CN105632516B - A kind of MP3 recording file source title method based on side information statistical property - Google Patents
A kind of MP3 recording file source title method based on side information statistical property Download PDFInfo
- Publication number
- CN105632516B CN105632516B CN201610020274.6A CN201610020274A CN105632516B CN 105632516 B CN105632516 B CN 105632516B CN 201610020274 A CN201610020274 A CN 201610020274A CN 105632516 B CN105632516 B CN 105632516B
- Authority
- CN
- China
- Prior art keywords
- parameter
- mobile phone
- brand
- side information
- recording file
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 239000002245 particle Substances 0.000 claims description 59
- 238000012549 training Methods 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 6
- 241000208340 Araliaceae Species 0.000 claims description 5
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 5
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 5
- 235000008434 ginseng Nutrition 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 230000002411 adverse Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 9
- DWDGSKGGUZPXMQ-UHFFFAOYSA-N OPPO Chemical compound OPPO DWDGSKGGUZPXMQ-UHFFFAOYSA-N 0.000 description 31
- 238000009826 distribution Methods 0.000 description 28
- 238000010586 diagram Methods 0.000 description 14
- 238000001514 detection method Methods 0.000 description 12
- 238000013139 quantization Methods 0.000 description 11
- 238000005516 engineering process Methods 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 5
- 230000006835 compression Effects 0.000 description 4
- 238000007906 compression Methods 0.000 description 4
- 230000004069 differentiation Effects 0.000 description 4
- 230000005236 sound signal Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000008187 granular material Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 101000969688 Homo sapiens Macrophage-expressed gene 1 protein Proteins 0.000 description 1
- 102100021285 Macrophage-expressed gene 1 protein Human genes 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Signal Processing For Digital Recording And Reproducing (AREA)
Abstract
The MP3 recording file source title method based on side information statistical property that the invention discloses a kind of, its mentality of designing is by analyzing MP3 recording file code stream, namely in frame in side information each parameter handling characteristics and statistical property, different series model mobile phone is therefrom found out when generating MP3 recording file using the tendency or feature of these parameters, to realize the identification to MP3 recording file source, distinguish that given mp3 file is that the mobile phone of which kind of serial model No. is recorded, this recognition methods has feature set building process relatively simple quickly, algorithm complexity is low, recognition accuracy is high, real-time is good, the advantages that convenient for operating.
Description
Technical field
The present invention relates to the recognition methods in MP3 recording file source, in particular to a kind of based on side information statistical property
MP3 recording file source title method.
Background technique
With the arriving of big data era, digital multimedia presentation increases explosively.But meanwhile multi-medium data magnitude
Growth with frequency of use also promotes authoring tool to be developed rapidly, this allow for multi-medium data forgery and
It distorts and becomes increasingly easy.Just because of this, is flooded with a large amount of untrue, incredible multimedia number in our life
According to.In order to verify primitiveness, the authenticity and integrity of multi-medium data, multi-media forensic technology is come into being.Currently, domestic
Image domains are concentrated mainly on to the research of digital multimedia evidence obtaining outside, and for the research of digital audio forensic technologies starting phase
To important component that is later, but collecting evidence as multimedia, actual demand is still constantly increasing, also by increasingly
More concerns.
The identification of audio source is the first step work of audio forensics, and the purpose is to the primitiveness to audio to verify, by
In the quick universal of smart phone, the behavior of recording becomes increasingly to facilitate.In daily life, people are more willing to use hand
Machine this moment portable equipment removes record sound;Therefore, identify a recording file from which kind of brand and model
Mobile phone is a hot issue in current audio source evidence forensics field, and also has a small amount of scholar and research group in recent years
Team has made some research work in this respect.Such as C. Hanilci et al. uses MFCC cepstrum (MFCC) as feature pair
The recording file of 14 different model mobile phones distinguishes, and the Detection accuracy of this method has reached 96.42%.On this basis,
They compared 4 kinds of acoustic features including MFCC, and (in addition 3 kinds are respectively linear cepstrum coefficient LFCC, Bark cepstrum system
Number BFCC and linear prediction residue error LPCC) performance in terms of equipment source identification.According to comparing result, they still recognize
It is best feature for MFCC.C. L. Kotropoulos by using rarefaction representation to 2049 dimension logarithm language spectrum signature and
The Gauss super vector feature of 2816 dimensions is successfully made dimensionality reduction, and using 850 peacekeeping, 120 dimensional feature after corresponding dimensionality reduction to 21 kinds
The mobile phone of different model is classified, and using 3 kinds of different classifications devices, the former is more than at Detection accuracy
94.84%, the latter is then 98.41%-100%.
Although these methods all achieve preferable recognition effect in terms of audio source device identification, according to our institutes
Know, realizes that the identification of mobile phone source has not been reported using the code flow structure and coding parameter characteristic of recording file.And at present absolutely
The recording format of most of smart phone defaults is compressed format, and compression standard is mainly MP3 and AAC;In addition, different factories
Family, even identical producer production different model equipment, the hardware and software part of audio-frequency module difference, compression
The specific implementation of algorithm and also have the characteristics that with the cooperation of hardware respective, this results in different brands model mobile phone to pickup
When voice signal carries out compressed encoding, there is differences for selection and use to various coding parameters, and usually these coding ginsengs
Number is all stored in the side information in the every frame of compressed format audio;Therefore, MP3 recording text is identified based on side information statistical property
Part source is undoubtedly a kind of very reliable recognition methods.
Summary of the invention
The technical problem to be solved by the present invention lies in overcoming the defects of the prior art and provides a kind of feature set and constructed
Journey is relatively simple quickly, algorithm complexity is low, recognition accuracy is high, real-time is good, is united convenient for one kind of operation based on side information
Count the MP3 recording file source title method of characteristic.
Technical problem of the invention is achieved through the following technical solutions:
A kind of MP3 recording file source title method based on side information statistical property, the recognition methods include following step
It is rapid:
Step 1: selecting the MP3 recording file that each serial model No. mobile phone is recorded under a plurality of brands as training sample, and make
The side information parameter of the training sample is extracted with MP3 codec Lame-3.99.5;The side information parameter includes master data
The side information of initial position, scale factor selection information and particle 0 and particle 1, and master data initial position and scale factor choosing
The common parameter that information constitutes two particles is selected, the side information of particle 0 or particle 1 is referred to as independent parameter, analyzes the side information
The service condition and statistical property of parameter, establish the common parameter service condition table of comparisons of mobile phone model and two particles, and make
The mobile phone building model with brand is directed to the part statistic of independent parameter;
Step 2: extracting the side information parameter of MP3 recording file to be measured using MP3 codec Lame-3.99.5, detect
Whether the value of master data initial position is all 0, and then the table of comparisons established with step 1 is compared, and tentatively draws a circle to approve MP3 to be measured
The mobile phone brand in recording file source;
Step 3: detecting L channel and the right side of the scale factor selection information in particle 0 in MP3 recording file side information to be measured
Value in sound channel is compared by the value with the table of comparisons that step 1 is established, and the MP3 to be measured record tentatively drawn a circle to approve from step 2
The specific series under specific mobile phone brand or some mobile phone brand are further selected in several mobile phone brands of sound document source;
Step 4: detecting L channel and the right side of the scale factor selection information in particle 1 in MP3 recording file side information to be measured
Value in sound channel is compared by the value with the table of comparisons that step 1 is established, and from step 3 determine mobile phone brand in into
One step, which determines in mobile phone series that specific series or step 3 determine, further determines that concrete model;
Step 5: continue to analyze the parameter value that each particle in MP3 recording file side information to be measured independently uses parameter,
The part statistic construction feature of the parameter value of parameter is independently used according to each particle, each particle is independently using ginseng
The part statistic of several parameter values is consistent with the statistic that training sample extracts is directed in step 1;On this basis, lead to
It crosses using LibSVM classifier, and combines in step 1 for the model of particular brand mobile phone building, finally determine MP3 record to be measured
The mobile phone of sound file which model under the mobile phone brand.
It includes 6 parameters, respectively part2_3_length, big_ that each particle, which independently uses parameter,
values、global_gain、scalefac_compress、region1_start、region2_start。
The speech samples library is divided into two set, and one of set is used as training set, another set, which is used as, to be surveyed
Examination collection, training set and test set respectively include the sample that 1480 durations are about 3 seconds, are in addition recorded using classifier to MP3 to be measured
Before the side information parameter attribute of sound file is trained and tests, place is all normalized to every one-dimensional characteristic of all samples
Reason, to reduce the inconsistent adverse effect to classifier performance of different characteristic value variation range.
Compared with prior art, mentality of designing of the invention is by analysis MP3 recording file code stream namely MP3 recording
In file frame in side information each parameter handling characteristics and statistical property, therefrom find out different brands model mobile phone generate MP3
Using the tendency of these parameters or feature when recording file, to realize the identification to MP3 recording file source, that is, distinguish given
Mp3 file be that the mobile phone of which kind of brand and model is recorded, this recognition methods has feature set building process relatively simple fast
Speed, algorithm complexity is low, recognition accuracy is high, real-time is good, is convenient for the advantages that operation.
Detailed description of the invention
Fig. 1 is the step flow diagram of the embodiment of the present invention.
Fig. 2 is the distribution schematic diagram of parameter ain_data_begin.
Fig. 3 is the subregion schematic diagram of MP3 block.
Fig. 4 is the distribution schematic diagram of parameter big_values.
Fig. 5 is the distribution schematic diagram of parameter global_gain.
The sub-district schematic diagram in the area Tu6Wei great Zhi.
Fig. 7 is the distribution schematic diagram of parameter region1_start.
Fig. 8 is the distribution schematic diagram of parameter region2_start.
Fig. 9 is the distribution schematic diagram of parameter part2_3_length.
Figure 10 is the distribution schematic diagram of parameter scalefac_compress.
Figure 11 is that the area great Zhi code table indexes distribution schematic diagram.
Figure 12 is that the area region0 code table indexes distribution schematic diagram.
Figure 13 is that the area region1 code table indexes distribution schematic diagram.
Figure 14 is that the area region2 code table indexes distribution schematic diagram.
Specific embodiment
It will elaborate again by above-mentioned attached drawing to the embodiment of the present invention below.
A kind of MP3 recording file source title method based on side information statistical property, what is involved is pass through research MP3 record
The handling characteristics and statistical property of coding parameter in sound file side information, to judge which brand a MP3 recording file is
The mobile phone of which model is recorded under brand.
The conceptual illustration about MP3 recording file involved in the recognition methods, specifically:
MP3 coding standard
MP3 full name is MPEG1 Layer-3, is the audio-frequency unit in MPEG standard.Though MP3 is lossy compression, it is close
The advantages such as sound quality, high compression ratio, opening and the ease for use of CD make it from birth once come the audio format that rapidly becomes mainstream,
It and is at present still most commonly used one of the format of digital audio field.The coding module of MP3 core is mainly by 5 part groups
At respectively sub-filter group, MDCT transformation, psychoacoustic model, quantization and coding, data stream.Detailed process is:
Input audio signal passes through 32 sub-filter groups all the way and MDCT transformation carries out time-frequency convert, while another way passes through the " heart
Reason acoustic model " calculates the signal energy and SMR of each subband." quantization and coding " module determines to distribute to son using SMR
The quantization digit of band signal, so that quantization noise is lower than masking threshold, finally by " data flow frame packaging " by the sample of subband
And other additional datas are assembled into a bit stream by the format of frame.
MP3 frame structure
MP3 data flow is packaged by minimum unit of frame, and every frame data generally comprise frame head (header), redundancy school
Test (CRC), side information (side information), master data (main data) and auxiliary data (ancillary data) 5
A part.Frame head is made of the data of 32 bits, and when decoding passes through the synchronization character in frame head first and determines the position that decoding starts,
And the relevant information that the MP3 is extracted from frame head is subsequent further decoding place mat, sample rate, sound channel mould such as audio signal
Formula, bit rate etc..CRC check is one 16 parity check words, for checking whether the frame data go out in transmission process
Mistake is showed.This is an optional information, shows there is CRC check when the guard bit in frame head is " 1 ", on the contrary then do not have.
Master data decoding is saved in side information needs various parameters to be used, such as quantization step, Huffman code table index, side information
Size be 130(monophonic) or 246(two-channel) position bit.And then side information is master data namely original audio sample
Data flow after this coding.The master data of each frame is divided into particle 0(granule 0) and particle 1(granule 1) two parts,
Every part is made of scale factor (scale factor) and Huffman code word (Huffman codes) again.Last supplementary number
According to being also optionally, to be defined by user oneself, the inside storage is some and decodes unrelated audio information, such as song title, singer, album
Etc. information.
Parameter in side information
The sound that sound pick-up outfit is recorded forms the process of mp3 file from PCM/PDM sampled value to the end by compressed encoding
Be all based on above-mentioned process, but different models of equipment when implementing the modules of MP3 standard there may be difference, because
The code flow structure and its parametric statistics characteristic of this MP3 ultimately produced also can different from.So the recognition methods phase of the invention
The statistical property for analysing in depth each parameter in the mp3 file side information that distinct device is recorded is hoped, so that finding out each equipment uses ginseng
Several features.
Side information is mainly stored for the decoded parameter of master data, generally includes master data initial position (main_
Data_begin), than column selecting predictors information (scfsi) and the side information of particle 0 and particle 1, as shown in table 1.
1. side information structure of table
It is described in detail according to table 1 to the effect of major parameter
The shared parameter of (1) two particle:
Main_data_begin: the offset of present frame master data starting position, because MP3 standard has used bit pond
Technology, the therefore therefore value can also be negative value, negative value this frame master data begin with may a frame in front, this indicates this
The forward offset in frame master data starting position indicates that this frame of master data starts if it is 0.
Private_bits: reserved bit, bit wide are 3bit in two-channel, and monophonic is 5bit, this can be by user
It oneself determines.
Scfsi: indicating whether two particles share scale factor, and " 1 " indicates that two particle fraction factors share, " 0 " table
Show that the scale factor of each particle is independent.
(2) parameter that each particle independently uses:
Part2_3_length: the bit number of scale factor and huffman data in master data is indicated.Because side information
Length is fixed, therefore the position that can be started by the position or next frame of this value calculating auxiliary data (if use)
It sets.
Big_values: the efficiency in order to improve Huffman encoding, MP3 standard is by every piece of 576 frequency lines by low frequency to height
Frequency is divided into 3 areas: the area great Zhi, the area little Zhi and zero area.The area great Zhi is exactly the relatively large subregion of frequency amplitude as its name suggests,
For area's coefficient of frequency with two for one group of coding, length is big_values × 2;The area little Zhi is by being worth the frequency system for -1,0 and 1
Array is at 4 one group of codings, length is (big_values × 2 count1 × 4-);Zero area coefficient of frequency is all 0, is not required to
It encodes.
Global_gain: the global proportionality factor, the quantization step that when inverse quantization uses be exactly by the Parameter Switch come
's.
Scalefac_compress: for determining bit number shared by scale factor in master data.
Table_select [region]: every piece of MP3 576 coefficient of frequencies are divided into 3 areas, the area great Zhi, the area little Zhi and
Zero area.Wherein the area great Zhi can be divided into 3 sub-districts, respectively region0, region1 and region2 again.Huffman decoding
When, this 3 sons, which are distinguished, to be decoded by a code table, and code table is coefficient of frequency and the current block letter according to the sub-district maximum amplitude
Number statistical property determine, table_select [region] indicates the code table index value of corresponding sub-district.
Count1table_select: it for the code table to small value fauna number encoder, is selected from specific two code tables.
Region1_start: the position that the area region1 starts is indicated.Region2_start: indicate that the area region2 starts
Position.
It is related to several conceptual descriptions of MP3 recording file, the concrete operations step of recognition methods of the present invention according to above-mentioned
It is rapid as follows:
Step 1: selecting the MP3 recording file that each serial model No. mobile phone is recorded under a plurality of brands as training sample, and make
The side information parameter of the training sample is extracted with MP3 codec Lame-3.99.5;The side information parameter includes master data
The side information of initial position, scale factor selection information and particle 0 and particle 1, the English being corresponding in turn to, which is explained, is respectively
Main_data_begin, scfsi, side_info_gr0, side_info_gr1, and master data initial position and scale factor
The side information of the common parameter of selection information two particles of composition, particle 0 or particle 1 is referred to as independent parameter, analyzes side letter
The service condition and statistical property for ceasing parameter, establish the common parameter service condition table of comparisons of mobile phone model and two particles, and
The mobile phone building model with brand is directed to using the part statistic of independent parameter;Meanwhile in above-mentioned retouching about selection mobile phone
In stating, it is to be distinguished according to brand, series, such a concept from big to small of model, i.e., is wrapped under each mobile phone brand
It again include concrete model under each series containing each series;
Step 2: extracting the side information parameter of MP3 recording file to be measured using MP3 codec Lame-3.99.5, detect
Whether the value of master data initial position is all 0, and then the table of comparisons established with step 1 is compared, and tentatively draws a circle to approve MP3 to be measured
The mobile phone brand in recording file source;
Step 3: detecting L channel and the right side of the scale factor selection information in particle 0 in MP3 recording file side information to be measured
Value in sound channel is compared by the value with the table of comparisons that step 1 is established, and the MP3 to be measured record tentatively drawn a circle to approve from step 2
The specific series under specific mobile phone brand or some mobile phone brand are further selected in several mobile phone brands of sound document source;
Step 4: detecting L channel and the right side of the scale factor selection information in particle 1 in MP3 recording file side information to be measured
Value in sound channel is compared by the value with the table of comparisons that step 1 is established, and from step 3 determine mobile phone brand in into
One step, which determines in mobile phone series that specific series or step 3 determine, further determines that concrete model;
Step 5: continue to analyze the parameter value that each particle in MP3 recording file side information to be measured independently uses parameter,
Each particle independently use parameter include 6 parameters, respectively part2_3_length, big_values, global_gain,
Scalefac_compress, region1_start, region2_start independently use the parameter of parameter according to each particle
The part statistic construction feature of value, each particle independently use the part statistic and step 1 of the parameter value of parameter
In for training sample extract statistic it is consistent;On this basis, by using LibSVM classifier, and step 1 is combined
In for the building of particular brand mobile phone model, MP3 recording file to be measured which model under the mobile phone brand finally determined
Mobile phone.
Embodiment
The present invention examines the specific identification step of the recognition methods again with a specific experiment:
Experimental setup
The speech samples library of specific experiment involved in the present invention is the MP3 recording that 10 sections of mobile phones as shown in table 2 are recorded
File is constituted.Recording environment is more quiet office;Participating in recording personnel is 9 male 8 female, totally 17 people;Duration 9 is recorded per capita
Minute or so, sample database duration about 2.5 hours altogether.The recording function of 10 mobile phones is kept while opened and closed as far as possible when recording
Can, and mobile phone location is each attached to the position apart from 1-1.2 meters of speaker.
The relevant information for 10 sections of mobile phones that the experiment of table 2. is selected
The present invention selects LibSVM as 3 sections of mobile phones (MLnote, MX2, MX4) for distinguishing Meizu brand, HMnote brand
2 sections of mobile phones (note1, note2) and OPPO brand 2 sections of mobile phones (OnePlus1, Find7) classifier.Sample database quilt
It is divided into two set, one of set is used as training set, is made of the mp3 file that 5 male 4 female record;Another set conduct
Test set is made of the mp3 file that remaining 4 male 4 female record.Each set is cut into the small sample that duration is about 3 seconds,
The training set and test set for being directed to every kind of equipment formation in this way include 1480 samples.In addition, in use LibSVM to feature
Before being trained and testing, every one-dimensional characteristic of all samples is all normalized, to reduce the change of different characteristic value
Change the inconsistent adverse effect to classifier performance of range.
According to above-mentioned Preparatory work of experiment, recognition methods of the invention operates in accordance with the following steps, detailed process institute referring to Fig.1
Show:
The side information parameter of MP3 recording file to be measured is extracted using MP3 codec Lame-3.995, and is believed according to the side
Breath parameter constructs the feature set of MP3 recording file to be measured.
The MP3 recording file of different model mobile phone recording under 10 sections of different brands is chosen as speech samples library, analytic language
Two public ginsengs of particle in MP3 recording file side information that different model mobile phone is recorded under 10 sections of different brands in sound sample database
Number, and the table of comparisons 3 is established according to two particle common parameter service conditions.
The value of parameter main-data-begin and scfsi in 10 sections of mobile phones that the experiment of table 3. is selected
Judge the service condition of bit pool technology, is that two particles are public in 10 kinds of MP3 recording file side informations shown in table 3
The service condition of parameter main_data_begin and scfsi, " * " indicate the not shown evident regularity of the value.The 2nd column can from table 3
To see, the main_data_begin value of the mobile phone of Meizu and OPPO brand is all 0, therefore their MP3 encoder is simultaneously
Do not use bit pond mechanism.And Mi and Mi HMnote brand has then used bit pond mechanism, their main_data_begin
The distribution of value is as shown in Figure 2.Therefore, by confirming whether MP3 recording file has used bit pool technology, it can be determined that this document
It is to be recorded by Mi brand mobile phone or Meizu or OPPO brand mobile phone is recorded.If so, illustrating MP3 recording file to be measured
It is recorded by the mobile phone of Meizu or OPPO brand;Otherwise it is recorded by the mobile phone of Mi or HMnote brand, it can be first by the step
A few money mobile phone brands in step delineation MP3 recording file to be measured source.
On this basis, target device can be further reduced by the value of the scfsi of particle 0 in the every frame of mp3 file
The range in source detects scale factor in MP3 recording file side information to be measured and selects information scfsi in the L channel of particle 0
With the value in right channel.As shown in the 3rd column in table 3, the scfsi value of the right channel of Meizu and Mi HMnote brand particle 1
It is 0(Lame decoder to the right channel scfsi of monophonic audio not assignment, therefore is default value 0), and Mi and OPPO brand
Corresponding scfsi value be -1(Lame decoder -1 is disposed as to the effective scfsi of particle 0, so that subsequent decoding passes through
Scfsi is worth positive and negative differentiation value to belong to particle 0 or particle 1).In addition, the scfsi value of Mi4 particle 1 is more special, it is left
The scfsi value of sound channel is equal with the scfsi value of right channel always (passing through the observation to great amount of samples).Therefore in conclusion we
A MP3 to be measured can be distinguished according to the value of main_data_begin and scfsi in MP3 recording file side information to be measured
Recording file comes from the mobile phone of Mi3 mobile phone, Mi4 mobile phone or HMnote brand, Meizu brand, OPPO brand.
So far, this method has been able to complete the primary work of MP3 recording file source identification, but also using every
The parameter that independently uses of grain is to 3 sections of mobile phones of Meizu brand, 2 sections of mobile phones of HMnote brand and 3 sections of hands of OPPO brand
Machine further discriminates between, and in this 8 sections of mobile phones, the mp3 file that OPPO R831S is recorded, the parameter that right channel uses is Lame
The default value of codec is all 0, part2_3_length value as global_gain value is all 210, big_values value
It is all 0 etc..And other remaining 7 sections of mobile phones no open-and-shut differentiating characteristics of presentation in the parameter of side information, especially belong to
In the mobile phone of same brand, has higher similitude to the use of parameter.This may be to produce because being designed by same producer
Mobile phone, even different model, for circuit, the components of selection and the signal processing of exploitation of audio-frequency module design
All there is biggish similitudes for algorithm etc..But the design requirement just because of different model mobile phone and market orientation be not yet simultaneously
Together, certainly there is certain difference, this species diversity is recorded eventually by equipment for the soft and hardware of internal audio frequency module and combination
Code flow structure, parameter or data statistics of MP3 recording file of generation etc. show.Therefore, the present invention passes through analysis
The statistical property of coding parameter in mp3 file side information, constructing being capable of accurate and effective differentiation 3 sections of mobile phone of Meizu brand
(MLnote, MX2, MX4), 2 sections of mobile phones (HMnote1, HMnote2) of HMnote brand and 2 sections of mobile phones of OPPO brand
The feature set of (OnePlus1, Find7)、With。
In addition, bit pond mechanism is not used when MP3 is encoded for Meizu brand and OPPO brand, so Fig. 2 only gives Mi
The distribution curve of 4 sections of equipment main_data_begin values of brand.The shape of this 4 curves is although similar as shown in Figure 2, and all
It is concentrated mainly between 450-510, but the corresponding curve of HMnote1 and HMnote2 still has biggish difference, such as
Probability of the main_data_begin value that HMnote1 is used between 460-480 is apparently higher than HMnote2, otherwise is in value
What the main_data_begin on the section 481-510, HMnote2 were used will be far more than HMnote1.So this method this two
Several points are picked on a section, and corresponding main_data_begin value (462,474,495,498 and 511) are occurred
Probability as distinguish HMnote1 and HMnote2 feature, be denoted as respectively.In addition, entire MP3
The mean value and variance of file main_data_begin value will also be calculated and as characteristic of division、。
(each particle is known as block to MP3 coding standard by Block, monophonic audio, and dual-channel audio is by each every piece
Left and right acoustic channels in grain are referred to as block) 576 bars frequency lines are divided into three regions: the area great Zhi (big_ by frequency from low to high
Values), the area little Zhi (count1) and complete 0th area (rzero), as shown in Figure 3.By the coefficient of frequency of higher magnitude in the area great Zhi
Composition, and with two for one group of carry out Huffman coding;The coefficient of frequency in the area little Zhi is 1, -1 or 0, with four for one group
Come coding of tabling look-up;Complete 0th area does not need then to encode.
Parameter big_values is the number for indicating to fall in the coefficient of frequency in the area great Zhi.Different to parameter main_data_
The characteristics of use of begin, different brands mobile phone has oneself to the use of big_values, as shown in Figure 4.Meizu brand
The big_values value of corresponding mp3 file is mainly distributed between 67-115, and 3 sections of equipment are to 77,78,86,87,95,
96,104,105 this 8 use probability being worth are obviously higher than other values, but probability value of the distinct device in the same value is deposited
In biggish difference.The distribution curve of HMnote brand is whole more steady, but both has in one section of successive range bright
Aobvious protrusion, HMnote1 is about in the range that big_values is 82-97, and HMnote2 is about 145-162 in big_values
Range.The distribution similarity highest of 2 sections of equipment big_values values of OPPO brand has in the range of 189-193 obvious
It increases, but the probability value of 190-192 is still able to more efficiently distinguish the two.Simultaneously, it has been found that OPPO brand uses
The value of big_values be significantly larger than Meizu brand and HMnote brand, this is because mp3 file that OPPO brand is recorded
Code rate is 320kbps, and much larger than the 64kbps of Meizu brand and HMnote brand, and code rate is higher, what audio signal was quantized
Finer, the biggish coefficient of frequency of amplitude will increase after quantization, so that the range in the area great Zhi is bigger.
Recognition methods of the invention select big_values equal to 86,87,104 and 105 probability value for, 85,87,149 and 151 probability value is, the probability value of 190-192 is.In addition, whole head MP3 is recorded from the point of view of each equipment corresponds to the location and shape of distribution curve
The mean value of all big_values values and variance can also be in promotion classifiers as feature in file (every head duration is about 3 seconds)
Energy aspect is contributed.So the feature of mean value and variance of the building based on big_values、、、、With。
Overall situation quantization gain global_gain is another representation of quantization step, its value range is 0-
210.Fig. 5 (a), (b), (c) respectively indicate Meizu brand, parameter in HMnote brand and OPPO brand mobile phone MP3 recording file
The distribution of global_gain.As shown in Figure 5, although these three brands global_gain value is all to be concentrated mainly on 125-185
Between, but also having respective characteristic distributions: Meizu brand mainly uses continuous in this section and 15 interval about 1
Value, wherein the probability of occurrence of 9 values has been more than 0.07;And the distribution of HMnote brand and OPPO brand is just comparatively equal
Even, the number of accepted value is more (25-40), and other than a peak value in HMnote1, other values are no more than 0.07.
In addition, as can be seen from Figure 5, same brand equipment using global_gain preference also different from.Therefore curve distribution phase
The biggish position of difference is selected as the feature of classification, and this method chooses global_gain and is equal to 142,144,157 and 159
Probability value conduct, 135,136,142 and 143 probability value conduct, 135,
136,142 and 143 probability value conduct;And the mean value and variance conduct of global_gain、、、、With。
In order to improve code efficiency and robustness, the area great Zhi is further divided into 3 sub-districts by MP3 coding standard, respectively
Region0, region1 and region2, as shown in Figure 6.As its name suggests, parameter region1_start and region2_start
What is meant that is the position of two critical points among 3 areas.
As can be seen from Figures 7 and 8, each brand equipment makes parameter region1_start and region2_start
With suffering from respective feature.But the similitude of a few money equipment homologous threads of same brand is relatively high, especially OPPO product
The distribution curve of 2 sections of equipment of board, region1_start and region2_start are almost overlapped.Other the one of OPPO brand
A feature is exactly that the use to the two parameters is more single, and region1_start value is mainly 15 and 18, region2_
Start value is mainly 64 and 288.For Meizu brand and HMnote brand, there is a small number of corresponding probability values of several values to same
The equipment of brand has certain distinction, if Meizu brand region1_start is 10,12, region2_start 37,45
When probability value, HMnote brand region1_start be 2,22,67 22, region2_start when probability value.Therefore,
This method is with regard to two parameters of region1 and region2 to Meizu brand construction feature, to HMnote brand
Construction feature。
Parameter part2_3_length indicates ratio of the current block for scale factor and Huffman data in main code data
Special number, this value is general and the directly proportional relationship of code rate of mp3 file namely the code rate of MP3 it is higher, part2_3_length
It is bigger.Fig. 9 demonstrates this relationship, and the code rate of the mp3 file of Meizu brand and HMnote brand is 64kbps, they
Part2_3_length value is mainly between 500-1000, and the code rate of OPPO brand is 320kbps, part2_3_length
Value is just corresponding much greater, is concentrated mainly in the range of 1500-2000.Although the part2_3_ of different brands equipment
Larger difference is distributed in length value, but the distribution curve of same brand different models of equipment is similar.The present invention according to
It is that various brands construct 4 dimensional features with the biggish position of difference in brand distinct device distribution map, adds mean value and variance,
Totally 6 dimensional feature, respectively、、。
Scalefac_compress is the index value of current block scale factor data length, it is long with scale factor data
The mapping relations of degree are as shown in table 4.
The corresponding relationship of table 4. parameter scalefac_compress and slen1, slen2
If current block is long block, slen1 indicates the bit number of scale factor used in the scale factor of 0-10;
Slen2 indicates the bit number of scale factor used in the scalefactor bands of 11-20;If current block is short block, slen1 indicates 0-
The bit number of scale factor used in 5 scale factor, slen2 indicate scale factor used in the scale factor of 6-10
Bit number.Since the mp3 file that Meizu brand equipment is recorded does not have scale factor data, scalefac_
The value of the parameter of compress also all 0, therefore this method only analyzes HMnote brand and OPPO brand scalefac_
The distribution of compress value.From the point of view of the distribution curve in Figure 10, HMnote2 is to 0-10,12 and No. 13 scalefac_
The probability that compress is used is above HMnote1, mutually opposes that 14 and No. 15 are significantly lower than HMnote1 using probability;And for
OPPO brand, when scalefac_compress value is in the range of 4-10, the corresponding curve of OnePlus1 is in the upper of Find7
Side, opposite rule is then presented when scalefac_compress value is in the range of 11-15.So identification side of the invention
Method has chosen probability value of the sacle_compress equal to 13 and 15 as feature to HMnote brand、,
Mean value and variance are characterized、;Sacle_compress etc. is increased compared with HMnote brand to OPPO brand
In 9,10 2 probability values, then plus mean value and variance, altogether6 dimensional features.
In MP3 coding standard, 34 Huffman code tables are shared, the area great Zhi has used No. 0-31 32 code tables, remaining
Two code tables are the dedicated code tables in the area little Zhi.And for every piece of 576 coefficient of frequencies, the region0 in the area great Zhi,
Coefficient in region1 and region2 is respectively encoded by a code table, the area little Zhi by one of coding in dedicated code table, therefore
Every block number completes Huffman encoding processes according to need to use 4 code tables.
Recognition methods of the invention mainly examines the characteristics of area great Zhi code table of each mp3 file, as shown in figure 11.
The distribution of HMnote brand and OPPO brand is more conform with the rule that code table uses in MP3 coding standard, the MP3 of HMnote brand
File code rate is lower (64kbps), and the mp3 file of OPPO brand is the high code rate MP3 of 320kbps, therefore the former should be more
Ground uses small directory, and as the peak value of curve in Figure 10 (b) nearly all concentrates on 15 and position below, and the latter is to large size
Code table is apparently higher than the former using probability, and does not almost have to No. 12 code tables below.Although having the section using several directories
The coefficient and current demand signal statistical property of middle amplitude maximum determine [ISO11172-3], but usual 0th~15 directory is used to pair
Maximum amplitude is no more than 15 Interval Coding, if to encode to the frequency values for being more than 15, need to just use the 15th~31 directory.Again
Will be more finer to the quantization of audio signal when meaning compressed encoding in addition code rate is higher, the big value of amplitude in coefficient of frequency
It just will increase, so theoretically the higher probability using big directory of code rate is bigger.However Meizu brand then disobeys this
Rule, although its corresponding mp3 file is also 64kbps, this three sections of equipment mainly use 13,15 and 24-28
Directory.In addition, Figure 12-Figure 14 is the code table distribution of three, the area great Zhi sub-district region0, region1 and region2 respectively,
No matter to the equipment of which brand it can be seen from this 3 width, to region2 from region0 to region1, big directory is used
Probability be gradually reduced, opposite small directory is used more and more.This with frequency in region0-region2 be by
The rule of low to high distribution is consistent, that is to say, that the value relative maximum of coefficient of frequency in region0, time in region1
It, it is then minimum in region2.
Recognition methods of the invention is finally directed in the entire area great Zhi and each area region0, region1 and region2
Probability using certain code tables is that three brand equipment construct several dimensional features, is specifically shown in Table 5.
Table 5. is directed to the feature of parameter codebook_index selection
Therefore, with reference to above-mentioned specific alignments, specific recognition methods of the invention can be divided into following 9 steps, tool
Body process is as shown in Figure 1.
(1) the side information parameter of MP3 recording file to be measured is extracted and according to above-mentioned side using codec Lame-3.995
Method construction feature collection、With。
(2) judge whether the service condition of bit pool technology, the i.e. value of detection parameters main_data_begin are all 0.If
It is then to illustrate that the recording file is recorded by the mobile phone of Meizu or OPPO brand, executes step (3);Otherwise by Mi or HMnote product
The mobile phone of board is recorded, and is executed step (6).
(3) value scfsi_gr0_ch0, scfsi_gr0_ of the detection parameters scfsi in the L channel and right channel of particle 0
ch1.If scfsi_gr0_ch0=- 1, scfsi_gr0_ch1=- 1, illustrate that the recording file is recorded by the mobile phone of OPPO brand,
It executes step (4);If scfsi_gr0_ch0=- 1, scfsi_gr0_ch1=0, illustrate the recording file by Meizu brand
Mobile phone is recorded, and is executed step (9).
(4) whether the value for detecting other parameters (selecting 6 parameters herein) corresponding right channel is all silent when Lame is decoded
Recognize value, i.e. whether the global_gain value of right channel is all whether 210, part2_3_length value is all 0, big_values
Whether value is all 0, saclefac_compress value and whether is all whether 0, region1_start value is all 2, region2_
Whether start value is all 4.If so, illustrating that the recording file is recorded by OPPO R831s mobile phone.
(5) feature set is usedDistinguishing the recording file with LibSVM classifier is by OPPO
What Find7 or OnePlus1 was recorded.
(6) value scfsi_gr0_ch0, scfsi_gr0_ of the detection parameters scfsi in the L channel and right channel of particle 0
ch1.If scfsi_gr0_ch0=- 1, scfsi_gr0_ch1=- 1, illustrates that the recording file is recorded by the mobile phone of Mi brand, hold
Row step (7);If scfsi_gr0_ch0=- 1, scfsi_gr0_ch1=0, illustrate the recording file by the hand of HMnote brand
Machine is recorded, and is executed step (8).
(7) value scfsi_gr1_ch0, scfsi_gr1_ of the detection parameters scfsi in the L channel and right channel of particle 1
ch1.If being all satisfied scfsi_gr1_ch0=scfsi_gr1_ch1 to all frames, illustrate that the recording file is recorded by 4 mobile phone of Mi
System;Otherwise illustrate that the recording file is recorded by 3 mobile phone of Mi.
(8) feature set is usedWith LibSVM classifier distinguish the recording file be by
What HMnote1 or HMnote2 was recorded.
(9) feature set is usedDistinguishing the recording file with LibSVM classifier is by Meizu
What MLnote, Meizu MX2 or Meizu MX4 were recorded.
By above-mentioned 9 steps it is found that by examine a head MP3 recording file side information in main_data_begin with
The service condition of scfsi parameter can tentatively reduce the investigation range of this document source device or successfully complete identification work
Make or in 3 sections of mobile phones (MLnote, MX2, MX4) of Meizu brand, 2 sections of mobile phones (note1, note2) of HMnote brand, and
It is further discriminated between in 2 sections of mobile phones (OnePlus1, Find7) of OPPO brand.Therefore, which mainly verifies for three
The classification performance of the feature of the corresponding building of brand.In addition, in order to analyze the feature of each parameter in side information to promotion whole detection
The contribution of accuracy rate, experiment give the classification performance of each parameter attribute, as shown in table 6.
The classifying quality of each parameter attribute of table 6. and its intersection feature to Meizu brand mobile phone
Table 6 is that the feature based on different parameters building is quasi- to the classification of Meizu brand three sections of mobile phones MLnote, MX2 and MX4
True rate, it can be seen that in addition to the feature of parameter part2_3_length and codebook_index, the detection of other 4 groups of features is quasi-
True rate has reached 90% or more.By further to the observation of confusion matrix, it has been found that the two parameter attribute performances are poor
Immediate cause be sample that a big chunk MLnote and MX4 sample has been identified as counterpart device.For codebook_
The feature of index has the MX4 sample of nearly half to be identified into MLnote sample;And use the spy of part2_3_length
When sign, or even has more than 60% MX4 sample and be classified into MLnote sample, while also thering is 26% MLnote sample to be differentiated into
The sample of MX4.But the two parameter attributes are difficult to differentiate between the basic reason of MLnote and MX4 or this two equipment is Meizu
Two very close mobile phone of Time To Market in brand, possible producer to the architecture design of their audio-frequency modules, part selection with
And algorithm optimization has a higher similitude, therefore the otherness between certain parameters of its recording file is with regard to relatively smaller, example
It is very close if the statistical property of part2_3_length and codebook_index parameter, as shown in Figure 8, Figure 9.This algorithm
The feature union that final choice is constructed based on all parameters(34 dimension) is as three sections of Meizu brand of differentiation
The feature set of equipment, this feature collection have outstanding classification performance, and Detection accuracy has reached 99.32%.
It is differentiation result of each parameter attribute to two mobile phone HMnote1 and HMnote2 of HMnote brand shown in table 7.By
Table 7 it is found that in addition to parameter part2_3_length feature, the detection effect of other features is all satisfactory, is based especially on
The feature of big_values and codebook_index building, classification accuracy have reached 99.5% or more.Identification of the invention
The set of all parameter attributes of method final choiceIt is able to achieve to HMnote1 and HMnote2
Sample 100% is accurately distinguished.Equally, what table 8 provided is each parameter attribute to two mobile phone Find7 of OPPO brand and
The identification performance of OnePlus1, in addition to the feature Detection accuracy of part2_3_length and scalefac_compress is lower
Except, other features suffer from preferable performance, and the classification accuracy of the feature of big_values is that 100% or even outline are excellent
In the feature set of recognition methods final choice of the invention.From the result of table 7 and table 8, we can be with
See, it is our although the Detection accuracy of a certain or a certain category feature is used only already close to being even up to 100%
Method still selects the feature intersection formed by all kinds of parameters of side information as final characteristic of division.Because it is desirable that building
Feature can comprehensively reflect the characteristics of distinct device is to mp3 file code stream as far as possible, to improve the stabilization of this algorithm performance
Property and to collection external equipment applicability.
The classifying quality of each parameter attribute of table 7. and its intersection feature to HMnote brand mobile phone
The classifying quality of each parameter attribute of table 8. and its intersection feature to OPPO brand mobile phone
The present invention has studied the feelings that each parameter is used in the mp3 file side information of 10 kinds of main brand model mobile phones recording
Condition and its statistical property, have found distinct device using the characteristics of certain parameters and based on these features successfully distinguished Mi3,
The equipment of Mi4 and other three brands.On this basis, the statistical property of partial parameters has been analysed in depth, and to Meizu brand
3 sections of mobile phones (MLnote, MX2, MX4), 2 sections of mobile phones (note1, note2) of HMnote brand and 2 sections of mobile phones of OPPO brand
(OnePlus1, Find7) targetedly constructs characteristic of division, and the results show, which can be relatively accurately
It distinguishes 10 kinds of common smart phones and records the MP3 recording file generated, classification accuracy is respectively 99.32%, 100.00% and
99.70%.In addition, the feature construction process of this method is relatively simple quickly, therefore the complexity of algorithm is lower, real-time is good.
Claims (3)
1. a kind of MP3 recording file source title method based on side information statistical property, it is characterised in that the recognition methods packet
Include following steps:
Step 1: selecting the MP3 recording file that each serial model No. mobile phone is recorded under a plurality of brands as training sample, and use
MP3 codec Lame-3.99.5 extracts the side information parameter of the training sample;The side information parameter rises comprising master data
Beginning position, scale factor selection information and particle 0 and particle 1 side information, and master data initial position and scale factor selection
The side information of the common parameter of information two particles of composition, particle 0 or particle 1 is referred to as independent parameter, analyzes side information ginseng
Several service conditions and statistical property, establishes the common parameter service condition table of comparisons of mobile phone model and two particles, and uses
The part statistic of independent parameter is directed to the mobile phone building model with brand;
Step 2: extracting the side information parameter of MP3 recording file to be measured using MP3 codec Lame-3.99.5, main number is detected
It whether is all 0 according to the value of initial position, and then the table of comparisons established with step 1 is compared, tentatively draws a circle to approve MP3 recording to be measured
The mobile phone brand of document source;
Select information in the L channel and right channel of particle 0 Step 3: detecting scale factor in MP3 recording file side information to be measured
In value, be compared by the value with the table of comparisons that step 1 is established, and tentatively drawn a circle to approve from step 2 MP3 to be measured recording text
The specific series under specific mobile phone brand or some mobile phone brand are further selected in several mobile phone brands in part source;
Select information in the L channel and right channel of particle 1 Step 4: detecting scale factor in MP3 recording file side information to be measured
In value, be compared by the value with the table of comparisons that step 1 is established, and from step 3 determine mobile phone brand in further
It determines in the mobile phone series of specific series or step 3 determination and further determines that concrete model;
Step 5: continue to analyze the parameter value that each particle in MP3 recording file side information to be measured independently uses parameter, according to
Each particle independently uses the part statistic construction feature of the parameter value of parameter, and each particle independently uses parameter
The part statistic of parameter value is consistent with the statistic that training sample extracts is directed in step 1;On this basis, by making
It with LibSVM classifier, and combines in step 1 for the model of particular brand mobile phone building, finally determines MP3 recording text to be measured
The mobile phone of part which model under the mobile phone brand.
2. a kind of MP3 recording file source title method based on side information statistical property according to claim 1, special
It includes 6 parameters that sign, which is that each particle independently uses parameter, respectively part2_3_length, big_values,
global_gain、scalefac_compress、region1_start、region2_start。
3. a kind of MP3 recording file source title method based on side information statistical property according to claim 1, special
Sign is that the training sample for the MP3 recording file production that each serial model No. mobile phone is recorded under a plurality of brands is speech samples
Library, the speech samples library are divided into two set, and one of set is used as training set, another set is used as test set, training
Collection and test set respectively include the sample that 1480 durations are about 3 seconds, in addition on the side using classifier to MP3 recording file to be measured
Before information parameter feature is tested, every one-dimensional characteristic of all samples is all normalized, to reduce different spies
The inconsistent adverse effect to classifier performance of value indicative variation range.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610020274.6A CN105632516B (en) | 2016-01-13 | 2016-01-13 | A kind of MP3 recording file source title method based on side information statistical property |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610020274.6A CN105632516B (en) | 2016-01-13 | 2016-01-13 | A kind of MP3 recording file source title method based on side information statistical property |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105632516A CN105632516A (en) | 2016-06-01 |
CN105632516B true CN105632516B (en) | 2019-07-30 |
Family
ID=56047357
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610020274.6A Active CN105632516B (en) | 2016-01-13 | 2016-01-13 | A kind of MP3 recording file source title method based on side information statistical property |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105632516B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106531159B (en) * | 2016-12-09 | 2019-06-18 | 宁波大学 | A kind of mobile phone source title method based on equipment background noise spectrum signature |
CN107507626B (en) * | 2017-07-07 | 2021-02-19 | 宁波大学 | Mobile phone source identification method based on voice frequency spectrum fusion characteristics |
CN107274912B (en) * | 2017-07-13 | 2020-06-19 | 东莞理工学院 | Method for identifying equipment source of mobile phone recording |
CN109088974A (en) * | 2018-06-28 | 2018-12-25 | 深圳市口袋网络科技有限公司 | A kind of data save method and its device, equipment, storage medium |
CN109935234B (en) * | 2019-02-22 | 2020-02-07 | 东莞理工学院 | Method for identifying source equipment of sound recording |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102394062B (en) * | 2011-10-26 | 2013-02-13 | 华南理工大学 | Method and system for automatically identifying voice recording equipment source |
CN102394065B (en) * | 2011-11-04 | 2013-06-12 | 中山大学 | Analysis method of digital audio fake quality WAVE file |
US9520138B2 (en) * | 2013-03-15 | 2016-12-13 | Broadcom Corporation | Adaptive modulation filtering for spectral feature enhancement |
-
2016
- 2016-01-13 CN CN201610020274.6A patent/CN105632516B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN105632516A (en) | 2016-06-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105632516B (en) | A kind of MP3 recording file source title method based on side information statistical property | |
US7460994B2 (en) | Method and apparatus for producing a fingerprint, and method and apparatus for identifying an audio signal | |
US7081581B2 (en) | Method and device for characterizing a signal and method and device for producing an indexed signal | |
CN103403710B (en) | Extraction and coupling to the characteristic fingerprint from audio signal | |
KR101101384B1 (en) | Parameterized temporal feature analysis | |
Herre et al. | Robust matching of audio signals using spectral flatness features | |
Luo et al. | Detection of double compressed AMR audio using stacked autoencoder | |
JP2009511954A (en) | Neural network discriminator for separating audio sources from mono audio signals | |
JP2004530153A6 (en) | Method and apparatus for characterizing a signal and method and apparatus for generating an index signal | |
WO2006004050A1 (en) | System for detection section including particular acoustic signal, method and program thereof | |
CN102723079B (en) | Music and chord automatic identification method based on sparse representation | |
Tsai et al. | Background music removal based on cepstrum transformation for popular singer identification | |
Nwe et al. | Automatic Detection Of Vocal Segments In Popular Songs. | |
CN110647656A (en) | Audio retrieval method utilizing transform domain sparsification and compression dimension reduction | |
KR101808810B1 (en) | Method and apparatus for detecting speech/non-speech section | |
Deshmukh et al. | North Indian classical music's singer identification by timbre recognition using MIR toolbox | |
CN108735230B (en) | Background music identification method, device and equipment based on mixed audio | |
CN102394065A (en) | Analysis method of digital audio fake quality WAVE file | |
Wang et al. | Low pass filtering and bandwidth extension for robust anti-spoofing countermeasure against codec variabilities | |
CN102610234B (en) | Method for selectively mapping signal complexity and code rate | |
Hu et al. | Singer identification based on computational auditory scene analysis and missing feature methods | |
Rizzi et al. | Genre classification of compressed audio data | |
CN105845132A (en) | Coding parameter statistical feature-based AAC sound recording document source identification method | |
Chen et al. | A robust feature extraction algorithm for audio fingerprinting | |
CN105632523A (en) | Method and device for regulating sound volume output value of audio data, and terminal |
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 | ||
EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20160601 Assignee: SANMEN SANYOU TECHNOLOGY Inc. Assignor: Ningbo University Contract record no.: X2022330000855 Denomination of invention: A Method of MP3 Recording File Source Recognition Based on the Statistical Properties of Edge Information Granted publication date: 20190730 License type: Common License Record date: 20221226 |
|
EE01 | Entry into force of recordation of patent licensing contract |