CN105070297A - MP3 audio compression history detection method - Google Patents
MP3 audio compression history detection method Download PDFInfo
- Publication number
- CN105070297A CN105070297A CN201510419213.2A CN201510419213A CN105070297A CN 105070297 A CN105070297 A CN 105070297A CN 201510419213 A CN201510419213 A CN 201510419213A CN 105070297 A CN105070297 A CN 105070297A
- Authority
- CN
- China
- Prior art keywords
- scale factor
- audio
- value
- audio frequency
- compression
- 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
Links
Abstract
The invention discloses an MP3 audio compression history detection method. According to the method, primary compressed MP3 audio samples, secondary compressed MP3 audio samples, and three-time compressed MP3 audio samples corresponding to uncompressed WAV audio samples are acquired, the coding bit rates of MP3 encoders employed by first compression for acquiring the secondary compressed MP3 audio samples and first compression for acquiring the three-time compressed MP3 audio samples are limited to be same, and the coding bit rate of the MP3 encoders employed by second compression for acquiring the primary compressed MP3 audio samples and the secondary compressed MP3 audio samples and the coding bit rate of the MP3 encoders employed by second compression and third compression for acquiring the three-time compressed MP3 audio samples are the same. By employing a training template obtained by the training of respective characteristic values of the MP3 audio samples, whether the to-be-detected MP3 audio samples are primary compressed MP3 audio samples or secondary compressed MP3 audio samples or three-time compressed MP3 audio samples can be well determined, MP3 audio compression history detection is realized, the detection accuracy is high, and the calculation complexity is low.
Description
Technical field
The present invention relates to a kind of DAB evidence collecting method, especially relate to a kind of MP3 audio compression history detection method.
Background technology
Along with the development of multimedia technology, various authoring tool arises at the historic moment, such as Coo1EditPro, GoldWave etc.The generation of these authoring tools to improving multimedia messages, to strengthen the audio visual effect of multimedia messages significant.But, also some negative impacts are brought while authoring tool brings convenience to people's lives, such as some lawless persons reach hidden object, carry out variously distorting and forging to multimedia messages by operation authoring tool, distort due to these and forge and often have malicious and not easily discovered, therefore can threaten to individual and even nation's security, can social stability be had a strong impact on.
Audio frequency is the important component part in multimedia messages, and it has easy acquisition, is convenient to the features such as storage.MP3 is as audio format most popular on current network, and judging whether MP3 audio frequency is tampered is current urgent problem.Because distorting of audio frequency could effectively must be carried out in the form of uncompressed domain, therefore as MP3 audio frequency, the WAV audio frequency that just first its solution must be pressed into uncompressed domain is distorted to compressed domain audio, then could increase this WAV audio frequency, delete, shear, splicing etc. other distort operation, finally again the WAV audio compression after distorting is become MP3 audio frequency.Can find from this series of distorting operating process, the MP3 audio frequency be tampered must live through repeatedly the process of Compress softwares, therefore by the compression histories of research MP3 audio frequency, effectively can infer whether MP3 audio frequency is tampered.And current compression histories detect delay is mainly for image domains, and audio compression history detect delay is mainly for the two compressed detected of MP3 audio frequency, detects concern very few to the compression histories of MP3 audio frequency more than twice.
Summary of the invention
Technical matters to be solved by this invention is to provide a kind of MP3 audio compression history detection method, it can detect the compression histories of MP3 audio frequency effectively, namely can determine that MP3 audio frequency to be measured is through once exactly, the MP3 audio frequency of secondary or three second compression.
The present invention solves the problems of the technologies described above adopted technical scheme: a kind of MP3 audio compression history detection method, is characterized in that comprising the following steps:
1. N number of style difference is chosen and unpressed WAV audio sample, wherein, N >=10;
2. the first compression MP3 audio sample corresponding with each unpressed WAV audio sample is obtained, detailed process is: utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression;
3. the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample is obtained, 3.-1 detailed process is:, utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression, each first compression MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device;-2 3., utilize MP3 scrambler to carry out compressed encoding to each WAV audio sample that solution in step 3.-1 is pressed into, obtain the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample;
Wherein, 3.-1 all 2. the middle MP3 scrambler used is identical with described step with the MP3 scrambler used in described step 3.-2 for described step, and the coding bit rate of the MP3 scrambler of described step 3. in-1 and described step 2. in the coding bit rate of MP3 scrambler identical or not identical, the coding bit rate of the MP3 scrambler of described step 3. in-2 and described step 2. in the coding bit rate of MP3 scrambler identical;
4. the three second compression MP3 audio sample corresponding with each unpressed WAV audio sample are obtained, 4.-1 detailed process is:, utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression, each first compression MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device; 4.-2, MP3 scrambler is utilized to carry out compressed encoding to each WAV audio sample that solution in step 4.-1 is pressed into, obtain the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample, each second-compressed MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device;-3 4., utilize MP3 scrambler to carry out compressed encoding to each WAV audio sample that solution in step 4.-2 is pressed into, obtain the three second compression MP3 audio sample corresponding with each unpressed WAV audio sample;
Wherein, described step 4.-1, 4.-2 all 2. the middle MP3 scrambler used is identical with described step with the MP3 scrambler used in described step 4.-3 for described step, described step 4.-1 with in described step 4.-2 use MP3 decoding device all identical with the MP3 decoding device used in described step 3.-1, and the coding bit rate of the MP3 scrambler of described step 4. in-1 is identical with the coding bit rate of the MP3 scrambler in described step 3.-1, described step is 4.-2 all identical with the coding bit rate of the MP3 scrambler in described step 3.-2 with the coding bit rate of the MP3 scrambler in described step 4.-3,
5. each first compression MP3 audio sample is labeled as-1, each second-compressed MP3 audio sample is labeled as 0, each three second compression MP3 audio sample are labeled as 1, again all first compression MP3 audio sample, all second-compressed MP3 audio sample and three all second compression MP3 audio sample are formed a training sample set, wherein, each subsample in training sample set is first compression MP3 audio sample or is second-compressed MP3 audio sample or be three second compression MP3 audio sample;
6. 47 eigenwerts of each subsample in training sample set are extracted, and 47 eigenwerts of each subsample in training sample set are formed a row vector, the row vector that 47 eigenwerts of the subsample of the kth in training sample set are formed is designated as F
k; Then adopt 47 eigenwerts of min-max method for normalizing to each subsample in training sample set to be normalized, obtain the eigenwert after 47 normalizeds of each subsample in training sample set; Wherein, the initial value of k is 1,1≤k≤K, K represent total number of the subsample comprised in training sample set, K=3N;
7. utilize LibSVM sorter to train the eigenwert after respective 47 normalizeds in all subsamples in training sample set, obtain training template; Wherein, adopt cross validation mode [2 in training process
-5, 2
5] choose best penalty parameter c and best RBF nuclear parameter g in interval, all the other parameters all Use Defaults;
8. M MP3 audio frequency to be detected is chosen arbitrarily, wherein, M >=1, each MP3 audio frequency to be detected is first compression MP3 audio frequency or is second-compressed MP3 audio frequency or be three second compression MP3 audio frequency, and in the coding bit rate of MP3 scrambler that in the acquisition process of each MP3 audio frequency to be detected, last compressed encoding uses and the acquisition process of the subsample in training sample set, the coding bit rate of MP3 scrambler that uses of compressed encoding is identical for the last time; Then according to step 6. in extract the process of 47 eigenwerts of each subsample in training sample set, extract 47 eigenwerts of each MP3 audio frequency to be detected in an identical manner; Then adopt 47 eigenwerts of min-max method for normalizing to each MP3 audio frequency to be detected to be normalized, obtain the eigenwert after 47 normalizeds of each MP3 audio frequency to be detected; Again the eigenwert after 47 normalizeds of each MP3 audio frequency to be detected is input in training template and detects, if the Output rusults of training template is-1, then determine that corresponding MP3 audio frequency to be detected is first compression MP3 audio frequency, if the Output rusults of training template is 0, then determine that corresponding MP3 audio frequency to be detected is second-compressed MP3 audio frequency, if the Output rusults of training template is 1, then determine that corresponding MP3 audio frequency to be detected is three second compression MP3 audio frequency.
Described step is middle F 6.
kacquisition process be:
6. be-1, long window coded frame by the frame definition utilizing long window coded system to carry out processing in the subsample of the kth in training sample set; Then the MP3 decoding device utilizing step 3. to use in-1 carries out decoding process to the every frame in the subsample of the kth in training sample set, obtain the kth decoded WAV audio frequency of sub-sample in training sample set, in decoding process, extract the position of the every frame length window coded frame in the kth subsample in training sample set; Then according to the position of the every frame length window coded frame in the subsample of the kth in training sample set, obtain the scale factor matrix of the position of all long window coded frame in the kth subsample in training sample set, be designated as sf
a, wherein, if the kth subsample in training sample set is monophonic audio, then sf
adimension be 2w × 21, if the kth subsample in training sample set is dual-channel audio, then sf
adimension be 4w × 21, w represents the totalframes of the long window coded frame in the kth subsample in training sample set;
6. the MP3 scrambler used in-2, utilizing step 2. carries out coded treatment to the decoded WAV audio frequency obtained in step 6.-1, obtain new MP3 audio frequency, in coding process, according to the position of the every frame length window coded frame in the kth subsample in the training sample set extracted in step 6.-1, the scale factor matrix of the position of all long window coded frame of decoded WAV audio frequency in coding process that 6. obtaining step obtains in-1, is designated as sf
bwherein, the coding bit rate of the MP3 scrambler in this step is identical with the coding bit rate of the MP3 scrambler that compressed encoding last in the acquisition process of the subsample of the kth in training sample set uses, if the kth subsample in training sample set is monophonic audio, then sf
bdimension be 2w × 21, if the kth subsample in training sample set is dual-channel audio, then sf
bdimension be 4w × 21;
6.-3, sf is calculated
ain scale factor 1 to 5 shift as sf
bin the transition probability matrix of scale factor 0 to 3, be designated as P, wherein, the dimension of P is the value of the 1st row the 1st column element in 5 × 4, P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 0, the value of the 1st row the 2nd column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 1, the value of the 1st row the 3rd column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 2, the value of the 1st row the 4th column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 3, the like, the value of the 5th row the 4th column element in P is sf
ain scale factor 5 shift as sf
bin the transition probability of scale factor 3; Then using the value of the element of 20 in P according to the order of sequence as front 20 eigenwerts of the subsample of the kth in training sample set;
6.-4, sf is calculated
awith sf
bmatrix of differences, be designated as Δ sf, Δ sf=sf
a-sf
b; Then the average of the value of all elements in Δ sf is calculated, using the 21st eigenwert of this average as the subsample of the kth in training sample set;
6.-5, sf is calculated
ain the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
ain the probability distribution of scale factor 0,
in the value of the 2nd element be sf
ain the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
ain the probability distribution of scale factor 8;
And calculate sf
bin the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
bin the probability distribution of scale factor 0,
in the value of the 2nd element be sf
bin the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
bin the probability distribution of scale factor 8;
6.-6, calculate
with
difference value vector, be designated as Δ P,
Δ P represents the probability of use incremental vector of scale factor 0 to 8; Then using the value of the element of 9 in Δ P according to the order of sequence as the 22nd eigenwert of the subsample of the kth in training sample set to the 30th eigenwert;
6.-7, calculate
with
quotient vector, be designated as dP,
dP represents the probability of use increased times vector of scale factor 0 to 8; Then using the value of the element of 9 in dP according to the order of sequence as the 31st eigenwert of the subsample of the kth in training sample set to the 39th eigenwert;
-8 6., the difference of the probability of use increased times of scale factor 0 and the respective probability of use increased times of scale factor 1 to 8 is calculated, using 8 differences obtaining according to the order of sequence as the 40th eigenwert of the subsample of the kth in training sample set to the 47th eigenwert;
6.-9,47 eigenwerts of the subsample of the kth in training sample set are formed row vector F
k.
Described step 8. in the leaching process of 47 eigenwerts of a MP3 audio frequency to be detected be:
8. be-1, long window coded frame by the frame definition utilizing long window coded system to carry out processing in MP3 audio frequency to be detected; Then the MP3 decoding device utilizing step 3. to use in-1 carries out decoding process to the every frame in MP3 audio frequency to be detected, obtain the WAV audio frequency after MP3 audio decoder to be detected, in decoding process, extract the position of the every frame length window coded frame in MP3 audio frequency to be detected; Then according to the position of the every frame length window coded frame in MP3 audio frequency to be detected, obtain the scale factor matrix of the position of all long window coded frame in MP3 audio frequency to be detected, be designated as sf
a', wherein, if MP3 audio frequency to be detected is monophonic audio, then sf
a' dimension be 2w' × 21, if MP3 audio frequency to be detected is dual-channel audio, then sf
a' dimension be 4w' × 21, w' represents the totalframes of the long window coded frame in MP3 audio frequency to be detected;
8. the MP3 scrambler used in-2, utilizing step 2. carries out coded treatment to the decoded WAV audio frequency obtained in step 8.-1, obtain new MP3 audio frequency, in coding process, according to the position of the every frame length window coded frame in the MP3 audio frequency to be detected extracted in step 8.-1, the scale factor matrix of the position of all long window coded frame of decoded WAV audio frequency in coding process that 8. obtaining step obtains in-1, is designated as sf
b', wherein, the coding bit rate of the MP3 scrambler in this step is identical with the coding bit rate of the MP3 scrambler that compressed encoding last in the acquisition process of MP3 audio frequency to be detected uses, if MP3 audio frequency to be detected is monophonic audio, then sf
b' dimension be 2w' × 21, if MP3 audio frequency to be detected is dual-channel audio, then sf
b' dimension be 4w' × 21;
8.-3, sf is calculated
a' in scale factor 1 to 5 shift as sf
b' in the transition probability matrix of scale factor 0 to 3, be designated as P', wherein, the dimension of P' is the value of the 1st row the 1st column element in 5 × 4, P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 0, the value of the 1st row the 2nd column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 1, the value of the 1st row the 3rd column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 2, the value of the 1st row the 4th column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 3, the like, the value of the 5th row the 4th column element in P' is sf
a' in scale factor 5 shift as sf
b' in the transition probability of scale factor 3; Then using the value of the element of 20 in P' according to the order of sequence as front 20 eigenwerts of MP3 audio frequency to be detected;
8.-4, sf is calculated
a' and sf
b' matrix of differences, be designated as Δ sf', Δ sf'=sf
a'-sf
b'; Then the average of the value of all elements in Δ sf' is calculated, using the 21st eigenwert of this average as MP3 audio frequency to be detected;
8.-5, sf is calculated
a' in the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
a' in the probability distribution of scale factor 0,
in the value of the 2nd element be sf
a' in the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
a' in the probability distribution of scale factor 8;
And calculate sf
b' in the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
b' in the probability distribution of scale factor 0,
in the value of the 2nd element be sf
b' in the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
b' in the probability distribution of scale factor 8;
8.-6, calculate
with
difference value vector, be designated as Δ P',
Δ P' represents the probability of use incremental vector of scale factor 0 to 8; Then using the value of the element of 9 in Δ P' according to the order of sequence as the 22nd eigenwert of MP3 audio frequency to be detected to the 30th eigenwert;
8.-7, calculate
with
quotient vector, be designated as dP',
dP' represents the probability of use increased times vector of scale factor 0 to 8; Then using the value of the element of 9 in dP' according to the order of sequence as the 31st eigenwert of MP3 audio frequency to be detected to the 39th eigenwert;
-8 8., the difference of the probability of use increased times of scale factor 0 and the respective probability of use increased times of scale factor 1 to 8 is calculated, using 8 differences obtaining according to the order of sequence as the 40th eigenwert of MP3 audio frequency to be detected to the 47th eigenwert.
Compared with prior art, the invention has the advantages that:
1) by obtaining first compression MP3 audio sample corresponding to unpressed WAV audio sample, second-compressed MP3 audio sample and three second compression MP3 audio sample, and the first time compression that restriction obtains second-compressed MP3 audio sample is identical with the coding bit rate that the first time obtaining three second compression MP3 audio sample compresses the MP3 scrambler used, obtain first compression MP3 audio sample, obtain the second time compression of second-compressed MP3 audio sample, the second time compression obtaining three second compression MP3 audio sample is identical with the coding bit rate compressing the MP3 scrambler used for the third time, utilize these MP3 audio sample eigenwert separately train the training template obtained can be good at judging MP3 audio sample to be detected as the MP3 audio sample of first compression or for the MP3 audio sample of second-compressed be still the MP3 audio sample of three second compression, achieve MP3 audio compression history to detect.
2) the inventive method to be encoded ultimate principle by analyzing MP3, according to the Changing Pattern of long window scale factor in MP3 many compression processes, extracts the foundation that the eigenwert based on long window scale factor detects as MP3 audio compression history.By the complexity utilizing the feature of long window scale factor effectively can reduce detection method.
3) the present invention is by the difference average of the long window scale factor before and after extraction two second compression, and utilize probabilistic method and single step Markov Transition Probabilities to extract the probability distribution of long window scale factor and the transition probability validity feature as this detection method respectively, the extraction of these features and utilize the accuracy rate that effectively can improve detection method.
Accompanying drawing explanation
Fig. 1 be the inventive method totally realize block diagram.
Embodiment
Below in conjunction with accompanying drawing embodiment, the present invention is described in further detail.
A kind of MP3 audio compression history detection method that the present invention proposes, it totally realizes block diagram as shown in Figure 1, and it comprises the following steps:
1. choose N number of style difference and unpressed WAV audio sample, wherein, N >=10, as got N=300.At this, unpressed WAV audio sample can select the audio frequency of the styles such as blues, pop, classical, country and folk.
2. the first compression MP3 audio sample corresponding with each unpressed WAV audio sample is obtained, detailed process is: utilize existing MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression.At this, existing MP3 scrambler is as adopted lame3.99.5MP3 scrambler, and coding bit rate is br1
1, br1
1it can be a kind of coding bit rate in 64kbps, 80kbps, 96kbps, 112kbps, 128kbps, 160kbps and 192kbps.
3. the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample is obtained, 3.-1 detailed process is:, utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression, recycle existing MP3 decoding device and each first compression MP3 audio sample solution obtained is pressed into WAV audio sample;-2 3., utilize MP3 scrambler to carry out compressed encoding to each WAV audio sample that solution in step 3.-1 is pressed into, obtain the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample.
Wherein, 3.-1 all 2. the middle MP3 scrambler used is identical with described step with the MP3 scrambler used in described step 3.-2 for described step, and the coding bit rate of the MP3 scrambler of described step 3. in-1 and described step 2. in the coding bit rate of MP3 scrambler identical or not identical, the coding bit rate of the MP3 scrambler of described step 3. in-2 and described step 2. in the coding bit rate of MP3 scrambler identical.If step 2. in the coding bit rate of MP3 scrambler that uses be br1
1for 128kbps, then the coding bit rate br2 of the MP3 scrambler of step 3. in-1
1can be a kind of coding bit rate in 64kbps, 80kbps, 96kbps, 112kbps, 128kbps, 160kbps and 192kbps, the coding bit rate br2 of the MP3 scrambler of step 3. in-2
2for 128kbps.
At this, existing MP3 decoding device is as adopted lame3.99.5MP3 demoder.
4. the three second compression MP3 audio sample corresponding with each unpressed WAV audio sample are obtained, 4.-1 detailed process is:, utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression, each first compression MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device; 4.-2, MP3 scrambler is utilized to carry out compressed encoding to each WAV audio sample that solution in step 4.-1 is pressed into, obtain the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample, each second-compressed MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device;-3 4., utilize MP3 scrambler to carry out compressed encoding to each WAV audio sample that solution in step 4.-2 is pressed into, obtain the three second compression MP3 audio sample corresponding with each unpressed WAV audio sample.
Wherein, described step 4.-1, 4.-2 all 2. the middle MP3 scrambler used is identical with described step with the MP3 scrambler used in described step 4.-3 for described step, described step 4.-1 with in described step 4.-2 use MP3 decoding device all identical with the MP3 decoding device used in described step 3.-1, and the coding bit rate of the MP3 scrambler of described step 4. in-1 is identical with the coding bit rate of the MP3 scrambler in described step 3.-1, described step is 4.-2 all identical with the coding bit rate of the MP3 scrambler in described step 3.-2 with the coding bit rate of the MP3 scrambler in described step 4.-3.If the coding bit rate br2 of the MP3 scrambler of step 3. in-1
1for 64kbps, the coding bit rate br2 of the MP3 scrambler of step 3. in-2
2for 128kbps, then the coding bit rate br3 of the MP3 scrambler of step 4. in-1
1for 64kbps, step 4.-2 and the coding bit rate br3 of the MP3 scrambler of step 4. in-3
2and br3
3be 128kbps.
5. each first compression MP3 audio sample is labeled as-1, each second-compressed MP3 audio sample is labeled as 0, each three second compression MP3 audio sample are labeled as 1, again all first compression MP3 audio sample, all second-compressed MP3 audio sample and three all second compression MP3 audio sample are formed a training sample set, wherein, each subsample in training sample set is first compression MP3 audio sample or is second-compressed MP3 audio sample or be three second compression MP3 audio sample.
6. 47 eigenwerts of each subsample in training sample set are extracted, and 47 eigenwerts of each subsample in training sample set are formed a row vector, the row vector that 47 eigenwerts of the subsample of the kth in training sample set are formed is designated as F
k; Then adopt 47 eigenwerts of existing min-max method for normalizing to each subsample in training sample set to be normalized, obtain the eigenwert after 47 normalizeds of each subsample in training sample set; Wherein, the initial value of k is 1,1≤k≤K, K represent total number of the subsample comprised in training sample set, K=3N.
In this particular embodiment, step 6. middle F
kacquisition process be:
6. be-1, long window coded frame by the frame definition utilizing long window coded system to carry out processing in the subsample of the kth in training sample set; Then the MP3 decoding device utilizing step 3. to use in-1 carries out decoding process to the every frame in the subsample of the kth in training sample set, obtain the kth decoded WAV audio frequency of sub-sample in training sample set, in decoding process, extract the position of the every frame length window coded frame in the kth subsample in training sample set; Then according to the position of the every frame length window coded frame in the subsample of the kth in training sample set, obtain the scale factor matrix of the position of all long window coded frame in the kth subsample in training sample set, be designated as sf
a, wherein, if the kth subsample in training sample set is monophonic audio, then sf
adimension be 2w × 21, if the kth subsample in training sample set is dual-channel audio, then sf
adimension be 4w × 21, w represents the totalframes of the long window coded frame in the kth subsample in training sample set, w>=1.
6. the MP3 scrambler used in-2, utilizing step 2. carries out coded treatment to the decoded WAV audio frequency obtained in step 6.-1, obtain new MP3 audio frequency, in coding process, according to the position of the every frame length window coded frame in the kth subsample in the training sample set extracted in step 6.-1, the scale factor matrix of the position of all long window coded frame of decoded WAV audio frequency in coding process that 6. obtaining step obtains in-1, is designated as sf
bwherein, the coding bit rate of the MP3 scrambler in this step is identical with the coding bit rate of the MP3 scrambler that compressed encoding last in the acquisition process of the subsample of the kth in training sample set uses, if the kth subsample in training sample set is monophonic audio, then sf
bdimension be 2w × 21, if the kth subsample in training sample set is dual-channel audio, then sf
bdimension be 4w × 21.
6.-3, sf is calculated
ain scale factor 1 to 5 shift as sf
bin the transition probability matrix of scale factor 0 to 3, be designated as P, wherein, the dimension of P is the value of the 1st row the 1st column element in 5 × 4, P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 0, the value of the 1st row the 2nd column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 1, the value of the 1st row the 3rd column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 2, the value of the 1st row the 4th column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 3, the like, the value of the 5th row the 4th column element in P is sf
ain scale factor 5 shift as sf
bin the transition probability of scale factor 3; Then using the value of the element of 20 in P according to the order of sequence as front 20 eigenwerts of the subsample of the kth in training sample set.
At this, sf
ain scale factor transfer for sf
bin the transition probability of scale factor obtain by existing single step Markov Transition Probabilities.
6.-4, sf is calculated
awith sf
bmatrix of differences, be designated as Δ sf, Δ sf=sf
a-sf
b, i.e. sf
ain the value of each element and sf
bin the value one_to_one corresponding of each element subtract each other, obtain the value of each element in Δ sf; Then the average of the value of all elements in Δ sf is calculated, using the 21st eigenwert of this average as the subsample of the kth in training sample set.
6.-5, sf is calculated
ain the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
ain the probability distribution of scale factor 0,
in the value of the 2nd element be sf
ain the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
ain the probability distribution of scale factor 8.
And calculate sf
bin the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
bin the probability distribution of scale factor 0,
in the value of the 2nd element be sf
bin the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
bin the probability distribution of scale factor 8.
At this, sf
ain the probability distribution of scale factor and sf
bin the probability distribution of scale factor all obtained by existing probabilistic method.
6.-6, calculate
with
difference value vector, be designated as Δ P,
Δ P represents the probability of use incremental vector of scale factor 0 to 8, and the value of the 1st element in Δ P is
in the 1st element value with
in the difference of value of the 1st element, the value of the 1st element in Δ P represents the probability of use increment of scale factor 0, the like, the value of the 9th element in Δ P is
in the 9th element value with
in the difference of value of the 9th element, the value of the 9th element in Δ P represents the probability of use increment of scale factor 8; Then using the value of the element of 9 in Δ P according to the order of sequence as the 22nd eigenwert of the subsample of the kth in training sample set to the 30th eigenwert.
6.-7, calculate
with
quotient vector, be designated as dP,
dP represents the probability of use increased times vector of scale factor 0 to 8, and the value of the 1st element in dP is
in the 1st element value divided by
in the quotient that obtains of the value of the 1st element, the value of the 1st element in dP represents the probability of use increased times of scale factor 0, the like, the value of the 9th element in dP is
in the 9th element value divided by
in the quotient that obtains of the value of the 9th element, the value of the 9th element in dP represents the probability of use increased times of scale factor 8; Then using the value of the element of 9 in dP according to the order of sequence as the 31st eigenwert of the subsample of the kth in training sample set to the 39th eigenwert.
-8 6., the difference of the probability of use increased times of scale factor 0 and the respective probability of use increased times of scale factor 1 to 8 is calculated, using 8 differences obtaining according to the order of sequence as the 40th eigenwert of the subsample of the kth in training sample set to the 47th eigenwert.
6.-9,47 eigenwerts of the subsample of the kth in training sample set are formed row vector F
k.
7. utilize existing LibSVM sorter to train the eigenwert after respective 47 normalizeds in all subsamples in training sample set, obtain training template; Wherein, adopt cross validation mode [2 in training process
-5, 2
5] choose best penalty parameter c and best RBF nuclear parameter g in interval, all the other parameters all Use Defaults.At this, the eigenwert after normalized is trained, testing result can be made more accurate.
8. M MP3 audio frequency to be detected is chosen arbitrarily from network or in electronic record, wherein, M >=1, each MP3 audio frequency to be detected is first compression MP3 audio frequency or is second-compressed MP3 audio frequency or be three second compression MP3 audio frequency, and in the coding bit rate of MP3 scrambler that in the acquisition process of each MP3 audio frequency to be detected, last compressed encoding uses and the acquisition process of the subsample in training sample set, the coding bit rate of MP3 scrambler that uses of compressed encoding is identical for the last time; Then according to step 6. in extract the process of 47 eigenwerts of each subsample in training sample set, extract 47 eigenwerts of each MP3 audio frequency to be detected in an identical manner; Then adopt 47 eigenwerts of existing min-max method for normalizing to each MP3 audio frequency to be detected to be normalized, obtain the eigenwert after 47 normalizeds of each MP3 audio frequency to be detected; Again the eigenwert after 47 normalizeds of each MP3 audio frequency to be detected is input in training template and detects, if the Output rusults of training template is-1, then determine that corresponding MP3 audio frequency to be detected is first compression MP3 audio frequency, if the Output rusults of training template is 0, then determine that corresponding MP3 audio frequency to be detected is second-compressed MP3 audio frequency, if the Output rusults of training template is 1, then determine that corresponding MP3 audio frequency to be detected is three second compression MP3 audio frequency.
In this particular embodiment, step 8. in the leaching process of 47 eigenwerts of a MP3 audio frequency to be detected be:
8. be-1, long window coded frame by the frame definition utilizing long window coded system to carry out processing in MP3 audio frequency to be detected; Then the MP3 decoding device utilizing step 3. to use in-1 carries out decoding process to the every frame in MP3 audio frequency to be detected, obtain the WAV audio frequency after MP3 audio decoder to be detected, in decoding process, extract the position of the every frame length window coded frame in MP3 audio frequency to be detected; Then according to the position of the every frame length window coded frame in MP3 audio frequency to be detected, obtain the scale factor matrix of the position of all long window coded frame in MP3 audio frequency to be detected, be designated as sf
a', wherein, if MP3 audio frequency to be detected is monophonic audio, then sf
a' dimension be 2w' × 21, if MP3 audio frequency to be detected is dual-channel audio, then sf
a' dimension be 4w' × 21, w' represents the totalframes of the long window coded frame in MP3 audio frequency to be detected, w'>=1.
8. the MP3 scrambler used in-2, utilizing step 2. carries out coded treatment to the decoded WAV audio frequency obtained in step 8.-1, obtain new MP3 audio frequency, in coding process, according to the position of the every frame length window coded frame in the MP3 audio frequency to be detected extracted in step 8.-1, the scale factor matrix of the position of all long window coded frame of decoded WAV audio frequency in coding process that 8. obtaining step obtains in-1, is designated as sf
b', wherein, the coding bit rate of the MP3 scrambler in this step is identical with the coding bit rate of the MP3 scrambler that compressed encoding last in the acquisition process of MP3 audio frequency to be detected uses, if MP3 audio frequency to be detected is monophonic audio, then sf
b' dimension be 2w' × 21, if MP3 audio frequency to be detected is dual-channel audio, then sf
b' dimension be 4w' × 21.
8.-3, sf is calculated
a' in scale factor 1 to 5 shift as sf
b' in the transition probability matrix of scale factor 0 to 3, be designated as P', wherein, the dimension of P' is the value of the 1st row the 1st column element in 5 × 4, P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 0, the value of the 1st row the 2nd column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 1, the value of the 1st row the 3rd column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 2, the value of the 1st row the 4th column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 3, the like, the value of the 5th row the 4th column element in P' is sf
a' in scale factor 5 shift as sf
b' in the transition probability of scale factor 3; Then using the value of the element of 20 in P' according to the order of sequence as front 20 eigenwerts of MP3 audio frequency to be detected.
8.-4, sf is calculated
a' and sf
b' matrix of differences, be designated as Δ sf', Δ sf'=sf
a'-sf
b', i.e. sf
a' in the value of each element and sf
b' in the value one_to_one corresponding of each element subtract each other, obtain the value of each element in Δ sf'; Then the average of the value of all elements in Δ sf' is calculated, using the 21st eigenwert of this average as MP3 audio frequency to be detected;
8.-5, sf is calculated
a' in the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
a' in the probability distribution of scale factor 0,
in the value of the 2nd element be sf
a' in the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
a' in the probability distribution of scale factor 8.
And calculate sf
b' in the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
b' in the probability distribution of scale factor 0,
in the value of the 2nd element be sf
b' in the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
b' in the probability distribution of scale factor 8.
8.-6, calculate
with
difference value vector, be designated as Δ P',
Δ P' represents the probability of use incremental vector of scale factor 0 to 8, and the value of the 1st element in Δ P' is
in the 1st element value with
in the difference of value of the 1st element, the value of the 1st element in Δ P' represents the probability of use increment of scale factor 0, the like, the value of the 9th element in Δ P' is
in the 9th element value with
in the difference of value of the 9th element, the value of the 9th element in Δ P' represents the probability of use increment of scale factor 8; Then using the value of the element of 9 in Δ P' according to the order of sequence as the 22nd eigenwert of MP3 audio frequency to be detected to the 30th eigenwert.
8.-7, calculate
with
quotient vector, be designated as dP',
dP' represents the probability of use increased times vector of scale factor 0 to 8, and the value of the 1st element in dP' is
in the 1st element value divided by
in the quotient that obtains of the value of the 1st element, the value of the 1st element in dP' represents the probability of use increased times of scale factor 0, the like, the value of the 9th element in dP' is
in the 9th element value divided by
in the quotient that obtains of the value of the 9th element, the value of the 9th element in dP' represents the probability of use increased times of scale factor 8; Then using the value of the element of 9 in dP' according to the order of sequence as the 31st eigenwert of MP3 audio frequency to be detected to the 39th eigenwert.
-8 8., the difference of the probability of use increased times of scale factor 0 and the respective probability of use increased times of scale factor 1 to 8 is calculated, using 8 differences obtaining according to the order of sequence as the 40th eigenwert of MP3 audio frequency to be detected to the 47th eigenwert.
In the present embodiment in steps in MP3 scrambler and MP3 decoding device all adopt at present more popular lame3.99.5, as long as experimenter by arranging the switching that different parameters can realize MP3 scrambler and MP3 decoding device on lame3.99.5.
For further illustrating feasibility and the validity of the inventive method, test.
In the present embodiment, MP3 scrambler adopts 64kbps, 80kbps, 96kbps, 112kbps, 128kbps, 160kbps, 192kbps seven kinds of coding bit rates to carry out selected 539 first unpressed WAV audio sample compressing the first compression MP3 audio sample obtaining correspondence successively, choose mode again according to the coding bit rate of the MP3 scrambler limited, obtain corresponding second-compressed MP3 audio sample and three second compression MP3 audio sample.49 groups of experiments are comprised altogether in the present embodiment, often organize experiment sample set and comprise 539 first compression MP3 audio sample, 539 second-compressed MP3 audio sample and 539 three second compression MP3 audio sample, and extract 47 eigenwerts of each sample and it is normalized.Train classification models is used for by 70% of the eigenwert after the normalized of all first compression MP3 audio sample often organized in experiment, eigenwert after the normalized of all second-compressed MP3 audio sample 70% for the eigenwert after the normalized of train classification models and all three second compression MP3 audio sample 70% for train classification models, be input in LibSVM sorter, by remaining 30% of the eigenwert after the normalized of all first compression MP3 audio sample, remaining 30% of eigenwert after the normalized of all second-compressed MP3 audio sample, eigenwert after the normalized of all three second compression MP3 audio sample remaining 30% for test, often organize experiment test and carry out 10 times.Finally by asking the mean value of the testing result of 10 times to calculate the predictablity rate often organizing experiment test, the result often organizing the predictablity rate of experiment test is as shown in table 1.
The result of the predictablity rate of experiment test often organized by table 1
In table 1, when BR1, BR2 and BR3 represent first time, the second time of three second compression MP3 audio sample respectively and compress for the third time, institute adopts coding bit rate.Second data representation of such as the first row is through 80kbps coding bit rate first compression MP3 sample, and the second-compressed MP3 sample through 64kbps and 80kbps coding bit rate twice is 81.85% with the predictablity rate through the MP3 sample of 64kbps, 80kbps and 80kbps coding bit rate three second compression.As can be seen from Table 1 under different coding bit rates, compression histories verification and measurement ratio is better, under being particularly useful for higher coding bit rate.
To sum up, MP3 audio compression history detection method of the present invention has validity, breaches the blank that MP3 audio compression history detects, and under being applicable to high code rate, MP3 audio compression history detects, and computation complexity is lower.
Claims (3)
1. a MP3 audio compression history detection method, is characterized in that comprising the following steps:
1. N number of style difference is chosen and unpressed WAV audio sample, wherein, N >=10;
2. the first compression MP3 audio sample corresponding with each unpressed WAV audio sample is obtained, detailed process is: utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression;
3. the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample is obtained, 3.-1 detailed process is:, utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression, each first compression MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device;-2 3., utilize MP3 scrambler to carry out compressed encoding to each WAV audio sample that solution in step 3.-1 is pressed into, obtain the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample;
Wherein, 3.-1 all 2. the middle MP3 scrambler used is identical with described step with the MP3 scrambler used in described step 3.-2 for described step, and the coding bit rate of the MP3 scrambler of described step 3. in-1 and described step 2. in the coding bit rate of MP3 scrambler identical or not identical, the coding bit rate of the MP3 scrambler of described step 3. in-2 and described step 2. in the coding bit rate of MP3 scrambler identical;
4. the three second compression MP3 audio sample corresponding with each unpressed WAV audio sample are obtained, 4.-1 detailed process is:, utilize MP3 scrambler to carry out compressed encoding to each unpressed WAV audio sample, obtain the first compression MP3 audio sample that each unpressed WAV audio sample is formed after first compression, each first compression MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device; 4.-2, MP3 scrambler is utilized to carry out compressed encoding to each WAV audio sample that solution in step 4.-1 is pressed into, obtain the second-compressed MP3 audio sample corresponding with each unpressed WAV audio sample, each second-compressed MP3 audio sample solution obtained is pressed into WAV audio sample by recycling MP3 decoding device;-3 4., utilize MP3 scrambler to carry out compressed encoding to each WAV audio sample that solution in step 4.-2 is pressed into, obtain the three second compression MP3 audio sample corresponding with each unpressed WAV audio sample;
Wherein, described step 4.-1, 4.-2 all 2. the middle MP3 scrambler used is identical with described step with the MP3 scrambler used in described step 4.-3 for described step, described step 4.-1 with in described step 4.-2 use MP3 decoding device all identical with the MP3 decoding device used in described step 3.-1, and the coding bit rate of the MP3 scrambler of described step 4. in-1 is identical with the coding bit rate of the MP3 scrambler in described step 3.-1, described step is 4.-2 all identical with the coding bit rate of the MP3 scrambler in described step 3.-2 with the coding bit rate of the MP3 scrambler in described step 4.-3,
5. each first compression MP3 audio sample is labeled as-1, each second-compressed MP3 audio sample is labeled as 0, each three second compression MP3 audio sample are labeled as 1, again all first compression MP3 audio sample, all second-compressed MP3 audio sample and three all second compression MP3 audio sample are formed a training sample set, wherein, each subsample in training sample set is first compression MP3 audio sample or is second-compressed MP3 audio sample or be three second compression MP3 audio sample;
6. 47 eigenwerts of each subsample in training sample set are extracted, and 47 eigenwerts of each subsample in training sample set are formed a row vector, the row vector that 47 eigenwerts of the subsample of the kth in training sample set are formed is designated as F
k; Then adopt 47 eigenwerts of min-max method for normalizing to each subsample in training sample set to be normalized, obtain the eigenwert after 47 normalizeds of each subsample in training sample set; Wherein, the initial value of k is 1,1≤k≤K, K represent total number of the subsample comprised in training sample set, K=3N;
7. utilize LibSVM sorter to train the eigenwert after respective 47 normalizeds in all subsamples in training sample set, obtain training template; Wherein, adopt cross validation mode [2 in training process
-5, 2
5] choose best penalty parameter c and best RBF nuclear parameter g in interval, all the other parameters all Use Defaults;
8. M MP3 audio frequency to be detected is chosen arbitrarily, wherein, M >=1, each MP3 audio frequency to be detected is first compression MP3 audio frequency or is second-compressed MP3 audio frequency or be three second compression MP3 audio frequency, and in the coding bit rate of MP3 scrambler that in the acquisition process of each MP3 audio frequency to be detected, last compressed encoding uses and the acquisition process of the subsample in training sample set, the coding bit rate of MP3 scrambler that uses of compressed encoding is identical for the last time; Then according to step 6. in extract the process of 47 eigenwerts of each subsample in training sample set, extract 47 eigenwerts of each MP3 audio frequency to be detected in an identical manner; Then adopt 47 eigenwerts of min-max method for normalizing to each MP3 audio frequency to be detected to be normalized, obtain the eigenwert after 47 normalizeds of each MP3 audio frequency to be detected; Again the eigenwert after 47 normalizeds of each MP3 audio frequency to be detected is input in training template and detects, if the Output rusults of training template is-1, then determine that corresponding MP3 audio frequency to be detected is first compression MP3 audio frequency, if the Output rusults of training template is 0, then determine that corresponding MP3 audio frequency to be detected is second-compressed MP3 audio frequency, if the Output rusults of training template is 1, then determine that corresponding MP3 audio frequency to be detected is three second compression MP3 audio frequency.
2. a kind of MP3 audio compression history detection method according to claim 1, is characterized in that described step 6. middle F
kacquisition process be:
6. be-1, long window coded frame by the frame definition utilizing long window coded system to carry out processing in the subsample of the kth in training sample set; Then the MP3 decoding device utilizing step 3. to use in-1 carries out decoding process to the every frame in the subsample of the kth in training sample set, obtain the kth decoded WAV audio frequency of sub-sample in training sample set, in decoding process, extract the position of the every frame length window coded frame in the kth subsample in training sample set; Then according to the position of the every frame length window coded frame in the subsample of the kth in training sample set, obtain the scale factor matrix of the position of all long window coded frame in the kth subsample in training sample set, be designated as sf
a, wherein, if the kth subsample in training sample set is monophonic audio, then sf
adimension be 2w × 21, if the kth subsample in training sample set is dual-channel audio, then sf
adimension be 4w × 21, w represents the totalframes of the long window coded frame in the kth subsample in training sample set;
6. the MP3 scrambler used in-2, utilizing step 2. carries out coded treatment to the decoded WAV audio frequency obtained in step 6.-1, obtain new MP3 audio frequency, in coding process, according to the position of the every frame length window coded frame in the kth subsample in the training sample set extracted in step 6.-1, the scale factor matrix of the position of all long window coded frame of decoded WAV audio frequency in coding process that 6. obtaining step obtains in-1, is designated as sf
bwherein, the coding bit rate of the MP3 scrambler in this step is identical with the coding bit rate of the MP3 scrambler that compressed encoding last in the acquisition process of the subsample of the kth in training sample set uses, if the kth subsample in training sample set is monophonic audio, then sf
bdimension be 2w × 21, if the kth subsample in training sample set is dual-channel audio, then sf
bdimension be 4w × 21;
6.-3, sf is calculated
ain scale factor 1 to 5 shift as sf
bin the transition probability matrix of scale factor 0 to 3, be designated as P, wherein, the dimension of P is the value of the 1st row the 1st column element in 5 × 4, P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 0, the value of the 1st row the 2nd column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 1, the value of the 1st row the 3rd column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 2, the value of the 1st row the 4th column element in P is sf
ain scale factor 1 shift as sf
bin the transition probability of scale factor 3, the like, the value of the 5th row the 4th column element in P is sf
ain scale factor 5 shift as sf
bin the transition probability of scale factor 3; Then using the value of the element of 20 in P according to the order of sequence as front 20 eigenwerts of the subsample of the kth in training sample set;
6.-4, sf is calculated
awith sf
bmatrix of differences, be designated as Δ sf, Δ sf=sf
a-sf
b; Then the average of the value of all elements in Δ sf is calculated, using the 21st eigenwert of this average as the subsample of the kth in training sample set;
6.-5, sf is calculated
ain the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
ain the probability distribution of scale factor 0,
in the value of the 2nd element be sf
ain the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
ain the probability distribution of scale factor 8;
And calculate sf
bin the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
bin the probability distribution of scale factor 0,
in the value of the 2nd element be sf
bin the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
bin the probability distribution of scale factor 8;
6.-6, calculate
with
difference value vector, be designated as Δ P,
Δ P represents the probability of use incremental vector of scale factor 0 to 8; Then using the value of the element of 9 in Δ P according to the order of sequence as the 22nd eigenwert of the subsample of the kth in training sample set to the 30th eigenwert;
6.-7, calculate
with
quotient vector, be designated as dP,
dP represents the probability of use increased times vector of scale factor 0 to 8; Then using the value of the element of 9 in dP according to the order of sequence as the 31st eigenwert of the subsample of the kth in training sample set to the 39th eigenwert;
-8 6., the difference of the probability of use increased times of scale factor 0 and the respective probability of use increased times of scale factor 1 to 8 is calculated, using 8 differences obtaining according to the order of sequence as the 40th eigenwert of the subsample of the kth in training sample set to the 47th eigenwert;
6.-9,47 eigenwerts of the subsample of the kth in training sample set are formed row vector F
k.
3. a kind of MP3 audio compression history detection method according to claim 1 and 2, is characterized in that the leaching process of 47 eigenwerts of a MP3 audio frequency to be detected during described step is 8. is:
8. be-1, long window coded frame by the frame definition utilizing long window coded system to carry out processing in MP3 audio frequency to be detected; Then the MP3 decoding device utilizing step 3. to use in-1 carries out decoding process to the every frame in MP3 audio frequency to be detected, obtain the WAV audio frequency after MP3 audio decoder to be detected, in decoding process, extract the position of the every frame length window coded frame in MP3 audio frequency to be detected; Then according to the position of the every frame length window coded frame in MP3 audio frequency to be detected, obtain the scale factor matrix of the position of all long window coded frame in MP3 audio frequency to be detected, be designated as sf
a', wherein, if MP3 audio frequency to be detected is monophonic audio, then sf
a' dimension be 2w' × 21, if MP3 audio frequency to be detected is dual-channel audio, then sf
a' dimension be 4w' × 21, w' represents the totalframes of the long window coded frame in MP3 audio frequency to be detected;
8. the MP3 scrambler used in-2, utilizing step 2. carries out coded treatment to the decoded WAV audio frequency obtained in step 8.-1, obtain new MP3 audio frequency, in coding process, according to the position of the every frame length window coded frame in the MP3 audio frequency to be detected extracted in step 8.-1, the scale factor matrix of the position of all long window coded frame of decoded WAV audio frequency in coding process that 8. obtaining step obtains in-1, is designated as sf
b', wherein, the coding bit rate of the MP3 scrambler in this step is identical with the coding bit rate of the MP3 scrambler that compressed encoding last in the acquisition process of MP3 audio frequency to be detected uses, if MP3 audio frequency to be detected is monophonic audio, then sf
b' dimension be 2w' × 21, if MP3 audio frequency to be detected is dual-channel audio, then sf
b' dimension be 4w' × 21;
8.-3, sf is calculated
a' in scale factor 1 to 5 shift as sf
b' in the transition probability matrix of scale factor 0 to 3, be designated as P', wherein, the dimension of P' is the value of the 1st row the 1st column element in 5 × 4, P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 0, the value of the 1st row the 2nd column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 1, the value of the 1st row the 3rd column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 2, the value of the 1st row the 4th column element in P' is sf
a' in scale factor 1 shift as sf
b' in the transition probability of scale factor 3, the like, the value of the 5th row the 4th column element in P' is sf
a' in scale factor 5 shift as sf
b' in the transition probability of scale factor 3; Then using the value of the element of 20 in P' according to the order of sequence as front 20 eigenwerts of MP3 audio frequency to be detected;
8.-4, sf is calculated
a' and sf
b' matrix of differences, be designated as Δ sf', Δ sf'=sf
a'-sf
b'; Then the average of the value of all elements in Δ sf' is calculated, using the 21st eigenwert of this average as MP3 audio frequency to be detected;
8.-5, sf is calculated
a' in the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
a' in the probability distribution of scale factor 0,
in the value of the 2nd element be sf
a' in the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
a' in the probability distribution of scale factor 8;
And calculate sf
b' in the ProbabilityDistribution Vector of scale factor 0 to 8, be designated as
wherein,
dimension be 1 × 9,
in the value of the 1st element be sf
b' in the probability distribution of scale factor 0,
in the value of the 2nd element be sf
b' in the probability distribution of scale factor 1, the like,
in the value of the 9th element be sf
b' in the probability distribution of scale factor 8;
8.-6, calculate
with
difference value vector, be designated as Δ P',
Δ P' represents the probability of use incremental vector of scale factor 0 to 8; Then using the value of the element of 9 in Δ P' according to the order of sequence as the 22nd eigenwert of MP3 audio frequency to be detected to the 30th eigenwert;
8.-7, calculate
with
quotient vector, be designated as dP',
dP' represents the probability of use increased times vector of scale factor 0 to 8; Then using the value of the element of 9 in dP' according to the order of sequence as the 31st eigenwert of MP3 audio frequency to be detected to the 39th eigenwert;
-8 8., the difference of the probability of use increased times of scale factor 0 and the respective probability of use increased times of scale factor 1 to 8 is calculated, using 8 differences obtaining according to the order of sequence as the 40th eigenwert of MP3 audio frequency to be detected to the 47th eigenwert.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510419213.2A CN105070297B (en) | 2015-07-16 | 2015-07-16 | A kind of MP3 audio compressions history detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510419213.2A CN105070297B (en) | 2015-07-16 | 2015-07-16 | A kind of MP3 audio compressions history detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105070297A true CN105070297A (en) | 2015-11-18 |
| CN105070297B CN105070297B (en) | 2018-10-23 |
Family
ID=54499650
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510419213.2A Active CN105070297B (en) | 2015-07-16 | 2015-07-16 | A kind of MP3 audio compressions history detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105070297B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105513600A (en) * | 2016-01-13 | 2016-04-20 | 宁波大学 | Duration-related mp3 double compression detection method under same bit rate |
| CN105845132A (en) * | 2016-03-22 | 2016-08-10 | 宁波大学 | Coding parameter statistical feature-based AAC sound recording document source identification method |
| CN105913855A (en) * | 2016-04-11 | 2016-08-31 | 宁波大学 | Long window scaling factor-based playback voice attack detection algorithm |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020194403A1 (en) * | 2001-06-15 | 2002-12-19 | Khein-Seng Pua | Audio video signal processing device |
| CN102097098A (en) * | 2011-02-25 | 2011-06-15 | 宁波大学 | Digital steganography and digital extraction methods with compressed audio as masking carrier |
| CN103106061A (en) * | 2013-03-05 | 2013-05-15 | 北京车音网科技有限公司 | Voice input method and device |
| CN103366753A (en) * | 2013-06-28 | 2013-10-23 | 宁波大学 | Moving picture experts group audio layer-3 (MP3) audio double-compression detection method under same code rate |
| CN104123935A (en) * | 2014-07-16 | 2014-10-29 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN104282310A (en) * | 2014-09-26 | 2015-01-14 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
-
2015
- 2015-07-16 CN CN201510419213.2A patent/CN105070297B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020194403A1 (en) * | 2001-06-15 | 2002-12-19 | Khein-Seng Pua | Audio video signal processing device |
| CN102097098A (en) * | 2011-02-25 | 2011-06-15 | 宁波大学 | Digital steganography and digital extraction methods with compressed audio as masking carrier |
| CN103106061A (en) * | 2013-03-05 | 2013-05-15 | 北京车音网科技有限公司 | Voice input method and device |
| CN103366753A (en) * | 2013-06-28 | 2013-10-23 | 宁波大学 | Moving picture experts group audio layer-3 (MP3) audio double-compression detection method under same code rate |
| CN104123935A (en) * | 2014-07-16 | 2014-10-29 | 武汉大学 | Double compression detection method towards MP3 (moving picture experts group audio Layer-3) digital audio file |
| CN104282310A (en) * | 2014-09-26 | 2015-01-14 | 宁波大学 | Steganography detection method for audio subjected to MP3Stego steganography |
Non-Patent Citations (3)
| Title |
|---|
| LIU Q. , SUNG A. H. , QIAO M.: "Detection of Double MP3 Compression", 《COGNITIVE COMPUTATION》 * |
| 余先敏: "基于相同压缩速率下的MP3双压缩检测方法", 《计算机工程与应用》 * |
| 马朋飞: "音频二次压缩分析方法与检测技术研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105513600A (en) * | 2016-01-13 | 2016-04-20 | 宁波大学 | Duration-related mp3 double compression detection method under same bit rate |
| CN105513600B (en) * | 2016-01-13 | 2019-02-05 | 宁波大学 | A kind of bis- compressed detected methods of same code rate mp3 relevant to duration |
| CN105845132A (en) * | 2016-03-22 | 2016-08-10 | 宁波大学 | Coding parameter statistical feature-based AAC sound recording document source identification method |
| CN105913855A (en) * | 2016-04-11 | 2016-08-31 | 宁波大学 | Long window scaling factor-based playback voice attack detection algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105070297B (en) | 2018-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104282310B (en) | Steganography detection method for audio subjected to MP3Stego steganography | |
| CN109325019B (en) | Data association relationship network construction method | |
| CN113435424B (en) | Method and system for identifying destroying granularity of confidential medium | |
| CN103366753B (en) | Moving picture experts group audio layer-3 (MP3) audio double-compression detection method under same code rate | |
| CN105070297A (en) | MP3 audio compression history detection method | |
| CN105609116A (en) | Speech emotional dimensions region automatic recognition method | |
| CN111291660A (en) | Anchor-free traffic sign identification method based on void convolution | |
| CN116185971B (en) | Intelligent processing system for electronic pressure weighing data | |
| CN105869658A (en) | Voice endpoint detection method employing nonlinear feature | |
| CN105070299A (en) | Hi-Fi tone quality identifying method based on pattern recognition | |
| CN113640380A (en) | Multi-stage classification method and system for rail damage detection | |
| CN110399613B (en) | Method and system for identifying internet news related to place names based on part-of-speech tagging | |
| CN103871405A (en) | AMR audio authenticating method | |
| CN103236265B (en) | A kind of Stego-detection method for MP3Stegz | |
| CN113703773B (en) | NLP-based binary code similarity comparison method | |
| CN105808602B (en) | Method and device for detecting junk information | |
| CN105072444B (en) | A kind of HEVC video second-compressed detection methods under different quantization parameters | |
| CN104637484A (en) | MP3 audio steganography detection method based on co-occurrence matrix analysis | |
| CN116204830A (en) | Welding abnormality real-time detection method based on path aggregation network | |
| CN115508615A (en) | Load transient characteristic extraction method based on induction motor | |
| CN104462458A (en) | Data mining method of big data system | |
| CN114155544A (en) | Wireless form identification method and device, computer equipment and storage medium | |
| CN101667423A (en) | Compressed domain high robust voice/music dividing method based on probability density ratio | |
| CN116453514B (en) | Multi-view-based voice keyword detection and positioning method and device | |
| CN113963357B (en) | Knowledge graph-based sensitive text detection method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |