CN103426439B - A kind of broadcast television audio signal content consistency detecting method - Google Patents

Abstract

The invention discloses a kind of broadcast television audio signal content consistency detecting method, comprise the following steps: step one, condition code extraction is carried out to audio sources, morphogenesis characters code storehouse; Step 2, signal transmission to be checked is gathered to the audio signal characteristic code of a timeslice, and the timestamp that record gathers; Step 3, audio signal characteristic code step 2 collected and timestamp are sent to inspection center's system; The condition code of condition code storehouse before and after timestamp in error time delay of the audio signal characteristic code received and audio sources contrasts by step 4, inspection center's system, detects, then think that signal is correct, otherwise think that signal transmission is wrong.The present invention can be applicable to the information source production mechanism such as broadcasting station, TV station in signals transmission and source signal consistance monitoring.

Description

A kind of broadcast television audio signal content consistency detecting method

Technical field

The present invention relates to audio signal characteristic code extractive technique, audio feature code retrieval calibration technology, propose a kind of method whether broadcast television audio signal of monitoring in transmitting procedure is consistent with its source signal.

Background technology

Function according to Broadcast and TV system all departments divides, broadcast television content is produced and is responsible for by various places TV and Radio Service, the transmission of signal is responsible for by transmitting station or the mechanism such as cable network company, internet operation company, audio-video signal is in transmitting procedure, due to a variety of causes (as mistake passes, intercuts etc.), audio-video signal that transmitted audio-video signal (hereinafter referred to as signal transmission) and TV and Radio Service export at first may be caused (hereinafter referred to as source signal, sound signal is wherein audio sources) inconsistent, cause mistake to broadcast, broadcast leakage.Therefore one is all needed whether correctly can to monitor and report to the police signal transmission content for TV and Radio Service or transport sector, and traditional signal level or field intensity monitoring, can only monitor the mass parameter of signal, monitoring judgement cannot be carried out to whether content is consistent with source signal.

State Intellectual Property Office of the People's Republic of China discloses on Dec 01st, 2010 patent documentation that publication number is CN101902677A, and title is audio detection device and method, and this device comprises: MIC electric circuit inspection module and phonation circuit detection module; According to amplitude and frequency, MIC electric circuit inspection module, for obtaining amplitude and the frequency of the sine wave signal after MIC circuit sampling, determines that whether MIC circuit is qualified, and sine wave output signal; Phonation circuit detection module comprises: change-over panel and the processor with MIC circuit; Change-over panel, for the simulating signal that phonation circuit of sampling exports, and is converted to digital signal by simulating signal, is sent to processor; Processor, for carrying out analyzing and processing to the amplitude of digital signal and frequency, judges that whether phonation circuit is qualified.This scheme is not suitable for monitors the sound signal of TV and Radio Service.

Summary of the invention

The present invention mainly solves whether consistent with the source signal technical matters of the content cannot play TV and Radio Service existing for prior art, provides the broadcast television audio signal content consistency detecting method whether a kind of signal can monitored in transmitting procedure is consistent with source signal.

The present invention is directed to that above-mentioned technical matters mainly solved by following technical proposals: a kind of broadcast television audio signal content consistency detecting method, comprises the following steps:

Step one, condition code extraction is carried out to audio sources, morphogenesis characters code storehouse;

Step 2, signal transmission to be checked is gathered to the audio signal characteristic code of a timeslice, and the timestamp that record gathers;

Step 3, audio signal characteristic code step 2 collected and timestamp are sent to inspection center's system;

The condition code of condition code storehouse before and after timestamp in error time delay of the audio signal characteristic code received and audio sources contrasts by step 4, inspection center's system, detects, then think that signal is correct, otherwise think that signal transmission is wrong.

As preferably, in step 4, error time delay is no more than 60 seconds.

As preferably, in step 2, the length of a timeslice is 5 to 60 seconds.

As preferably, when signal transmission to be checked is multiple signals, time-sharing multiplex mode is adopted to carry out collection audio signal characteristic code to every road suspect signal in turn.

As preferably, in step one and step 2, the condition code of sound signal is obtained by condition code extraction equipment, and condition code extraction equipment comprises:

Processor module, for controlling whole equipment;

Audio signal sample module, for gathering audio signal data, and by sending to audio feature code extraction module after processor module process;

Audio feature code extraction module, for carrying out vectorized process to the sound signal received, extracts condition code data;

Mixed-media network modules mixed-media, for being sent to center detection system by the timestamp of condition code and extraction condition code.

As preferably, described mixed-media network modules mixed-media comprises RJ45 interface and WIFI unit.

As preferably, the condition code of sound is extracted to comprise and is carried out successively quantizing, pre-emphasis, take advantage of Hamming window and carry out linear predictor coefficient analysis, and the linear regression single order prediction MFCC cepstrum obtained is as sound characteristic code.

The present invention utilizes every section audio signal to have the principle of unique characteristic information, by comparing to the ID Code of signal transmission and source signal, realizes the consistency analysis of two signals, thus realizes monitoring the content consistency of signal transmission.It is that audio signal characteristic code extracts and condition code calibration technology that the method realizes the gordian technique that signal transmission and source signal content consistency detect automatically.

The quantification of sound can be divided into two large classes: a class is scalar quantization, and another kind of is vector quantization.Scalar quantization be by sampling after signal value one by one quantize.And vector quantization several sampled signals is divided into-group, namely form a vector, then this vector Progressive symmetric erythrokeratodermia is quantized.Generally there is the loss of information in vector quantization, but regulate by quantified precision while packed data.By vector quantization, significantly can reduce the size of the quantification file of sound, thus reduce the amount of calculation of Sound Match identification.

Assuming that X is a K dimensional vector, each dimension component is real-valued random variable.Vectorial X is mapped to another K dimensional vector Y during vector quantization, is expressed as:

Y==VQ(X)

Y is value in a finite set, and a finite set is a code book, is denoted as CB={CW _i,

1≤i≤NC, NC is codebook size.VQ(VectorQuantization) process is exactly the mapping of sample space X to finite space CB:

$x\∈X\⋐E^K\→Y=\mathrm{VQ}\left(x\right)\∈\mathrm{CB}\⋐E^K---\left(3.1\right)$

There is a quantizing distortion between X and Y or claim distance metric d (X, Y).The quality available average quantization distortion D of quantizer VQ () expresses.

$D=\underset{x\∈X}{\mathrm{\Σ}}d(x,\mathrm{XQ}\left(x\right))/\left|X\right|---\left(3.2\right)$

Wherein | X| represents the number of element in set X.

When quantizer (the NN principle) of form below selecting,

Y=VQ (X)=CW _iiffd (X, CW _j), to all j ≠ i

Concerning a certain sample set X, code book and quantizer are one to one.

The selection of code book is realized by clustering algorithm, and VQ process is easy to realize by arest neighbors (NN) principle.Sample (training) the collection x of specific area (as acoustic domains) is E _ka proper subclass, X is larger, coverage rate is wider (relative to this field), more helpful to cluster.

In practical application, vectorial X is quantized into CW to a certain code book CB _iafter, only represent quantized result with the numbering i of this code word in CB.Like this, VQ can be expressed as:

Y=VQ (X)=iiffd (X, CW _i) < d (X, CW _j), to all j ≠ i

As in audio communication, in original code rate be 24Kbit/s lipreder basis on, if by the vector quantizations of 10 of every frame reflection coefficients in addition 10 dimensions, code rate just can be made to be reduced to 800bit/s, and sound quality does not decline substantially.Segment vocoders adopts vector quantization, and bit rate can be made to be reduced to 150bit/s.

Sound characteristic extraction is the key issue of pattern-recognition, and the quality of characteristic parameter has a great impact for voice recognition precision.It is generally acknowledged, sound is forced through glottis by air and produces recurrent pulse and formed through sound channel filtering, and pronunciation modeling is recurrent pulse passes through time-varying linear filter.The common method becoming sound characteristic during description adopts short-time spectrum, and cepstrum is conventional one.Cepstrum can by recurrent pulse and sound channel isolated, obtain audio parameter.

Cepstrum coefficient directly can be tried to achieve by the definition of cepstrum, and also can be obtained (LPCC) by linear predictor coefficient (LPC) recursion, wherein the calculated amount of LPCC is less.Voice signal can represent with all-pole modeling, and P rank forward prediction expression formula is:

$x=\underset{l=1}{\overset{p}{\mathrm{\&Sigma;}}}{a}_{1}x(n-1)---\left(2.1\right)$

In formula: a _lfor predictive coefficient.Recursion cepstrum coefficient c can be gone out by forward prediction coefficient _c

$\left\{\begin{array}{c}{C}_l=a_l\\ C_n=\underset{l-1}{\overset{n-1}{\mathrm{\&Sigma;}}}(1-\frac{1}{n})\\ 1<n<p\end{array}{a}_1{c}_{n-1}+a_n\right.---\left(2.2\right)$

In order to improve accuracy of identification, carry out non-thread conversion by Mei Er (MEL) yardstick, the computing formula of LPCMCC (LPC MFCC cepstrum) is:

${\mathrm{MC}}_k\left(n\right)=\left\{\begin{array}{c}{C}_{-n}+\mathrm{aMC}(n-1)\\ (1-a^2){\mathrm{MC}}_0(N-1)+{\mathrm{aMC}}_1(n-1)\\ {\mathrm{MC}}_{k-1}(n-1)+a({\mathrm{MC}}_k(n-1)-{\mathrm{MC}}_{k-1}\left(n\right))\end{array}\right.---\left(2.3\right)$

C _krepresent cepstrum coefficient; MC _krepresent Mei Er coefficient reciprocal.When sampling frequency equals 8KHZ, α gets 0.31.

We also ask for the LPCMCC linear regression coeffficient on γ rank by following formula in addition, and comparative experiments proves, the dynamic parameter utilizing formula (2.4) to obtain, can obtain good recognition result.

$R_r(t,T,\mathrm{\&Delta;T},N)=\frac{\underset{X=1}{\overset{N}{\mathrm{\&Sigma;}}}{R}_r(x,N)C[t+[\frac{x-1}{N-1}-\frac{1}{2}\left](T-\mathrm{\&Delta;T})\right]}{\underset{x-1}{\overset{N}{\mathrm{\&Sigma;}}}{P}_r^2(x,N)}---\left(2.4\right)$

Frame number used when wherein C (t), T, Δ T and N are illustrated respectively in the LPC MFCC cepstrum of moment t, window when calculating time width, the sound characteristic Parameter analysis of regression coefficient moves and calculate regression coefficient; P _r(X, N) represents weighting function when asking for γ rank linear regression coeffficient, and the weighting function of single order and second order is respectively

P ₁(x,N)＝x(2.5)

$P_2(x,N)=x^2-\frac{1}{12}(N^2-1)---\left(2.6\right)$

Feature extraction preprocessing process is: voice signal is sampled through 8KHZ, 1-0.98Z ^-1pre-emphasis, take advantage of the long 21.33ms(256 point of window), window moves the Hamming window of 10ms, then carries out 14 rank lpc analysis, using 14 linear regression single order LPCMCC (the Δ Δ LPCMCC) coefficients tieed up as sound characteristic parameter (totally 30 dimensions).

Segmentation short-time analysis is carried out to one section of sound import, a characteristic sequence can be obtained the sound that the training stage inputs is analyzed, obtain each stack features sequence and be called reference template, be designated as: j=1,2 ... V, in formula, j be template corresponding order numbering: J is the bulk analysis frame number in this order; V is the total template number in system template storehouse, can be equal to or greater than order number to be identified.Test template is called to the characteristic sequence that the sound analysis of cognitive phase input obtains, is designated as t={t ₁, t ₂, t ₃..., t _i, I is the frame number inputting sound to be identified.Template matches process compares exactly between reference template R and test template T, calculates its similarity degree.

Now each entry characteristic sequence stored in template is called reference model, a reference model can be expressed as R={R (1), R (2) ..., R (m) ..., R (M) }.Wherein, m is the order label of training voiced frame, and m=1 ~ M, M are voiced frame sums, and R (m) is m frame acoustic feature vector.Input entry sound to be identified is called test pattern, and test pattern is expressed as T={T (1), T (2), ..., T (n) ... T (N) }, n is test sound frame number, and comprise N frame sound altogether, T (n) is the n-th frame feature vector.

By calculating the degree of distortion between T and R, differentiate the similarity of T and R.During calculated distortion, from T and R, the distortion of corresponding frame is counted, if n and m is respectively optional frame number in T and R, D [T (n), R (m)] represent the degree of distortion of this two frame feature vector, then obtain the distortion in template between every frame according to different situations, thus calculate total distortion D (T, R).

$D[T,R]=\underset{n=m=1}{\overset{N}{\mathrm{\&Sigma;}}}D[T\left(n\right),R\left(m\right)]---\left(4.1\right)$

1) N=M, this be can calculate the distortion between n=m=N=M frame according to this and get and, can total distortion be obtained

2) N ≠ M. supposes N > M, then can by R={R (1), R (2), R (m) ..., R (M) } and map a N frame sequence { R (1) with linear expansion, R (2), ... R (m) ... R (M) }, then calculate the latter and { T (1), T (2), ..., T (n) ... T (N) between total distortion, this calculating just can be carried out according to the first situation frame by frame, and the calculating root formula that linear expansion maps is:

$R\left(n\right)=\left\{\begin{array}{cc}R\left(m\right),& n=m=1\\ (1-\frac{\mathrm{nM}}{N}+m)R\left(m\right)+(\frac{\mathrm{nM}}{N}-m)R(m+1),& M\&le;\frac{\mathrm{nN}}{N}\&le;m+1,n=\mathrm{2,3}...,N\end{array}\right.---\left(4.2\right)$

Suppose N < M, then can by T (1), T (2) ..., T (n), ... T (N) } be mapped as a frame sequence by linear expansion method, then calculate the latter and { R (1), R (2), R (m) ..., R (M) } between total distortion.

Mate by simple template the error calculating test sound and reference voice comparatively large, in order to reach optimum efficiency, we adopt the method for dynamic programming.

If test sound pattern has N frame vector, reference template has M frame vector, and N ≠ M.Dynamic time is reformed and will be found a time and to reform function m=ω (n), by the time shaft of test vector being mapped on the time shaft of template non-linearly, and function is met

$D=\underset{\mathrm{\&omega;}\left(n\right)}{\lim }\underset{n-1}{\overset{N}{\mathrm{\&Sigma;}}}d[T\left(n\right),R(\mathrm{\&omega;}\left(n\right)],$

In formula, d [T (n), R (ω (n)] is test template n-th frame vector T (n) and reference template m frame vector R(m) between distance measure; D is then the distance of two vectors in situation of reforming for optimal time.N-th sample of T is aimed at m the sample of R, and obtaining a slope when T with M is identical is the straight line of l.As shown in Figure 4, when T and M is incomplete same, for making m the sample of R aim at n-th sample of T, then corresponding point not on this straight line, and forms a curve, and the function that this curve is corresponding is reformed function exactly.The distance that DTW constantly calculates two vectors to find optimum coupling path, thus ensures the maximum Acoustic Similarity that exists between them.

The condition code of the condition code of extraction and radio and television source signal by extracting the collection of the broadcast television audio signal in transmitting procedure and condition code, and is compared by this method, thus realizes the consistance monitoring to transmitting procedure sound intermediate frequency signal and source signal.

The present invention can be applicable to the information source production mechanism such as broadcasting station, TV station in signals transmission and source signal consistance monitoring.

The substantial effect that the present invention brings is, (1) proposes a kind of audio signal characteristic code verification algorithm, the sound signal of same information source, and in each transmission link of signal, its condition code will be kept; (2) propose a kind of equipment, condition code extraction is carried out to the sound signal in the broadcast TV program in transmitting procedure, carry out verifying with Detection of content consistance with the condition code storehouse of information source.

Accompanying drawing explanation

Fig. 1 is a kind of assay device structures schematic diagram of the present invention;

Fig. 2 is a kind of total system structure of the present invention;

Fig. 3 is that a kind of timesharing of the present invention utilizes detection algorithm schematic diagram;

Fig. 4 is that one of the present invention is dynamically reformed functional arrangement;

In figure: 101, processor module, 102, audio collection module, 103, audio feature code extraction module, 104, mixed-media network modules mixed-media, 105, user interface, 106, memory module, 201 is the audio feature code sequence of the continuous times that signal source A produces, the audio feature code sequence of 301 continuous times produced for signal source B, this partial data is by information source collecting unit system creation; 202,203 ... for the condition code time slice sequence of broadcast television audio signal A in transmission, 302,303 ... for the condition code time slice sequence of broadcast television audio signal B in transmission.202 ', 203 ' be 202,203 check field between feature code sequence timeslice; 302 ', 303 ' be 302,303 check field between feature code sequence timeslice.

Embodiment

Below by embodiment, and by reference to the accompanying drawings, technical scheme of the present invention is described in further detail.

Embodiment: a kind of broadcast television audio signal content consistency detecting method, comprises the following steps:

Step one, condition code extraction is carried out to audio sources, morphogenesis characters code storehouse;

Step 2, signal transmission to be checked is gathered to the audio signal characteristic code of a timeslice, and the timestamp that record gathers;

Step 3, audio signal characteristic code step 2 collected and timestamp are sent to inspection center's system;

The condition code of condition code storehouse before and after timestamp in error time delay of the audio signal characteristic code received and audio sources contrasts by step 4, inspection center's system, detects, then think that signal is correct, otherwise think that signal transmission is wrong.

In step 4, error time delay is no more than 60 seconds.

In step 2, the length of a timeslice is 5 to 60 seconds.

When signal transmission to be checked is multiple signals, time-sharing multiplex mode is adopted to carry out collection audio signal characteristic code to every road suspect signal in turn.

In step one and step 2, the condition code of sound signal is obtained by condition code extraction equipment, and condition code extraction equipment comprises:

Processor module, for controlling whole equipment;

Audio signal sample module, for gathering audio signal data, and by sending to audio feature code extraction module after processor module process;

Audio feature code extraction module, for carrying out vectorized process to the sound signal received, extracts condition code data;

Mixed-media network modules mixed-media, for being sent to center detection system by the timestamp of condition code and extraction condition code.

Described mixed-media network modules mixed-media comprises RJ45 interface and WIFI unit.

The condition code of sound is extracted to comprise and is carried out successively quantizing, pre-emphasis, take advantage of Hamming window and carry out linear predictor coefficient analysis, and the linear regression single order prediction MFCC cepstrum obtained is as sound characteristic code.

Quasi real time monitor broadcasting channel A, the channel B correctness situation at the signal content of A, B, C tri-positions of signal coverage areas.

Deployment signal collections and audio signal characteristic code extraction equipment one respectively three positions, radio station On-premise information source collecting device one and centring system a set of, system architecture is as shown in Figure 2.

Signals collecting and audio signal characteristic code extraction equipment structure as shown in Figure 1, comprise 101 processor modules, 102 audio collection module, 103 audio feature code extraction modules, 104 mixed-media network modules mixed-medias, 105 user interfaces, 106 storage modules.102 audio collection module connect 101 processor modules, and collection signal voice data is mentioned to 101 processor modules; 103 audio feature code extraction modules connect 101 processor modules, receive the voice data of 101 processor modules submissions and carry out audio feature code extraction computing; 104 mixed-media network modules mixed-medias connect 101 processor modules, have been responsible for the data communication between 101 processor modules and centring system, comprise the submission of audio frequency characteristics code data, time stamp data, channel information data etc. and the reception of configuration information; 106 storage modules are connected with 101 processor modules, complete the storage of configuration data.

Processor module 101 is also connected with user interface 105, and user can carry out optimum configurations by user interface 105 pairs of equipment.

Signals collecting and audio signal characteristic code extraction equipment issue configuration information by user interface 105 module or center-side configuration by 104 mixed-media network modules mixed-medias, configures the channel information that this equipment is monitored simultaneously.

Owing to detecting two-way information, in transmission, broadcast television audio signal characteristic code extraction equipment is by time-sharing multiplex sense channel, and timesharing and detection algorithm are as shown in Figure 3.In center-side by source signal collecting unit, gather the information source audio frequency of channel A, B and extract continuous print audio feature code sequence 201 and 301, the signal audio frequency feature code sequence timeslice 202,203 and 302,303 of channel A, B is extracted in signals collecting and the timesharing of audio signal characteristic code extraction equipment, and each timeslice length is fixedly set to T.Condition code data 202 and timeslice start time Ta are passed back to center by signal transmission collection and audio signal characteristic code extraction equipment, the condition code subordinate channel that basis is uploaded by center and time stamp data are from 201, extract a timeslice 202 ', timeslice length is T '=T+2*t, and wherein t is the maximum value possible of signal transmission delay error.The start time of 202 ' is Ta-t, end time is Ta-t+T ', the concept that calculating 202 sequence occurs by center in 202 ', if the probability threshold G1(that its probability of occurrence exceedes setting is generally set to 70% to 80%), to think this moment point, signal transmission conforms to source signal, otherwise thinks that signal transmission is rub-out signal and carries out alert process.

Time-sharing multiplex is specially: assuming that be T to the time of certain road signals collecting monitoring, so the time often crosses T, just switches a signal, as: T is 10 seconds, detect echo signal and have A, B, C are at individual signal, then the 1st is detected a-signal state (comprising acquisition module, vector module) in 10 seconds, 2nd is detected B signal condition in 10 seconds, 3rd is detected C signal condition in 10 seconds, and the 4th is detected a-signal in 10 seconds again, if cycle assignment detection time always.

Specific embodiment described herein is only to the explanation for example of the present invention's spirit.Those skilled in the art can make various amendment or supplement or adopt similar mode to substitute to described specific embodiment, but can't depart from spirit of the present invention or surmount the scope that appended claims defines.

Although more employ the terms such as vector quantization, condition code extraction, timeslice herein, do not get rid of the possibility using other term.These terms are used to be only used to describe and explain essence of the present invention more easily; The restriction that they are construed to any one additional is all contrary with spirit of the present invention.

Claims (6)

1. a broadcast television audio signal content consistency detecting method, is characterized in that comprising the following steps:

Step one, condition code extraction is carried out to audio sources, morphogenesis characters code storehouse;

Step 2, signal transmission to be checked is gathered to the audio signal characteristic code of a timeslice, and the timestamp that record gathers;

Step 3, audio signal characteristic code step 2 collected and timestamp are sent to inspection center's system;

The condition code of condition code storehouse before and after timestamp in error time delay of the audio signal characteristic code received and audio sources contrasts by step 4, inspection center's system, detects, then think that signal is correct, otherwise think that signal transmission is wrong;

Sound characteristic code extracts to comprise and carries out successively quantizing, pre-emphasis, take advantage of Hamming window and carry out linear predictor coefficient analysis, and the linear regression single order prediction MFCC cepstrum obtained is as sound characteristic code;

Comparison process is as follows:

To the training stage, the sound of input is analyzed, and obtains each stack features sequence and is called reference template, be designated as: j=1,2 ... V, in formula, j be template corresponding order numbering: J is the bulk analysis frame number in this order; V is the total template number in system template storehouse; Test template is called to the characteristic sequence that the sound analysis of cognitive phase input obtains, is designated as t={t ₁, t ₂, t ₃..., t _i, I is the frame number inputting sound to be identified;

Each entry characteristic sequence stored in template is called reference model, and a reference model is expressed as R={R (1), R (2) ..., R (m) ..., R (M) }; Wherein, m is the order label of training voiced frame, and m=1 ~ M, M are voiced frame sums, and R (m) is m frame acoustic feature vector; Input entry sound to be identified is called test pattern, and test pattern is expressed as T={T (1), T (2), ..., T (n) ... T (N) }, n is test sound frame number, and comprise N frame sound altogether, T (n) is the n-th frame feature vector;

By calculating the degree of distortion between T and R, differentiate the similarity of T and R; During calculated distortion, from T and R, the distortion of corresponding frame is counted, if n and m is respectively optional frame number in T and R, D [T (n), R (m)] represent the degree of distortion of this two frame feature vector, then obtain the distortion in template between every frame according to different situations, thus calculate total distortion D (T, R);

$D\&lsqb;T,R\&rsqb;=\underset{n=m=1}{\overset{N}{\&Sigma;}}D\&lsqb;T\left(n\right),R\left(m\right)\&rsqb;---\left(4.1\right)$

1) N=M, at this moment calculate the distortion between n=m=N=M frame according to this and get with, namely obtain total distortion;

2) N ≠ M, if N > is M, then by R={R (1), R (2) ..., R (m), R (M) } with linear expansion map a N frame sequence R (1), R (2) ... R (m), ... R (M) }, calculate again the latter with T (1), T (2) ..., T (n), ... T (N) } between total distortion, this calculating is carried out frame by frame according to the first situation, linear expansion map calculating root formula be:

$R\left(n\right)=\left\{\begin{array}{cc}R\left(m\right),& n=m=1\\ (1-\frac{nM}{N}+m)R\left(m\right)+(\frac{nM}{N}-m)R(m+1),& M\&le;\frac{nN}{N}\end{array}\right.\&le;m+1,n=2,3...,N---\left(4.2\right)$

If N < is M, then incite somebody to action T (1), T (2) ..., T (n), ... T (N) } be mapped as a frame sequence by linear expansion method, then calculate the latter and { R (1), R (2), R (m) ..., R (M) } between total distortion;

Test sound pattern has N frame vector, and reference template has M frame vector, and N ≠ M; Dynamic time is reformed and will be found a time and to reform function m=ω (n), by the time shaft of test vector being mapped on the time shaft of template non-linearly, and function is met

$D=\underset{\mathrm{\&omega;}\left(n\right)}{\lim }\underset{n-1}{\overset{N}{\&Sigma;}}d\&lsqb;T\left(n\right),R\left(\mathrm{\&omega;}\right(n)\&rsqb;,$

In formula, d [T (n), R (ω (n)] is the distance measure between test template n-th frame vector T (n) and reference template m frame vector R (m); D is then the distance of two vectors in situation of reforming for optimal time; N-th sample of T is aimed at m the sample of R, and obtaining a slope when T with M is identical is the straight line of l.

2. a kind of broadcast television audio signal content consistency detecting method according to claim 1, it is characterized in that, in step 4, error time delay is no more than 60 seconds.

3. a kind of broadcast television audio signal content consistency detecting method according to claim 1 and 2, is characterized in that, in step 2, the length of a timeslice is 5 to 60 seconds.

4. a kind of broadcast television audio signal content consistency detecting method according to claim 1 and 2, is characterized in that, when signal transmission to be checked is multiple signals, adopts time-sharing multiplex mode to carry out collection audio signal characteristic code to every road suspect signal in turn.

5. a kind of broadcast television audio signal content consistency detecting method according to claim 1, is characterized in that, in step one and step 2, the condition code of sound signal is obtained by condition code extraction equipment, and condition code extraction equipment comprises:

Processor module, for controlling whole equipment;

Audio signal sample module, for gathering audio signal data, and by sending to audio feature code extraction module after processor module process;

Audio feature code extraction module, for carrying out vectorized process to the sound signal received, extracts condition code data;

Mixed-media network modules mixed-media, for being sent to center detection system by the timestamp of condition code and extraction condition code.

6. a kind of broadcast television audio signal content consistency detecting method according to claim 5, it is characterized in that, described mixed-media network modules mixed-media comprises RJ45 interface and WIFI unit.