signal X is received by n microphonej(t), Xj
(t) and Si(t) shown in relationship such as formula (2),
In above formula (2): m is the number of source sound；N is the number of microphone；T is the time；Xj(t) it is received for microphone array
Voice signal；VjIt (t) is the engine noise and road noise in environment；SiIt (t) is source voice signal；ajiFor sound mix system
Hybrid parameter；
The voice signal Y isolatedi(t) and Xj(t) shown in relationship such as formula (3):
In above formula (3): m is the number of source sound；N is the number of microphone；T is the time；Xj(t) it is received for microphone array
Voice signal；Yi(t) the source voice signal to isolate；bijFor the separation parameter of sound separation system；
The hybrid parameter a of hybrid system is found out using fast independent component analysis (FastICA) algorithmjiWith the separation of separation system
Parameter bij, source voice signal Y is found out using the corresponding relationship of formula (2) and formula (3)i(t), to complete point of voice signal
From.
7. the recognition methods of automatic driving vehicle traffic environment special type sound according to claim 3, which is characterized in that institute
The extracting method for stating time and frequency domain characteristics parameter is first to find out the time-frequency spectrum of abnormal sound, then uses the maximum power of each time point
Spectrum density tracks time frequency signal, seeks the number of peaks of characteristic parameter unit time；Then the signal that tracking obtains is carried out
Smoothing processing finds out characteristic parameter time-frequency derivative；
The specific method is as follows for the extraction of scramble characteristic of field:
(1) abnormal sound signal is set as S (t), the extraordinary sound letter of every frame is obtained after filtering, preemphasis, framing, adding window
Number S (t), does discrete Fourier transform (DFT) to S (t) and obtains frequency domain signal X (ω), time domain is converted into frequency domain；The X acquired
(ω) is indicated are as follows:
In above formula (4): t is the time；ω is frequency；X (ω) is frequency-region signal；X (t) is time-domain signal after framing；N is Fourier
Convert siding-to-siding block length；
(2) square of X (ω), i.e. energy spectrum are asked；By the filter group that is made of M filter to energy spectral filter, m-th
The centre frequency of filter is f (m), m=1,2 ..., M；M-th of triangular filter transmission function are as follows:
In above formula (5), Hm(ω) is triangular filter transmission function,
(3) the energy logarithm superposition for calculating m-th of filter output is denoted as S (m), and expression formula is as follows:
In formula (6): S (m) is the energy logarithm of superposition；M is filter group number；X (ω) is frequency-region signal；Hm(ω) is triangle
Filter transfer function；
(4) mel cepstrum coefficients (MFCC) is obtained through discrete cosine transform (DCT) again:
In formula (7): C (n) is mel cepstrum coefficients；S (m) is the energy logarithm of superposition；M is filter group number；L falls for Meier
Spectral coefficient dimension.
8. the recognition methods of automatic driving vehicle traffic environment special type sound according to claim 7, which is characterized in that plum
You take L=12 by the dimension of cepstrum coefficient herein；The number M of filter and the number of critical band are close in filter group, and M takes 22-
26；Filter uses triangular filter.
9. the recognition methods of automatic driving vehicle traffic environment special type sound according to claim 3, which is characterized in that institute
It is specific as follows to state voice recognition process:
(1) first order identifies, according to the time-frequency characteristics of extraction: when by band limits, the number of peaks and unit of unit time
Interior zero-crossing rate tentatively identifies mechanic sound and voice；Furthermore according to the unit time in the time-frequency domain of extraction number of peaks and
Zero-crossing rate in unit time identifies the motion state of sound, specific recognition methods are as follows: the number of peaks and unit of unit time
Zero-crossing rate in time becomes larger explanation and is proximate to state, and gradually becoming smaller explanation is far from state, and two kinds are basically unchanged explanation
It is to maintain relative static conditions；
(2) second level identifies, then will be preliminary according to the MFCC parameter training identification model for the standardized special sound that test obtains
The extraordinary voice signal of classification is input in trained SVM identification model, completes voice recognition.
10. the recognition methods of automatic driving vehicle traffic environment special type sound according to claim 9, which is characterized in that
The SVM identification model algorithm is as follows:
Svm classifier model is established, characteristic parameter C (n) the training svm classifier of the sample special type sound stored in storage unit is passed through
The characteristic parameter C ' (n) of the abnormal sound obtained in real time is carried out Classification and Identification, specific SVM nonlinear algorithm as input by device
It is as follows:
Training sampleX in formulai∈R1It is i-th of input pattern, di∈ {+1, -1 } is its corresponding output
It is expected that；Pass through Nonlinear Mapping φ (x): R1→Rn, former space input vector is mapped to N-dimensional feature space, it is optimal super flat
Face:W is hyperplane normal in formula, is adjustable weight vector, and b is biasing, is determined opposite
In the optimal location of origin；Can be converted to when meeting certain condition, minimize | | W | | the problem of, it is expressed as form:
In formula, C is constant, influences decision accuracy, and ξ is slack variable；
Above-mentioned optimization problem is solved using Langrange multiplier, expression formula:
In formula, αi, βiFor non-negative Lagrange coefficient, enable:
It obtains:
For solving high-dimensional feature space dot product φ (xi)T·φ(xj), without knowing Nonlinear Mapping, always find an energy
Enough meet the kernel function in Mercer condition original space, so that: K (xi, xj)=φ (xi)T·φ(xj), therefore again describe problem
Are as follows:
The Selection of kernel function Gauss kernel function:
σ is bandwidth in formula, controls radial effect range.
After sample data training, obtains the classifier that optimal hyperlane and relevant parameter are constituted and classifies according to the following formula:
To complete the Classification and Identification of extraordinary sound.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910141320.1A CN109920448A (en) | 2019-02-26 | 2019-02-26 | A kind of identifying system and method for automatic driving vehicle traffic environment special type sound |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910141320.1A CN109920448A (en) | 2019-02-26 | 2019-02-26 | A kind of identifying system and method for automatic driving vehicle traffic environment special type sound |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109920448A true CN109920448A (en) | 2019-06-21 |
Family
ID=66962272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910141320.1A Pending CN109920448A (en) | 2019-02-26 | 2019-02-26 | A kind of identifying system and method for automatic driving vehicle traffic environment special type sound |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109920448A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675892A (en) * | 2019-09-24 | 2020-01-10 | 北京地平线机器人技术研发有限公司 | Multi-position voice separation method and device, storage medium and electronic equipment |
| CN110838305A (en) * | 2019-11-15 | 2020-02-25 | 中国汽车工程研究院股份有限公司 | Intelligent vehicle ADAS early warning test method and system based on voice recognition |
| CN111009261A (en) * | 2019-12-10 | 2020-04-14 | Oppo广东移动通信有限公司 | Arrival reminding method, device, terminal and storage medium |
| CN111409644A (en) * | 2020-04-10 | 2020-07-14 | 新石器慧通（北京）科技有限公司 | Autonomous vehicle and sound feedback adjusting method thereof |
| CN112233694A (en) * | 2020-10-10 | 2021-01-15 | 中国电子科技集团公司第三研究所 | Target identification method and device, storage medium and electronic equipment |
| CN113487909A (en) * | 2021-06-07 | 2021-10-08 | 惠州市德赛西威汽车电子股份有限公司 | Curve opposite vehicle early warning method and system based on voiceprint recognition |
| CN115116230A (en) * | 2022-07-26 | 2022-09-27 | 浪潮卓数大数据产业发展有限公司 | Traffic environment monitoring method, equipment and medium |
| CN117289208A (en) * | 2023-11-24 | 2023-12-26 | 北京瑞森新谱科技股份有限公司 | Sound source positioning method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104916289A (en) * | 2015-06-12 | 2015-09-16 | 哈尔滨工业大学 | Quick acoustic event detection method under vehicle-driving noise environment |
| CN107045784A (en) * | 2017-01-22 | 2017-08-15 | 苏州奇梦者网络科技有限公司 | A kind of electronic traffic police system |
| KR20180026243A (en) * | 2016-09-02 | 2018-03-12 | 엘지전자 주식회사 | Autonomous vehicle and control method thereof |
-
2019
- 2019-02-26 CN CN201910141320.1A patent/CN109920448A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104916289A (en) * | 2015-06-12 | 2015-09-16 | 哈尔滨工业大学 | Quick acoustic event detection method under vehicle-driving noise environment |
| KR20180026243A (en) * | 2016-09-02 | 2018-03-12 | 엘지전자 주식회사 | Autonomous vehicle and control method thereof |
| CN107045784A (en) * | 2017-01-22 | 2017-08-15 | 苏州奇梦者网络科技有限公司 | A kind of electronic traffic police system |
Non-Patent Citations (1)
| Title |
|---|
| 彭赛阳,王振华,朱元清: "盲源分离现状及发展", 《舰船电子对抗》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675892A (en) * | 2019-09-24 | 2020-01-10 | 北京地平线机器人技术研发有限公司 | Multi-position voice separation method and device, storage medium and electronic equipment |
| CN110838305A (en) * | 2019-11-15 | 2020-02-25 | 中国汽车工程研究院股份有限公司 | Intelligent vehicle ADAS early warning test method and system based on voice recognition |
| CN110838305B (en) * | 2019-11-15 | 2022-03-18 | 中国汽车工程研究院股份有限公司 | Intelligent vehicle ADAS early warning test method and system based on voice recognition |
| CN111009261A (en) * | 2019-12-10 | 2020-04-14 | Oppo广东移动通信有限公司 | Arrival reminding method, device, terminal and storage medium |
| CN111009261B (en) * | 2019-12-10 | 2022-11-15 | Oppo广东移动通信有限公司 | Arrival reminding method, device, terminal and storage medium |
| CN111409644A (en) * | 2020-04-10 | 2020-07-14 | 新石器慧通（北京）科技有限公司 | Autonomous vehicle and sound feedback adjusting method thereof |
| CN112233694A (en) * | 2020-10-10 | 2021-01-15 | 中国电子科技集团公司第三研究所 | Target identification method and device, storage medium and electronic equipment |
| CN112233694B (en) * | 2020-10-10 | 2024-03-05 | 中国电子科技集团公司第三研究所 | Target identification method and device, storage medium and electronic equipment |
| CN113487909A (en) * | 2021-06-07 | 2021-10-08 | 惠州市德赛西威汽车电子股份有限公司 | Curve opposite vehicle early warning method and system based on voiceprint recognition |
| CN115116230A (en) * | 2022-07-26 | 2022-09-27 | 浪潮卓数大数据产业发展有限公司 | Traffic environment monitoring method, equipment and medium |
| CN117289208A (en) * | 2023-11-24 | 2023-12-26 | 北京瑞森新谱科技股份有限公司 | Sound source positioning method and device |
| CN117289208B (en) * | 2023-11-24 | 2024-02-20 | 北京瑞森新谱科技股份有限公司 | Sound source positioning method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109920448A (en) | A kind of identifying system and method for automatic driving vehicle traffic environment special type sound | |
| Cao et al. | Convolutional neural network with second-order pooling for underwater target classification | |
| EP2907121B1 (en) | Real-time traffic detection | |
| CN112735473B (en) | Method and system for identifying unmanned aerial vehicle based on voice | |
| CN104916289A (en) | Quick acoustic event detection method under vehicle-driving noise environment | |
| CN111724770B (en) | Audio keyword identification method for generating confrontation network based on deep convolution | |
| Socoró et al. | Development of an Anomalous Noise Event Detection Algorithm for dynamic road traffic noise mapping | |
| CN109884591A (en) | A kind of multi-rotor unmanned aerial vehicle acoustical signal Enhancement Method based on microphone array | |
| CN108831506A (en) | Digital audio based on GMM-BIC distorts point detecting method and system | |
| CN110544482A (en) | single-channel voice separation system | |
| CN113640768B (en) | Low-resolution radar target identification method based on wavelet transformation | |
| CN103994820A (en) | Moving target identification method based on micro-aperture microphone array | |
| Djukanović et al. | Acoustic vehicle speed estimation from single sensor measurements | |
| Bulatović et al. | Mel-spectrogram features for acoustic vehicle detection and speed estimation | |
| US20210312912A1 (en) | Audio processing apparatus and method for audio scene classification | |
| Cheng et al. | Spectrogram-based classification on vehicles with modified loud exhausts via convolutional neural networks | |
| Zhao et al. | Event classification for living environment surveillance using audio sensor networks | |
| CN112541533A (en) | Modified vehicle identification method based on neural network and feature fusion | |
| Kaur et al. | Traffic state detection using smartphone based acoustic sensing | |
| CN115826042B (en) | Edge cloud combined distributed seismic data processing method and device | |
| Dadula et al. | Neural network classification for detecting abnormal events in a public transport vehicle | |
| CN104991245A (en) | Unmanned aerial vehicle early warning apparatus and early warning method thereof | |
| CN205003281U (en) | Unmanned vehicles early warning device | |
| Bhandarkar et al. | Vehicular mechanical condition determination and on road traffic density estimation using audio signals | |
| Olteanu et al. | Fusion of speech techniques for automatic environmental sound recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for 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: 20190621 |