CN111027453A - Automatic non-cooperative underwater target identification method based on Gaussian mixture model - Google Patents

Automatic non-cooperative underwater target identification method based on Gaussian mixture model Download PDF

Info

Publication number
CN111027453A
CN111027453A CN201911237928.0A CN201911237928A CN111027453A CN 111027453 A CN111027453 A CN 111027453A CN 201911237928 A CN201911237928 A CN 201911237928A CN 111027453 A CN111027453 A CN 111027453A
Authority
CN
China
Prior art keywords
data
gmm
feature vector
miss
threshold value
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.)
Granted
Application number
CN201911237928.0A
Other languages
Chinese (zh)
Other versions
CN111027453B (en
Inventor
曾向阳
乔彦
王海涛
杨爽
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN201911237928.0A priority Critical patent/CN111027453B/en
Publication of CN111027453A publication Critical patent/CN111027453A/en
Application granted granted Critical
Publication of CN111027453B publication Critical patent/CN111027453B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Complex Calculations (AREA)

Abstract

The invention discloses a non-cooperative underwater target automatic identification method based on a Gaussian mixture model, which is used for solving the technical problem of poor practicability of the existing non-cooperative underwater target identification method. The technical scheme is that when an unknown class of target is faced, starting from the angle of a mode of in-set and out-set identification, Mel frequency cepstrum coefficients, namely MFCC coefficients, which can describe the non-linear characteristics of human ear hearing, are extracted by utilizing a known in-set data structure, a Gaussian mixture model is trained, and a proper threshold value is selected, so that a target identification system is constructed.

Description

Automatic non-cooperative underwater target identification method based on Gaussian mixture model
Technical Field
The invention relates to an underwater target identification method, in particular to a non-cooperative underwater target automatic identification method based on a Gaussian mixture model.
Background
The document 'application progress and signal processing of deep learning in passive recognition of underwater targets, 2019, Vol35(9), p 1460-1475' discloses a passive recognition method of underwater targets based on deep learning, which is characterized in that after the steps of a typical pattern classification recognition system of preprocessing sample data and feature extraction are carried out, a specific deep neural network structure is adopted to realize a classifier design and a classifier link, a certain number of samples are used to complete model training of a deep neural network, or the features with good feature learning capability of the deep neural network are directly utilized, and the deep neural network is used to weaken or completely replace the more key feature extraction link in the traditional sense. However, in the document, the modeling is realized by using a deep learning method, a sufficient amount of data with label information is needed, and the modeling effect is seriously influenced by the lack of data. However, in practical application scenarios, it is difficult to collect enough sample data for each underwater target, and the underwater acoustic target data collection is expensive and the existing accumulation is rather weak, and most targets will be in a data-starvation state.
Disclosure of Invention
In order to overcome the defect that the existing non-cooperative underwater target identification method is poor in practicability, the invention provides a non-cooperative underwater target automatic identification method based on a Gaussian mixture model. When the method faces to an unknown class of target, starting from the angle of a mode of in-set and out-set identification, Mel frequency cepstrum coefficients, namely MFCC coefficients, which can describe the non-linear characteristics of human ear hearing, are extracted by using a known in-set data structure, a Gaussian mixture model is trained, a proper threshold value is selected, so that a target identification system is constructed, samples which are difficult to obtain label information are substituted into the target identification system to distinguish whether the samples belong to the known sample set, the class of the samples is judged, the class of a non-cooperative target is preliminarily judged, and meanwhile, the requirements on the label information and the data volume of sample data are reduced. The invention constructs a non-cooperative underwater target automatic identification system based on the human auditory perception mechanism and the Gaussian mixture model by combining MFCC characteristic parameters capable of representing the human auditory characteristics, does not need to acquire specific labels or classes of samples when training the model, and simultaneously reduces the data volume of the samples required to be collected for each underwater target. Through the method of internal and external identification, whether a non-cooperative target which is difficult to obtain label information belongs to a sample in a known sample set or not is distinguished, the category of the target is preliminarily identified, the computing efficiency of the system is high, and the identification performance and robustness are good.
The technical scheme adopted by the invention for solving the technical problems is as follows: a non-cooperative underwater target automatic identification method based on a Gaussian mixture model is characterized by comprising the following steps:
step one, preprocessing training samples observed in a data set, wherein the preprocessing comprises three parts of pre-emphasis, framing and windowing.
And step two, extracting MFCC characteristics from the preprocessed training samples.
And step three, training the GMM model by using the MFCC features extracted from the data in the set as feature vectors. Where for a feature vector, its mixture probability density in the GMM is defined as:
Figure BDA0002305365850000021
where x is the feature vector, λ is GMM, p (x | λ) represents the likelihood that the feature vector appears in the model λ, M is the degree of mixing of the GMM, pi(x) Probability density, ω, for each univariate Gaussian density functioniThe mixed weight occupied by each univariate Gaussian density function is satisfied, and the defined relation is required to be satisfied:
Figure BDA0002305365850000022
the mixed probability density is M p in GMMi(x) Weighted linear combination of (3). p is a radical ofi(x) Is represented by the formula (3):
Figure BDA0002305365850000023
where D is the dimension of the feature vector, μiD-dimensional mean vector, Σ, representing the ith univariate gaussian density functioniRepresenting the ith univariate Gaussian mixtureA covariance matrix of D × D of the degree function, where i ═ 1, 2. Wherein the mixing degree M is selected in advance and is larger than the number of categories, and μi、∑iAnd ωiThen, an expectation-maximization algorithm, namely an EM algorithm, is used to train a GMM model of each category, and the specific steps are as follows:
(a) for a given signature sequence X ═ X1,x2,x3,...,xTSelecting the mixing degree M of the GMM to be trained, setting initial values for the weight, the mean value and the variance of the ith Gaussian distribution, and calculating p (x) according to the formula (1)t| λ), then calculate the log-likelihood function according to equation (4):
Figure BDA0002305365850000024
(b) calculating the posterior probability of the feature vector according to the formula (5) according to the given initial values of the weight, the mean value and the variance;
Figure BDA0002305365850000031
(c) recalculating new weights, means and variances from the a posteriori probabilities calculated in step (b) and calculating new log-likelihood functions by substituting equation (4):
for the ith gaussian density function, the number of eigenvectors contained is:
Figure BDA0002305365850000032
the weight is:
Figure BDA0002305365850000033
the mean value is:
Figure BDA0002305365850000034
the variance is:
Figure BDA0002305365850000035
(d) iteratively iterating steps (b) and (c) until the log-likelihood function or estimated parameters converge.
(e) The three parameters of weight, mean value and variance obtained at this time make P (X | lambda) reach the maximum, complete the parameter estimation of GMM, train GMM model.
Step four, preprocessing and extracting MFCC characteristics in the step one and the step two are carried out on data which are not observed in the set and the non-collected data which are used as test samples.
And step five, substituting the test sample into the trained GMM to calculate the likelihood ratio. For the feature vector x representing the target to be recognized, the determination method can be determined by the following formula:
Λ(x)＝logp(x|λ) (10)
and comparing the log-likelihood ratio of the feature vector belonging to the set with the threshold value by giving a threshold value, wherein if the log-likelihood ratio is greater than the threshold value, the sample belongs to the set, and otherwise, the sample belongs to the outside of the set.
Step six, respectively calculating the false alarm probability E of the system when testing the systemfaAnd probability of missed alarm EmissThe definition is as follows:
Figure BDA0002305365850000036
Figure BDA0002305365850000041
wherein n isfaRepresenting the number of times that off-set data is misinterpreted as in-set data, nimposterIs the number of data outside the set, nmissIs the number of times that the data in the set is overlooked and judged as data out of the set, ntargetRefers to the number of data in the set.
The detection error relation curve, namely DET curve, is obtained according to different decision threshold values within the score range of the test setfaAnd EmissGraph of the corresponding relationship of (a). The closer the DET curve is to the lower left and the closer to the origin, the corresponding EfaAnd EmissThe lower the probability of error occurrence, the higher the model accuracy and the better the accuracy of system identification. With decreasing threshold, EfaIncrease and EmissDecreasing, the DET curve tends to fall, when the threshold decreases to a certain point, Efa＝EmissThe corresponding threshold is the equal error rate threshold. And finding out the equal false recognition rate and the equal false recognition rate threshold value for the actual use stage of the system by drawing the DET curve.
The invention has the beneficial effects that: when the method faces to an unknown class of target, starting from the angle of a mode of in-set and out-set identification, Mel frequency cepstrum coefficients, namely MFCC coefficients, which can describe the non-linear characteristics of human ear hearing, are extracted by using a known in-set data structure, a Gaussian mixture model is trained, a proper threshold value is selected, so that a target identification system is constructed, samples which are difficult to obtain label information are substituted into the target identification system to distinguish whether the samples belong to the known sample set, the class of the samples is judged, the class of a non-cooperative target is preliminarily judged, and meanwhile, the requirements on the label information and the data volume of sample data are reduced. The invention constructs a non-cooperative underwater target automatic identification system based on the human auditory perception mechanism and the Gaussian mixture model by combining MFCC characteristic parameters capable of representing the human auditory characteristics, does not need to acquire specific labels or classes of samples when training the model, and simultaneously reduces the data volume of the samples required to be collected for each underwater target. Through the method of internal and external identification, whether a non-cooperative target which is difficult to obtain label information belongs to a sample in a known sample set or not is distinguished, the category of the target is preliminarily identified, the computing efficiency of the system is high, and the identification performance and robustness are good.
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
Drawings
FIG. 1 is a flow chart of the non-cooperative underwater target automatic identification method based on the Gaussian mixture model.
Fig. 2 is a flow chart of MFCC feature extraction in fig. 1.
Fig. 3 is a diagram of the GMM calculation structure in the method of the present invention.
Fig. 4 is a DET curve plotted for an embodiment of the method of the present invention.
Detailed Description
Reference is made to fig. 1-4.
The data set used in this embodiment includes 3 types of underwater sound targets, 15 pieces of sound are collected for each type, the length of each piece of sound is about 6 seconds, total 45 sample files are respectively recorded as type i, type ii, and type iii, and the sampling frequency is 8000 Hz. During testing, the class I and class II data are used as the data in the set, and the class III data are used as the data out of the set. And selecting a part of the I-type and II-type data as a training set for training and establishing the GMM model, and using the rest part of the I-type and II-type data and the III-type data as a test set for testing the trained GMM model.
The automatic identification method of the non-cooperative underwater target based on the Gaussian mixture model comprises the following specific steps:
the method comprises the following steps: and preprocessing the training samples observed in the data set, wherein the preprocessing comprises three parts of pre-emphasis, framing and windowing.
The method comprises the steps of adopting MATLAB as an algorithm implementation platform, selecting 7 samples from class I and class II data respectively, taking 14 samples as training samples, completing frame division, windowing, filter bank construction and other processing by utilizing related functions in a voicebox toolbox, and completing preprocessing, wherein the duration of each frame is 0.1s during framing processing.
Step two: MFCC features are extracted from the preprocessed training samples.
And extracting 13-dimensional MFCC characteristics of the preprocessed result to obtain a 1667 x 13 characteristic matrix.
Step three: calling a fitgmist function in MATLAB by using the MFCC characteristics extracted from the training samples, setting the mixing degree M to be 5, finishing parameter estimation, and training a GMM model, wherein the mixing weight of the GMM is an array of 1 multiplied by 5, the mean value is an array of 5 multiplied by 13, and the covariance is an array of 13 multiplied by 5.
And training the GMM model by using the MFCC features extracted from the data in the set as feature vectors. Where for a feature vector, its mixture probability density in the GMM is defined as:
Figure BDA0002305365850000051
where x is the feature vector, λ is GMM, p (x | λ) represents the likelihood that the feature vector appears in the model λ, M is the degree of mixing of the GMM, pi(x) Probability density, ω, for each univariate Gaussian density functioniThe mixed weight occupied by each univariate Gaussian density function is satisfied, and the defined relation is required to be satisfied:
Figure BDA0002305365850000052
the mixed probability density is M p in GMMi(x) Weighted linear combination of (3). p is a radical ofi(x) Is represented by the formula (3):
Figure BDA0002305365850000053
where D is the dimension of the feature vector, μiD-dimensional mean vector, Σ, representing the ith univariate gaussian density functioniA covariance matrix of D × D representing the ith univariate gaussian density function, where i ═ 1, 2. Wherein the mixing degree M is selected in advance and is larger than the number of categories, and μi、∑iAnd ωiThen, using expectation maximization algorithm, that is, EM algorithm, it may be estimated, so as to train GMM models of various classes, and the specific steps are as follows:
(a) for a given signature sequence X ═ X1,x2,x3,...,xTSelecting the mixing degree M of the GMM to be trained, setting initial values for the weight, the mean value and the variance of the ith Gaussian distribution, and calculating p (x) according to the formula (1)t| λ), then calculate the log-likelihood function according to equation (4):
Figure BDA0002305365850000061
(b) calculating the posterior probability of the feature vector according to the formula (5) according to the given initial values of the weight, the mean value and the variance;
Figure BDA0002305365850000062
(c) and (3) calculating new weight, mean value and variance according to the posterior probability calculated in the last step, and substituting formula (4) to calculate new log-likelihood function:
for the ith gaussian density function, the number of eigenvectors contained is:
Figure BDA0002305365850000063
the weight is:
Figure BDA0002305365850000064
the mean value is:
Figure BDA0002305365850000065
the variance is:
Figure BDA0002305365850000066
(d) iteratively iterating steps (b) and (c) until the log-likelihood function or estimated parameters converge.
(e) The three parameters of weight, mean value and variance obtained at this time can make P (X | lambda) reach the maximum, thus can complete the parameter estimation of GMM, train the GMM model.
Step four: and taking the rest samples of each class in the 3 classes of underwater sound targets as test samples, similarly performing the steps of preprocessing and extracting 13-dimensional MFCC features, obtaining a 1905 x 13 feature matrix belonging to the set from the rest 16 samples of the class I and class II data, and obtaining an 1800 x 13 feature matrix belonging to the outside of the set from 15 samples of the class III.
And (4) performing preprocessing in the first step and the second step and extracting MFCC features on the data which are not observed in the set and the non-collected data as test samples.
Step five: and substituting the test sample into the trained GMM to calculate the likelihood ratio. For the feature vector x representing the target to be recognized, the determination method can be determined by the following formula:
Λ(x)＝logp(x|λ) (10)
and comparing the log-likelihood ratio of the feature vector belonging to the set with the threshold value by giving a threshold value, wherein if the log-likelihood ratio is greater than the threshold value, the sample belongs to the set, and otherwise, the sample belongs to the outside of the set.
Step six: when testing the system, respectively calculating the false alarm probability (E) of the systemfa) And probability of missed alarm (E)miss) The definition is as follows:
Figure BDA0002305365850000071
Figure BDA0002305365850000072
wherein n isfaRepresenting the number of times that off-set data is misinterpreted as in-set data, nimposterIs the number of data outside the set, nmissIs the number of times that the data in the set is overlooked and judged as data out of the set, ntargetRefers to the number of data in the set.
The detection error relation curve, namely DET curve, is obtained according to different decision threshold values within the score range of the test setfaAnd EmissGraph of the corresponding relationship of (a). The closer the DET curve is to the lower left and the closer to the origin, the corresponding EfaAnd EmissThe lower the probability of error occurrence, the higher the model accuracy and the better the accuracy of system identification. With decreasing threshold, EfaIncrease and EmissDecreasing, the DET curve tends to fall, when the threshold decreases to a certain point, Efa＝EmissThe corresponding threshold is the equal error rate threshold. And finding out the equal false recognition rate and the equal false recognition rate threshold value for the actual use stage of the system by drawing the DET curve.

Claims (1)

1. A non-cooperative underwater target automatic identification method based on a Gaussian mixture model is characterized by comprising the following steps:
the method comprises the following steps that firstly, training samples observed in a data set are preprocessed, and the preprocessing comprises three parts, namely pre-emphasis, framing and windowing;
step two, extracting MFCC characteristics from the preprocessed training samples;
step three, training a GMM model by using MFCC features extracted from the data in the set as feature vectors; where for a feature vector, its mixture probability density in the GMM is defined as:
Figure FDA0002305365840000011
where x is the feature vector, λ is GMM, p (x | λ) represents the likelihood that the feature vector appears in the model λ, M is the degree of mixing of the GMM, pi(x) Probability density, ω, for each univariate Gaussian density functioniThe mixed weight occupied by each univariate Gaussian density function is satisfied, and the defined relation is required to be satisfied:
Figure FDA0002305365840000012
the mixed probability density is M p in GMMi(x) Weighted linear combination of (3); p is a radical ofi(x) Is represented by the formula (3):
Figure FDA0002305365840000013
where D is the dimension of the feature vector, μiD-dimensional mean vector, Σ, representing the ith univariate gaussian density functioniDxD covariance matrix representing the ith univariate Gaussian density functionWherein i 1, 2.. said, M; wherein the mixing degree M is selected in advance and is larger than the number of categories, and μi、∑iAnd ωiThen, an expectation-maximization algorithm, namely an EM algorithm, is used to train a GMM model of each category, and the specific steps are as follows:
(a) for a given signature sequence X ═ X1,x2,x3,...,xTSelecting the mixing degree M of the GMM to be trained, setting initial values for the weight, the mean value and the variance of the ith Gaussian distribution, and calculating p (x) according to the formula (1)t| λ), then calculate the log-likelihood function according to equation (4):
Figure FDA0002305365840000014
(b) calculating the posterior probability of the feature vector according to the formula (5) according to the given initial values of the weight, the mean value and the variance;
Figure FDA0002305365840000021
(c) recalculating new weights, means and variances from the a posteriori probabilities calculated in step (b) and calculating new log-likelihood functions by substituting equation (4):
for the ith gaussian density function, the number of eigenvectors contained is:
Figure FDA0002305365840000022
the weight is:
Figure FDA0002305365840000023
the mean value is:
Figure FDA0002305365840000024
the variance is:
Figure FDA0002305365840000025
(d) iteratively iterating steps (b) and (c) until the log-likelihood function or estimated parameters converge;
(e) the three parameters of weight, mean value and variance obtained at the moment make P (X | lambda) reach the maximum, complete the parameter estimation of GMM, train out GMM model;
step four, preprocessing and extracting MFCC characteristics in the step one and the step two are carried out on data which are not observed in the set and the non-collected data which are used as test samples;
substituting the test sample into the trained GMM, and calculating the likelihood ratio; for the feature vector x representing the target to be recognized, the determination method can be determined by the following formula:
Λ(x)＝logp(x|λ) (10)
comparing the log-likelihood ratio of the feature vector belonging to the set with the threshold value by giving a threshold value, wherein if the log-likelihood ratio is greater than the threshold value, the sample belongs to the set, otherwise, the sample belongs to the outside of the set;
step six, respectively calculating the false alarm probability E of the system when testing the systemfaAnd probability of missed alarm EmissThe definition is as follows:
Figure FDA0002305365840000026
Figure FDA0002305365840000031
wherein n isfaRepresenting the number of times that off-set data is misinterpreted as in-set data, nimposterIs the number of data outside the set, nmissIs the number of times that the data in the set is overlooked and judged as data out of the set, ntargetThe number of data in the set;
the detection error relation curve, namely DET curve, is obtained according to different decision threshold values within the score range of the test setfaAnd EmissA graph of the corresponding relationship of (a); the closer the DET curve is to the lower left and the closer to the origin, the corresponding EfaAnd EmissThe lower the error condition is, the smaller the possibility of the error condition is, the higher the model precision is, and the accuracy of system identification is better; with decreasing threshold, EfaIncrease and EmissDecreasing, the DET curve tends to fall, when the threshold decreases to a certain point, Efa＝EmissWhen the EER is equal to the EER, the corresponding threshold value is the equal error recognition rate threshold value; and finding out the equal false recognition rate and the equal false recognition rate threshold value for the actual use stage of the system by drawing the DET curve.
CN201911237928.0A 2019-12-06 2019-12-06 Automatic non-cooperative underwater target identification method based on Gaussian mixture model Active CN111027453B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911237928.0A CN111027453B (en) 2019-12-06 2019-12-06 Automatic non-cooperative underwater target identification method based on Gaussian mixture model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911237928.0A CN111027453B (en) 2019-12-06 2019-12-06 Automatic non-cooperative underwater target identification method based on Gaussian mixture model

Publications (2)

Publication Number Publication Date
CN111027453A true CN111027453A (en) 2020-04-17
CN111027453B CN111027453B (en) 2022-05-17

Family

ID=70207325

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911237928.0A Active CN111027453B (en) 2019-12-06 2019-12-06 Automatic non-cooperative underwater target identification method based on Gaussian mixture model

Country Status (1)

Country Link
CN (1) CN111027453B (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002029785A1 (en) * 2000-09-30 2002-04-11 Intel Corporation Method, apparatus, and system for speaker verification based on orthogonal gaussian mixture model (gmm)
EP2328363A1 (en) * 2009-09-11 2011-06-01 Starkey Laboratories, Inc. Sound classification system for hearing aids
CN102324232A (en) * 2011-09-12 2012-01-18 辽宁工业大学 Method for recognizing sound-groove and system based on gauss hybrid models
CN102820033A (en) * 2012-08-17 2012-12-12 南京大学 Voiceprint identification method
WO2014153800A1 (en) * 2013-03-29 2014-10-02 京东方科技集团股份有限公司 Voice recognition system
US9892726B1 (en) * 2014-12-17 2018-02-13 Amazon Technologies, Inc. Class-based discriminative training of speech models
CN108269573A (en) * 2017-01-03 2018-07-10 蓝盾信息安全技术有限公司 Speaker Recognition System based on vector quantization and gauss hybrid models
CN109800700A (en) * 2019-01-15 2019-05-24 哈尔滨工程大学 A kind of underwater sound signal target classification identification method based on deep learning
CN110197286A (en) * 2019-05-10 2019-09-03 武汉理工大学 A kind of Active Learning classification method based on mixed Gauss model and sparse Bayesian

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002029785A1 (en) * 2000-09-30 2002-04-11 Intel Corporation Method, apparatus, and system for speaker verification based on orthogonal gaussian mixture model (gmm)
EP2328363A1 (en) * 2009-09-11 2011-06-01 Starkey Laboratories, Inc. Sound classification system for hearing aids
CN102324232A (en) * 2011-09-12 2012-01-18 辽宁工业大学 Method for recognizing sound-groove and system based on gauss hybrid models
CN102820033A (en) * 2012-08-17 2012-12-12 南京大学 Voiceprint identification method
WO2014153800A1 (en) * 2013-03-29 2014-10-02 京东方科技集团股份有限公司 Voice recognition system
US9892726B1 (en) * 2014-12-17 2018-02-13 Amazon Technologies, Inc. Class-based discriminative training of speech models
CN108269573A (en) * 2017-01-03 2018-07-10 蓝盾信息安全技术有限公司 Speaker Recognition System based on vector quantization and gauss hybrid models
CN109800700A (en) * 2019-01-15 2019-05-24 哈尔滨工程大学 A kind of underwater sound signal target classification identification method based on deep learning
CN110197286A (en) * 2019-05-10 2019-09-03 武汉理工大学 A kind of Active Learning classification method based on mixed Gauss model and sparse Bayesian

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
M V UNNIKRISHNAN ET AL: "Mimicking voice recognition using MFCC-GMM framework", 《2017 INTERNATIONAL CONFERENCE ON TRENDS IN ELECTRONICS AND INFORMATICS (ICEI)》 *
李慧慧: "基于深度学习的短语音说话人识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 *
申小虎等: "一种改进动态特征参数的话者语音识别系统", 《计算机仿真》 *
胡益平: "基于GMM的说话人识别技术研究与实现", 《中国优秀硕士学位论文全文数据库》 *

Also Published As

Publication number Publication date
CN111027453B (en) 2022-05-17

Similar Documents

Publication Publication Date Title
CN107610707B (en) A kind of method for recognizing sound-groove and device
Hansen et al. Speaker recognition by machines and humans: A tutorial review
US9489965B2 (en) Method and apparatus for acoustic signal characterization
Huang et al. Predicting speech recognition confidence using deep learning with word identity and score features
US20030225719A1 (en) Methods and apparatus for fast and robust model training for object classification
CN110120230B (en) Acoustic event detection method and device
CN106991312B (en) Internet anti-fraud authentication method based on voiceprint recognition
CN101452701B (en) Confidence degree estimation method and device based on inverse model
CN109378014A (en) A kind of mobile device source discrimination and system based on convolutional neural networks
CN111816185A (en) Method and device for identifying speaker in mixed voice
CN108831506A (en) Digital audio based on GMM-BIC distorts point detecting method and system
CN111477219A (en) Keyword distinguishing method and device, electronic equipment and readable storage medium
CN112560596A (en) Radar interference category identification method and system
CN110827809B (en) Language identification and classification method based on condition generation type confrontation network
CN115331661A (en) Voiceprint recognition backdoor attack defense method based on feature clustering analysis and feature dimension reduction
Beritelli et al. The role of voice activity detection in forensic speaker verification
CN111243621A (en) Construction method of GRU-SVM deep learning model for synthetic speech detection
Perera et al. Optimization of the DET curve in speaker verification under noisy conditions
CN116705063B (en) Manifold measurement-based multi-model fusion voice fake identification method
Woubie et al. Voice quality features for replay attack detection
CN110415707B (en) Speaker recognition method based on voice feature fusion and GMM
CN111027453B (en) Automatic non-cooperative underwater target identification method based on Gaussian mixture model
CN108694950B (en) Speaker confirmation method based on deep hybrid model
US7437289B2 (en) Methods and apparatus for the systematic adaptation of classification systems from sparse adaptation data
CN112820318A (en) Impact sound model establishment and impact sound detection method and system based on GMM-UBM

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
GR01 Patent grant
GR01 Patent grant