CN104732970B - A kind of ship-radiated noise recognition methods based on comprehensive characteristics - Google Patents
A kind of ship-radiated noise recognition methods based on comprehensive characteristics Download PDFInfo
- Publication number
- CN104732970B CN104732970B CN201310713525.5A CN201310713525A CN104732970B CN 104732970 B CN104732970 B CN 104732970B CN 201310713525 A CN201310713525 A CN 201310713525A CN 104732970 B CN104732970 B CN 104732970B
- Authority
- CN
- China
- Prior art keywords
- power
- frequency
- feature
- ship
- radiated noise
- 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.)
- Active
Links
Abstract
The ship-radiated noise recognition methods based on comprehensive characteristics that the present invention relates to a kind of, comprising: construct known ships radiated noise signal training set;In constructed training set, the feature of ship-radiated noise is extracted;Wherein, the feature includes aural signature and statistical nature, the aural signature includes spectrum flux, highest spectrum peak and Center of Time Gravity, the statistical nature include spectrum peak, mean power, the frequency of spectrum peak, average frequency, frequency bandwidth, frequency kurtosis, the standard deviation of power, power pitch, in frequency the gradient of power, power kurtosis, in frequency power kurtosis;Using the feature of ship-radiated noise as the feature of target identification training classifier;Read signal to be identified;Extract the feature of signal to be identified;Extracted feature is input to classifier, to do Classification and Identification to signal to be identified.
Description
Technical field
The present invention relates to Noise Identification field, in particular to a kind of ship-radiated noise identification side based on comprehensive characteristics
Method.
Background technique
The fast development of the underwater sound and electronic information technology, so that carrying out target classification identification using ship-radiated noise becomes
One important research topic, it is an important component of underwater information system, is constantly subjected to many to its research
The very big concern of scholar, engineers and technicians and military service.
Ship-radiated noise be it is extremely complex, it is closely related with marine environment and ship itself motion state.They with
Different sea areas, the different time and constantly change.These all to the type of sonic propagation and ship-radiated noise generate compared with
It is big to influence, bigger difficulty is brought to the Classification and Identification of ship-radiated noise.However, different naval vessels are due to Ship Structure, ship
The difference of the immanent structures such as type, propeller size, the number of blade, power device, the noise radiated are also different.Therefore, Ke Yitong
The classification to ship-radiated noise is crossed to identify the type on naval vessel.
High-order statistic, wavelet transformation in modern signal processing technology, fractal geometry, artificial neural network, information
Fusion and data mining scheduling theory and method have been widely used in ship-radiated noise identification.But because of the office of various methods
It is sex-limited, the correctness of target identification is affected, recognition result in the actual environment is unsatisfactory.The mankind make an uproar for naval vessel radiation
The identification of sound can achieve very high level, be at present qualitative and empirical, shortage to the aural signature description of the mankind still
The description and quantitative analysis of image.
The auditory system of the mankind has very excellent natural scale and robustness to the decomposition and processing of voice signal, right
Sound source characteristic has selectivity well, and has good adaptability to ambient noise.It therefore, can be from human auditory system feature
It sets out, research is suitable for the new feature extractive technique of underwater sound signal, finds the effective feature volume in human ear subjectivity sense of hearing amount, reaches
Improve the purpose of Underwater Targets Recognition rate.The quantitative or pictute of aural signature is to probing into what the mankind identified ship noise
Mechanism and Underwater Targets Recognition all have great importance.
Still lack the method that ship-radiated noise is identified according to human auditory system mechanism in the prior art.
Summary of the invention
It is an object of the invention to overcome to lack in the prior art to identify ship-radiated noise according to human auditory system mechanism
The defect of method, to provide a kind of ship-radiated noise recognition methods based on comprehensive characteristics.
To achieve the goals above, the ship-radiated noise recognition methods based on comprehensive characteristics that the present invention provides a kind of,
Include:
Ships radiated noise signal training set known to step 1), building;
In step 2, the training set constructed by step 1), the feature of ship-radiated noise is extracted;Wherein, the feature
Including aural signature and statistical nature, the aural signature includes spectrum flux, highest spectrum peak and Center of Time Gravity, the statistics
Feature includes spectrum peak, mean power, the frequency of spectrum peak, average frequency, frequency bandwidth, frequency kurtosis, power
Standard deviation, power pitch, in frequency the gradient of power, power kurtosis, in frequency power kurtosis;
Step 3) is classified using the feature of the obtained ship-radiated noise of step 2 as the training of the feature of target identification
Device;
Step 4) reads signal to be identified;
Step 5), the feature for extracting signal to be identified, the feature include aural signature and statistical nature, and the sense of hearing is special
Sign includes that spectrum flux, highest spectrum peak and Center of Time Gravity, the statistical nature include spectrum peak, mean power, power
The frequency of spectrum peak, average frequency, frequency bandwidth, frequency kurtosis, the standard deviation of power, power pitch, the power in frequency
Gradient, power kurtosis, in frequency power kurtosis;
Step 6), the classifier that feature extracted in step 5) is input to step 3) training, thus to be identified
Signal does Classification and Identification.
It further include using the feature of ship-radiated noise as target in the step 3) in above-mentioned technical proposal
Before knowing another characteristic training classifier, further includes being carried out to the value of the feature regular, convert them between 0-1
Value.
In above-mentioned technical proposal, in the step 2) or step 4), extracting the spectrum flux includes:
Step 2-1-1-1), normalized is done to time domain waveform first;
Step 2-1-1-2), determine the points of every frame sampling sequence, be discrete FFT after adding hamming window to every frame sequence, take
Mould square obtains discrete power spectrum;
Step 2-1-1-3), calculate according to the discrete power of adjacent two frame spectrum the spectral vectors of adjacent two frame, and then calculate
Pearson correlation coefficients between adjacent two frames spectral vectors;Its calculation formula is as follows:
Wherein, X, Y are the spectral vectors of adjacent two frames n dimension, Xi, YiI-th of value in respectively two vectors,Respectively
For the mean value of two vectors;
Step 2-1-1-4), according to the Pearson correlation coefficients between each adjacent two frames spectral vectors, calculate noise letter
Number spectrum flux;Its calculation formula is as follows:
Wherein, M is the quantity of the be divided into time frame of noise signal;rk,rk-1Be two neighboring time frame spectral vectors it
Between Pearson correlation coefficients.
In above-mentioned technical proposal, in the step 2) or step 4), extracting highest spectrum peak includes:
Step 2-1-2-1), time domain waveform is normalized first;
Step 2-1-2-2), time domain sequences are added and are discrete FFT after hamming window, modulus square obtains the discrete function of signal
Rate spectrum, and be decibel by the unit conversion of power;
Step 2-1-2-3), finally find out the corresponding frequency value F of noise power spectrum peak-peakm。
In above-mentioned technical proposal, in the step 2) or step 4), extracting the Center of Time Gravity includes:
Step 2-1-3-1), time domain waveform is normalized first;
Step 2-1-3-2), find out the signal energy at each moment;
Step 2-1-3-3), calculate the time domain center of gravity of noise signal, calculation formula is as follows:
Wherein, E (t) is energy value corresponding to t moment on time-domain diagram.
In above-mentioned technical proposal, the classifier uses SVM support vector machines.
The present invention has the advantages that
The invention comprehensively utilizes the subjective characteristics of simulation human auditory and based on the objective characteristics of statistics, pass through three spies
Sign quantitatively states the mankind for the subjective feeling of sound, and passes through the objective information that statistical nature reflects sound, comprehensive utilization
These two types of features, and feature combination is optimized, recognition performance can be effectively improved, so that identification process is closer to the mankind
Identification process.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention;
Fig. 2 (a) is the schematic diagram that different types of ship-radiated noise is identified using spectrum flux;
Fig. 2 (b) is the schematic diagram that different types of ship-radiated noise is identified using Center of Time Gravity;
Fig. 3 (a) shows the distribution situation on the eight class naval vessels under st2-st5 X-Y scheme;
Fig. 3 (b) shows the distribution situation on the eight class naval vessels under st3-st5 X-Y scheme.
Specific embodiment
Now in conjunction with attached drawing, the invention will be further described.
Ship-radiated noise recognition methods of the invention is realized based on passive sonar.Passive sonar passively receives naval vessel
After the signal of target generates in equal water radiated noise and underwater sound equipment transmitting, for these noises and signal, with reference to Fig. 1, originally
The method of invention realizes the identification to ship-radiated noise using following steps.
Ships radiated noise signal training set known to step 1), building.
Pretreatment is normalized etc. to the ships radiated noise signal of known class first, it then will be after normalization
The ship-radiated noise of known class is configured to training set.
In step 2, the training set constructed by step 1), the feature of ship-radiated noise is extracted.
In the present invention, the feature of extracted ship-radiated noise includes two major classes, and respectively aural signature and statistics is special
Sign.The extraction process of this two major classes feature is illustrated separately below.
Step 2-1), extract aural signature.
Feature extraction is exactly to extract some parameters that can characterize target physical properties, to obtain most characterizing target signature
Substantive characteristics.The aural signature to be extracted in this step includes three classes: spectrum flux, highest spectrum peak and Center of Time Gravity.
Step 2-1-1), extract spectrum flux
The spectrum flux of voice signal describes human ear at any time to the impression degree of sound, i.e., frequency mistake on a timeline
Cross characteristic.It simulates the non-linear resolution characteristic of human auditory system, reflects a large amount of and important characteristic information of voice signal, by force
Subjective tone color is affected strongly, closer to the auditory perception of people.Pearson correlation coefficients reflect two linear variable displacement correlations
Degree.According to the calculation method of spectrum flux, the extraction for composing flux characteristics is carried out by frame, and specific extraction step is as follows:
Step 2-1-1-1), time domain waveform (i.e. noise signal) is pre-processed first, i.e., place is normalized to waveform
Reason;
Step 2-1-1-2), determine every frame sampling sequence points (such as sample rate be 8000Hz when, 600 points of every frame,
300 points of overlappings), every frame sequence is added and is 2048 points of discrete FFT after hamming window, modulus square obtains discrete power spectrum;
Step 2-1-1-3), calculate according to the discrete power of adjacent two frame spectrum the spectral vectors of adjacent two frame, and then calculate
Pearson correlation coefficients between adjacent two frames spectral vectors;Its calculation formula is as follows:
Wherein, X, Y are the spectral vectors of adjacent two frames n dimension, Xi, YiI-th of value in respectively two vectors,Respectively
For the mean value of two vectors.
Step 2-1-1-4), according to the Pearson correlation coefficients between each adjacent two frames spectral vectors, calculate noise letter
Number spectrum flux;Its calculation formula is as follows:
Wherein, M is the quantity of the be divided into time frame of noise signal;rk,rk-1Be two neighboring time frame spectral vectors it
Between Pearson correlation coefficients.
Step 2-1-2), extract highest spectrum peak.
Power spectrum signal reflects signal energy random distribution situation, and noise power composes the corresponding frequency of peak-peak
It is worth the maximum frequency values of representation signal energy.
The extraction step of highest spectral peak value tag is as follows:
Step 2-1-2-1), first to time domain waveform pre-process, i.e., waveform is normalized;
Step 2-1-2-2), time domain sequences are added and are 2048 points of discrete FFT after hamming window, modulus square obtains signal
Discrete power spectrum, and be decibel by the unit conversion of power;
Step 2-1-2-3), finally find out the corresponding frequency value F of noise power spectrum peak-peakm。
Step 2-1-3), extraction time center of gravity.
The Center of Time Gravity of noise signal i.e. the center of gravity of temporal envelope, reflect the time domain specification of signal, specific
Extraction step is as follows:
Step 2-1-3-1), first to time domain waveform pre-process, i.e., waveform is normalized;
Step 2-1-3-2), find out the signal energy at each moment;
Step 2-1-3-3), calculate the time domain center of gravity of noise signal, calculation formula is as follows:
Wherein, E (t) is energy value corresponding to t moment on time-domain diagram.
Step 2-2), extract statistical nature.
For the ease of the extraction of ship-radiated noise statistical nature, noise signal is first divided into T time frame, and calculate
The short Fourier transform of 2F point (STFT) of every frame.In this way, signal will be represented by the F spectral coefficient by T frame.In following public affairs
In formula, pt,fIndicate power of the signal at moment t frequency f.
The statistical nature to be extracted includes:
Spectrum peak, i.e., the maximum value of all time frame general powers:
st2=M=max (pt)
Mean power, i.e., the average value of all time frame general powers:
The frequency of spectrum peak, i.e. frequency corresponding to spectrum peak:
Average frequency, the i.e. average frequency of signal, wherein P indicates total power signal:
RMS bandwidth, i.e. frequency bandwidth:
Frequency kurtosis, i.e. average frequency kurtosis:
Power SD, the i.e. standard deviation of power, wherein F is the number of spectral coefficient, and T is the number of time frame:
Power pitch, the i.e. gradient of power:
Power pitch, i.e., the gradient of power in frequency:
Power kurtosis, the i.e. kurtosis of power:
Power kurtosis F, i.e., the kurtosis of power in frequency:
Above-mentioned 11 statistical natures are the features that can most reflect ship-radiated noise, thus by above-mentioned statistical nature obtain as
Lower 11 dimension statistical nature (st2,st3,st5,st6,st7,st9,st14,st17,st19,st20,st22).
Step 2-3), by step 2-1) three obtained aural signature and step 2-2) obtained statistical nature construction is such as
Down for indicating the vector of the feature of ship-radiated noise:
V={ SF, Fm,TC,st2,st3,st5,st6,st7,st9,st14,st17,st19,st20,st22}。
The feature of the obtained ship-radiated noise of previous step is inputted classification by step 3)
Device, to train classifier.
As a kind of preferred implementation, a certain feature value is excessive and flood the contribution of other features in order to prevent,
Will ship-radiated noise feature input classifier before to characteristic value carry out it is regular, convert them into the value between 0-1.
SVM support vector machines can be used in classifier in this step, and SVM support vector machines is in the case of linear separability
What optimal separating hyper plane developed, mechanism and treatment process are equivalent to the former input space transforming to a new feature
Space, and optimum linearity classification plane is solved in new space.
Wherein, the principal mode of kernel function has following four kinds: linear kernel function, Polynomial kernel function, radial kernel function and
Sigmoid kernel function, the application can also use other classes in other embodiments using the Polynomial kernel function of 6 ranks
The kernel function of type.
The classification function that SVM training data is formed has lower surface properties: SVM is one group using supporting vector as the non-thread of parameter
The linear combination of property function, therefore the expression formula of classification function is only related with the quantity of supporting vector, and independently of the dimension in space
Degree.When handling the classification of high-dimensional input spaces, SVM is especially effective.
Step 4) reads signal to be identified;
Step 5) extracts signal characteristic to be identified;The feature includes that 3 aural signatures above-mentioned and 11 statistics are special
Sign.
Feature extracted in step 5) is input to trained classifier by step 6), thus to letter to be identified
Number do Classification and Identification.
Below by experiments have shown that the method for the present invention effect.The radiated noise on eight class naval vessels is done in an experiment
Detection, as shown in Figure 2, wherein ' asterisk ' represents active Sonar signal;' square ' represents 10,000 tons naval vessel, and ' circle ' represents
Freighter;' plus sige ' represents another large ship;' diamond shape ' represents small-scale fishing vessel;' right triangle ' represents medium-sized surface vessel;' under
Triangle ' represent seanoise;' five-pointed star ' represents small-sized aircraft.It can be seen that spectrum flux (Fig. 2 in acoustic feature
(a)) and Center of Time Gravity (Fig. 2 (b)) can be used for distinguishing different types of ship-radiated noise, in contrast, compose flux to not
The distinction of the ship-radiated noise of same type is better than Center of Time Gravity.
Fig. 3 (a) shows the distribution situation on the eight class naval vessels under st2-st5 X-Y scheme, and Fig. 3 (b) is shown in st3-st5
The distribution situation on eight class naval vessels under X-Y scheme.The two figures illustrate that the separability of these statistical natures is all preferable.
It should be noted last that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting.Although ginseng
It is described the invention in detail according to embodiment, those skilled in the art should understand that, to technical side of the invention
Case is modified or replaced equivalently, and without departure from the spirit and scope of technical solution of the present invention, should all be covered in the present invention
Scope of the claims in.
Claims (5)
1. a kind of ship-radiated noise recognition methods based on comprehensive characteristics, comprising:
Step 1) constructs known ships radiated noise signal training set;
In step 2), the training set constructed by step 1), the feature of ship-radiated noise is extracted;Wherein, the feature includes
Aural signature and statistical nature, the aural signature include spectrum flux, highest spectrum peak and Center of Time Gravity, the statistical nature
Including spectrum peak, mean power, the frequency of spectrum peak, average frequency, frequency bandwidth, frequency kurtosis, power mark
Quasi- poor, power pitch, in frequency the gradient of power, power kurtosis, in frequency power kurtosis;
Step 3) trains classifier using the feature of the obtained ship-radiated noise of step 2) as the feature of target identification;
Step 4) reads signal to be identified;
Step 5), the feature for extracting signal to be identified, the feature includes aural signature and statistical nature, the aural signature packet
Including spectrum flux, highest spectrum peak and Center of Time Gravity, the statistical nature includes spectrum peak, mean power, power spectral peak
The frequency of value, average frequency, frequency bandwidth, frequency kurtosis, the standard deviation of power, power pitch, in frequency the gradient of power,
Power kurtosis, in frequency power kurtosis;
Step 6), the classifier that feature extracted in step 5) is input to step 3) training, thus to signal to be identified
Do Classification and Identification;
In the step 2) or step 4), extracting the spectrum flux includes:
Step 2-1-1-1), normalized is done to time domain waveform first;
Step 2-1-1-2), determine the points of every frame sampling sequence, be discrete FFT after adding hamming window to every frame sequence, modulus
Square obtain discrete power spectrum;
Step 2-1-1-3), calculate according to the discrete power of adjacent two frame spectrum the spectral vectors of adjacent two frame, and then calculate adjacent
Pearson correlation coefficients between two frame spectral vectors;Its calculation formula is as follows:
Wherein, X, Y are the spectral vectors of adjacent two frames n dimension, Xi, YiI-th of value in respectively two vectors,Respectively two
The mean value of a vector;
Step 2-1-1-4), according to the Pearson correlation coefficients between each adjacent two frames spectral vectors, calculate noise signal
Compose flux;Its calculation formula is as follows:
Wherein, M is the quantity of the be divided into time frame of noise signal;rk,rk-1It is between the spectral vectors of two neighboring time frame
Pearson correlation coefficients.
2. the ship-radiated noise recognition methods according to claim 1 based on comprehensive characteristics, which is characterized in that described
Step 3) in, further include using the feature of ship-radiated noise as target identification feature training classifier before, also wrap
The value progress included to the feature is regular, converts them into the value between 0-1.
3. the ship-radiated noise recognition methods according to claim 1 or 2 based on comprehensive characteristics, which is characterized in that
In the step 2) or step 5), extracting highest spectrum peak includes:
Step 2-1-2-1), time domain waveform is normalized first;
Step 2-1-2-2), time domain sequences are added and are discrete FFT after hamming window, modulus square obtains the discrete power of signal
Spectrum, and be decibel by the unit conversion of power;
Step 2-1-2-3), finally find out the corresponding frequency value F of noise power spectrum peak-peakm。
4. the ship-radiated noise recognition methods according to claim 1 or 2 based on comprehensive characteristics, which is characterized in that
In the step 2) or step 5), extracting the Center of Time Gravity includes:
Step 2-1-3-1), time domain waveform is normalized first;
Step 2-1-3-2), find out the signal energy at each moment;
Step 2-1-3-3), calculate the time domain center of gravity of noise signal, calculation formula is as follows:
Wherein, E (t) is energy value corresponding to t moment on time-domain diagram.
5. the ship-radiated noise recognition methods according to claim 1 or 2 based on comprehensive characteristics, which is characterized in that institute
Classifier is stated using SVM support vector machines.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310713525.5A CN104732970B (en) | 2013-12-20 | 2013-12-20 | A kind of ship-radiated noise recognition methods based on comprehensive characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310713525.5A CN104732970B (en) | 2013-12-20 | 2013-12-20 | A kind of ship-radiated noise recognition methods based on comprehensive characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104732970A CN104732970A (en) | 2015-06-24 |
CN104732970B true CN104732970B (en) | 2018-12-04 |
Family
ID=53456810
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310713525.5A Active CN104732970B (en) | 2013-12-20 | 2013-12-20 | A kind of ship-radiated noise recognition methods based on comprehensive characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104732970B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106483520B (en) * | 2016-09-27 | 2018-12-07 | 哈尔滨工程大学 | A kind of Ship Radiated-Noise index of modulation estimation method |
CN106895905B (en) * | 2016-12-21 | 2019-07-19 | 西北工业大学 | A kind of ship-radiated noise detection method |
CN107103282B (en) * | 2017-03-22 | 2020-05-26 | 中国科学院遥感与数字地球研究所 | Ultra-high resolution synthetic aperture radar image classification method |
CN108694346B (en) * | 2017-04-05 | 2021-12-07 | 中国科学院声学研究所 | Ship radiation noise signal identification method based on two-stage CNN |
EP3579020B1 (en) * | 2018-06-05 | 2021-03-31 | Elmos Semiconductor SE | Method for recognition of an obstacle with the aid of reflected ultrasonic waves |
CN110390949B (en) * | 2019-07-22 | 2021-06-15 | 苏州大学 | Underwater sound target intelligent identification method based on big data |
CN111157095B (en) * | 2020-01-17 | 2022-03-01 | 上海索辰信息科技股份有限公司 | Automatic frequency extraction method of noise source |
CN111488801A (en) * | 2020-03-16 | 2020-08-04 | 天津大学 | Ship classification method based on vibration noise identification |
CN111553207B (en) * | 2020-04-14 | 2022-09-06 | 哈尔滨工程大学 | Statistical distribution-based ship radiation noise characteristic recombination method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5734793A (en) * | 1994-09-07 | 1998-03-31 | Motorola Inc. | System for recognizing spoken sounds from continuous speech and method of using same |
CN101685446A (en) * | 2008-09-25 | 2010-03-31 | 索尼(中国)有限公司 | Device and method for analyzing audio data |
CN102498514A (en) * | 2009-08-04 | 2012-06-13 | 诺基亚公司 | Method and apparatus for audio signal classification |
CN102760444A (en) * | 2012-04-25 | 2012-10-31 | 清华大学 | Support vector machine based classification method of base-band time-domain voice-frequency signal |
CN103000172A (en) * | 2011-09-09 | 2013-03-27 | 中兴通讯股份有限公司 | Signal classification method and device |
-
2013
- 2013-12-20 CN CN201310713525.5A patent/CN104732970B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5734793A (en) * | 1994-09-07 | 1998-03-31 | Motorola Inc. | System for recognizing spoken sounds from continuous speech and method of using same |
CN101685446A (en) * | 2008-09-25 | 2010-03-31 | 索尼(中国)有限公司 | Device and method for analyzing audio data |
CN102498514A (en) * | 2009-08-04 | 2012-06-13 | 诺基亚公司 | Method and apparatus for audio signal classification |
CN103000172A (en) * | 2011-09-09 | 2013-03-27 | 中兴通讯股份有限公司 | Signal classification method and device |
CN102760444A (en) * | 2012-04-25 | 2012-10-31 | 清华大学 | Support vector machine based classification method of base-band time-domain voice-frequency signal |
Also Published As
Publication number | Publication date |
---|---|
CN104732970A (en) | 2015-06-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104732970B (en) | A kind of ship-radiated noise recognition methods based on comprehensive characteristics | |
CN109800700A (en) | A kind of underwater sound signal target classification identification method based on deep learning | |
Gupta et al. | Feature extraction using MFCC | |
CN108711436B (en) | Speaker verification system replay attack detection method based on high frequency and bottleneck characteristics | |
CN104732978B (en) | The relevant method for distinguishing speek person of text based on combined depth study | |
Mielke et al. | A method for automated individual, species and call type recognition in free-ranging animals | |
CN110245608A (en) | A kind of Underwater targets recognition based on semi-tensor product neural network | |
CN108875592A (en) | A kind of convolutional neural networks optimization method based on attention | |
WO2018223727A1 (en) | Voiceprint recognition method, apparatus and device, and medium | |
CN110148408A (en) | A kind of Chinese speech recognition method based on depth residual error | |
CN103854660B (en) | A kind of four Mike's sound enhancement methods based on independent component analysis | |
CN103021406A (en) | Robust speech emotion recognition method based on compressive sensing | |
CN111724770B (en) | Audio keyword identification method for generating confrontation network based on deep convolution | |
CN110349588A (en) | A kind of LSTM network method for recognizing sound-groove of word-based insertion | |
Sun et al. | Underwater single-channel acoustic signal multitarget recognition using convolutional neural networks | |
Weißkirchen et al. | Recognition of emotional speech with convolutional neural networks by means of spectral estimates | |
CN115240651A (en) | Land-air communication speaker role identification method and device based on feature fusion | |
CN104156628A (en) | Ship radiation signal recognition method based on multi-kernel learning and discriminant analysis | |
CN115393933A (en) | Video face emotion recognition method based on frame attention mechanism | |
CN114722812A (en) | Method and system for analyzing vulnerability of multi-mode deep learning model | |
Wu et al. | The DKU-LENOVO Systems for the INTERSPEECH 2019 Computational Paralinguistic Challenge. | |
Chakravarty et al. | Data augmentation and hybrid feature amalgamation to detect audio deep fake attacks | |
CN113076878B (en) | Constitution identification method based on attention mechanism convolution network structure | |
CN108573711A (en) | A kind of single microphone speech separating method based on NMF algorithms | |
CN113111786A (en) | Underwater target identification method based on small sample training image convolutional network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |