CN102831893B - Method for rapidly destroying broadcast audio file viruses - Google Patents

Abstract

The invention provides a method for rapidly destroying broadcast audio file viruses, which adjusts the quantized bits of the audio frames in the coded stream one by one, and rewrites the weighted coded audio frames into the coded stream; when participating in the current audio frame In all the channel sub-bands of encoding, first select 1 channel sub-band to adjust the bit allocation minus 1, and then select 1 or more channel sub-bands to perform bit allocation plus 1 adjustment. The present invention does not need to decode the audio frequency into PCM form, but It simply processes samples of a few channel subbands in the compressed domain, its computational complexity is very low, and it can change audio files in a large range, thereby ensuring the destruction of viruses hidden in audio files.

Description

A Quick Way to Destroy Broadcast Audio File Viruses

technical field

The invention belongs to digital audio processing technology, in particular to broadcast audio file processing technology.

Background technique

After decades of development, my country's radio network has gradually formed a network system with broadcast network (intranet) and office network (external network) as the core. In order to ensure the safety of radio business and program broadcasting, the State Administration of Radio, Film and Television stipulates that the internal network of radio stations must be isolated from the external network to prevent malicious attacks and virus infections. However, on the other hand, a large amount of material information such as music programs, interview recordings, and news articles need to be exchanged between the internal and external networks. The closed internal network will seriously affect the work efficiency and program quality of the radio station. In order to solve this problem, the security isolation network gate product came into being and developed rapidly, and has been widely used in domestic radio stations.

Security threats from the Internet mainly include hacker attacks and virus infections. The gatekeeper provides protection against the former hacker attack, but is powerless against the latter virus infection. Viruses usually use files as carriers to spread. Many gatekeeper devices prevent file viruses by checking file extensions, file headers, and other formats. However, these methods cannot fundamentally prevent virus infection, because viruses may be hidden in file data ( such as audio data).

The current audio file format industry standard of my country's radio stations is the MPEG-I second layer format (mp2 for short) in the ISO11172-3 standard. This format file is generated by compressing the original audio data by using an audio data compression coding layer 2 standard method developed by the Moving Pictures Expert Group (MPEG). The mp2 audio file consists of several audio frames, and the frame information of each audio frame includes six parts: frame header, bit allocation, scaling factor selection, scaling factor, channel subband sample codeword and auxiliary data, as shown in Figure 1. Among them, the channel subband sample codeword part contains the encoding information of 32 frequency subband samples of the left and right channels, and the number of bits used for quantizing each channel subband sample is determined by the previous bit allocation part. The larger the value of the bit allocation, the more quantization bits are allocated to the corresponding channel subband, and the more accurate the expression of the sample value of the channel subband is. The channel sub-band mentioned in this document refers to the sub-band of the corresponding channel determined by the channel number and the sub-band number together. For example, the channel sub-band (i, j) is represented as the sub-band corresponding to the channel number i and the sub-band number j.

In the standard, 36 samples of a channel subband are divided into 12 sections (granule), each section contains 3 consecutive samples. As can be seen from Figure 1, the sample codewords of each channel subband are arranged alternately in sections in the audio frame, that is, the first section of each channel subband is arranged first, then their second section, and finally the 12th section . The channel subband sample codeword is the main part of the audio frame, and this part has no verification mechanism and can be tampered without being detected. Therefore, audio files can be excellent vectors for viruses. Viruses hidden in audio files can exploit loopholes in playback software to attack intranet machines, which cannot be prevented by format review methods. Currently, methods for preventing file virus intrusion can be roughly divided into two categories: detecting viruses and destroying viruses.

The method for detecting viruses is to realize defense by detecting viruses. Typical methods are comparative method and anti-tampering method. The comparative method discovers viruses by comparing file data with virus signatures. Similar to anti-virus software, this method needs to constantly update virus characteristics; anti-tampering method is to add authentication or verification code to the file in advance, and judge whether it is infected with a virus by detecting whether the file has been tampered with. It is also very inconvenient to use for processing at the source.

The method of destroying viruses is a better active defense method, and its goal is to directly destroy viruses. This kind of method can be divided into two categories: non-compressed domain processing and compressed domain processing. Uncompressed domain processing first decodes the compressed audio into the original Pulse Code Modulation (PCM) data, then changes the PCM data in a certain way, and finally restores the compressed file through encoding. This method has undergone decoding and re-encoding process, and the computational complexity is relatively high; compressed domain processing directly changes the compressed coded bit stream, such as adding watermarks, changing scaling factors or encoding codewords, etc. Although the compression domain processing method is simple, it often has great damage to the sound quality. In short, the current virus destruction method mainly faces two technical difficulties, that is, how to reduce the computational complexity and how to ensure the sound quality as much as possible while destroying the virus. These two issues have not been well addressed in existing methods.

Contents of the invention

The technical problem to be solved by the invention is to provide a new, fast virus destruction method capable of ensuring the sound quality of broadcast audio files.

The technical scheme adopted by the present invention for solving the above-mentioned technical problems is: a method for rapidly destroying broadcast audio file viruses, which is characterized in that the audio frames in the encoded stream are adjusted to the quantization bits one by one, and the weighted encoded The audio frame of the audio frame is rewritten into the encoded stream, and the bit adjustment of the audio frame includes the following steps:

a. Parsing the frame header and subband encoding information of the current audio frame to obtain the bit allocation and sample codewords of all channel subbands in the current audio frame;

b. Among all the channel subbands participating in the encoding of the current audio frame, first select 1 channel subband for bit allocation minus 1 adjustment, and then select 1 or more channel subbands for bit allocation plus 1 adjustment; the selected channel subband The condition to be met is: the number of sample codeword bits released by the channel sub-bands that are adjusted for bit allocation reduction after the bit allocation is reduced is greater than or equal to the sample codes that all channel sub-bands that are adjusted for bit allocation increase need to consume after the bit allocation is increased The total number of bits of the word, and the sub-band number for which bit allocation reduction adjustment is performed is different from the sub-band number for which adjustment is performed for increase;

c. Decode and dequantize the sample codewords of all channel subbands adjusted by bit allocation to generate frequency samples of the corresponding channel subbands, and then requantize and encode these frequency samples according to the adjusted allocation bits to generate new The sample code word of

d. Generate changed audio frames by organizing all frame information including new bit allocation of channel subbands and new sample codewords according to standards.

It can be seen that the present invention does not need to decode the audio into PCM form, but simply processes samples of a few channel subbands in the compressed domain, and its computational complexity is very low.

Since the sample codewords of different channel subbands in the audio frame are arranged alternately by section, when the sample codeword of a certain channel subband in the audio frame changes, the changed codeword will be distributed in the entire channel of the audio frame The sub-band sample code word area occupies most of the audio frame code stream, so the present invention can change the audio file in a large range, thereby ensuring the destruction of viruses hidden in the audio file.

In addition, the present invention does not change the scaling factor in the audio frame, scaling factor selection and other information that has a large impact on the quantization error, but only adjusts the bit allocation of one channel sub-band by minus 1, so that the loss of sound quality is small. At the same time, the present invention allocates the released bits to other channel subbands as much as possible, that is, adjusts the bit allocation of one or more channel subbands by adding 1, so that the sound quality loss is compensated.

In order to further reduce the loss of sound quality, when selecting the channel subband for bit allocation reduction adjustment, the channel subband with the smallest decrease in Signal to Noise Ratio (SNR) caused by the reduction of bit allocation by 1 is given priority. This is inspired by the bit allocation principle of encoding. In the mp2 standard encoding algorithm, bit allocation is done iteratively. At each iteration, the algorithm assigns 1 unit of bits to the channel subband with the minimum Mask to Noise Ratio (MNR). The masked-to-noise ratio indicates the signal-to-noise ratio after the current signal is masked by other signals, and the smaller the value, the bigger the noise after masking and the worse the sound quality. The masking-to-noise ratio is determined by the difference between the signal-to-noise ratio and the signal-to-mask ratio (Signal to Mask Ratio, SMR), namely:

MNR=SNR-SMR

Among them, the SNR is determined by the quantization bits, which can be directly obtained by looking up the table, while the SMR is calculated according to the acoustic model, and can be regarded as fixed after encoding. Based on this result, this method selects the channel sub-band with the smallest SNR drop, and reduces its bit allocation by 1, which also reduces the bit allocation of the channel sub-band with the smallest MNR drop by 1, that is, from the perspective of the acoustic model, it ensures that the re-bit allocation causes The subjective sound quality loss is minimal.

Further, when selecting the channel sub-band for bit allocation increase adjustment, the channel sub-band with the largest increase in signal-to-noise ratio caused by increasing the bit allocation by 1 is preferentially selected, so that the loss of sound quality can be compensated the most.

Furthermore, in order to make up for the loss of sound quality to the greatest extent, step a also calculates the number of unused bits of the current audio frame at the time of initial encoding; in step b, the condition satisfied by the selected channel subband becomes: perform bit allocation The sum of the number of sample codeword bits released after the reduction of the bit allocation and the number of unused bits in the initial encoding of the channel sub-bands adjusted by the reduction is greater than or equal to the cost of all channel sub-bands that are adjusted by the increase of the bit allocation after the increase of the bit allocation. The total number of bits in the sample codeword.

The beneficial effect of the present invention is that the computational complexity is very low, and the audio file can be changed in a large range to ensure that the virus hidden in the audio file is destroyed; compared with the original file, the subjective sound quality of the processed file is almost No loss; After the bit allocation is adjusted, it cannot be restored, which can prevent the method itself from being hacked; the processed file maintains the original attributes of the original file such as size, channel number, sampling rate, bit rate, etc., and can be Any standard codec decodes.

Description of drawings

Fig. 1 is the bit stream data frame format of the second layer of MPEG-I;

Fig. 2 is a flow chart of the method of the present invention;

Fig. 3 is a comparison of the data of the 2000th frame before and after being processed by the method of the present invention.

Detailed ways

The present invention obtains the original bit allocation information of each channel sub-band from the coded stream, selects the channel sub-band for minor adjustment of the bit allocation according to the psychoacoustic model, decodes and inversely quantizes the sample codewords of the corresponding sub-band and then re-encodes them, Finally, reorganize the synthetic encoded stream and write it into the file, thereby changing the original file data to achieve the purpose of destroying the virus. As shown in Figure 2, the audio file is processed frame by frame, wherein the processing steps for each audio frame include:

1. Parsing the frame header: According to the ISO11172-3 standard, the relevant audio format information is obtained by parsing the audio frame;

2. Analyzing sub-band encoding information: According to the ISO11172-3 standard, the sub-band bit allocation and sample codeword information of each channel is obtained by analyzing the audio frame, where the bit allocation of the j-th sub-band of the i-th channel is expressed as A(i, j ); Calculate the unused number of bits adb in the initial encoding;

3. Select channel subbands and adjust bit allocation: According to the number sublimit of the highest subband currently participating in encoding, select 2 or more channel subbands among all channel subbands whose subband numbers belong to [1, sublimit], and assign them to Record to the selected channel sub-band set M, and adjust the bit allocation of the selected channel sub-band, the specific steps are as follows:

3.1. For each channel sub-band, calculate the decrease in signal-to-noise ratio caused by reducing the current bit allocation by 1, and record it into the key value member of a structure array snr_dec. The structure array also records the channel number ch and subband number sub information corresponding to the key value key; the snr_dec array is sorted according to the size of the key value, from small to large; for items with the same key value, the order is randomly determined;

3.2. For each channel subband except that the bit allocation is 0 and the maximum allocation has been reached, calculate the increase in the signal-to-noise ratio caused by increasing the current bit allocation by 1, and record it to the key value of another structure array snr_enc members. The structure array also records the channel number ch and subband number sub information corresponding to the key value key; the snr_enc array is sorted from large to small according to the size of the key value; for items with the same key value, the sequence is randomly determined;

3.3. Let m and n be the browse subscripts of the arrays snr_dec and snr_enc respectively. Set the initial value m=1, n=1, M is an empty set;

3.4． Take snr_dec[m], the mth item of snr_dec, which corresponds to the channel sub-band with the smallest SNR drop, and set its channel number and sub-band number as snr_dec[m].ch and snr_dec[m].sub respectively, and calculate if The number of bits DEC that can be released after reducing its current bit allocation by 1; calculate the current number of available bits avail_bits as the sum of the current number of releasable bits plus the number of unused bits in the initial encoding, that is, avail_bits=DEC+adb;

3.5． For each item n of snr_enc: take the channel number and subband number of the nth item snr_enc[n], express it as snr_enc[n].ch and snr_enc[n].sub respectively, and calculate if the current bit allocation is increased by 1 Finally, the number of bits to be consumed is ENC, and it is judged that if the number of currently available bits is greater than or equal to the number of bits to be consumed (that is, avail_bits≥ENC) and the subband numbers of the released and consumed bits are different, then select the channel snr_enc[n].ch , The subband snr_enc[n].sub is used as a channel subband that will be adjusted for bit allocation increase, and its band is added to M, and avail_bits is updated at the same time, so that avail_bit=avail_bits-ENC;

3.6 If M is not empty, select the channel snr_dec[m].ch and the subband snr_dec[m].sub as the channel subband that will be adjusted for bit allocation reduction, and add it to M, otherwise set m=m +1, and go to step 3.4 for the next round of selection;

3.7 Subtract 1 from the bit allocation of one channel subband in M for which bit allocation reduction adjustment will be performed; add 1 to the bit allocation of one or more channel subbands in M that will undergo bit allocation increase adjustment;

4. Decoding inverse quantization sample codewords: according to the ISO11172-3 standard, decode and dequantize the sample codewords of all channel subbands in M to obtain corresponding frequency samples;

5. Requantized encoding frequency samples: According to the adjusted allocation bits, the frequency samples of all channel subbands in M are requantized and encoded according to the ISO11172-3 standard to generate new sample codewords;

6. Frame encoding stream formatting: Write all frame information including new bit allocation of channel subbands and new sample codewords into the encoding stream according to the ISO11172-3 standard.

Example

Take the standard 256kbit/s mp2 audio "Fei Xiang-Hometown Cloud.mp2" file as an example of the input original audio. The audio duration is 4 minutes and 24 seconds, and the total number of frames is 11020 frames. In this embodiment, the input audio file is processed in units of coded frames, one frame at a time, until all 11020 frames are processed. Taking 2000 frames as an example, the specific implementation process of the processing is as follows:

1. Parsing the frame header: According to the ISO11172-3 standard, the number of channels obtained from the audio frame analysis is 2, the sampling rate is 48000kHz, the bit rate is 256kbps, the sublimit of the highest subband participating in encoding is 27, and the decoding table is the 0th table, etc. information;

2. Analyzing subband encoding information: According to the ISO11172-3 standard, the bit allocation and sample code words of each channel subband are analyzed from the audio frame, and the value of each channel subband bit allocation array A is {{6, 7, 5 ,7,8,7,6,5,5,4,3,2,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0}, {6, 7, 5, 8, 8, 6, 5, 6, 4, 5, 4, 3, 2, 3, 3, 2, 2, 1, 1, 1, 0, 0, 0, 0, 0 , 0, 0}}, where the two sets of data correspond to the bit allocation of 27 sub-bands in the left and right channels, for example, the value of A(2, 13) is 2, which means that the bit allocation of the current channel sub-band (2, 13) is 2; Calculate the unused number of bits adb of the frame at the time of initial encoding to be 4 bits;

3. Select channel subbands and adjust bit allocation: According to the psychoacoustic model, select channel subbands from 54 candidates from subband 1 to subband 27 including left and right channels, and adjust their bit allocation. The specific steps are as follows:

3.1. For each channel subband, calculate the decrease in the signal-to-noise ratio caused by reducing the current bit allocation by 1, fill in the key value of the structure array snr_dec, the channel number and the subband number 3 members, and use the quick sort method to sort the snr_dec key value Sorting from small to large, the ordered snr_dec is obtained; the value of the first item snr_dec[1] is {4.00, 2, 13}, indicating that the key value (or SNR drop) is 4.00, the channel number is 2 and the subband number is 13. This item corresponds to the current minimum SNR drop.

3.2. For the 11 channel subbands except for the bit allocation of 0 and the maximum allocation, calculate the increase in signal-to-noise ratio caused by increasing the current bit allocation by 1, and fill in the key value, channel number and subband of the structure array snr_enc Number 3 members, use the quick sort method to sort snr_enc key values from large to small, and get ordered snr_enc; the value of the first item snr_enc[1] is {4.84, 1, 11}, indicating the key value (or SNR growth) is 4.84, the channel number is 1 and the subband number is 11, which corresponds to the current maximum SNR growth;

3.3. Set the browse subscript m=1 of the array snr_dec, the browse subscript n=1 of snr_enc, and M is an empty set;

3.4． Examine the mth item of the snr_dec array. For the first time, that is, for the 13th subband of the second channel represented by snr_dec[1], calculate the number of bits DEC that can be released if the bit allocation is reduced from 2 to 1. is 24, and the calculated avail_bits is 28;

3.5． Examine each item n of snr_enc one by one, calculate the number of bits ENC that need to be consumed if the current bit allocation is increased by 1, and judge the condition that avail_bits≥ENC and snr_enc[n].sub≠snr_dec[m].sub. After the investigation, it is found that only the first item meets the conditions, and its corresponding ENC is 12, the channel number is 1, and the subband number is 11, then the 11th subband of the first channel is selected as the channel subband for bit increase adjustment , and the channel subband (1, 11) is added to M, and avail_bits is updated to 16;

3.6 Since M is not empty, select the 13th subband of the second channel as the channel subband for bit reduction adjustment, and add the channel subband (2, 13) to M;

3.7 Decrease the bit allocation of the channel subband (2, 13) by 1, that is, A (2, 13) = A (2, 13)-1; increase the bit allocation of the channel subband (1, 11), that is, A (1,11)=A(1,11)+1;

4. Decoding dequantized sample codewords: according to the ISO11172-3 standard, decode and dequantize the sample codewords of channel subbands (2, 13) and channel subbands (1, 11) to obtain corresponding frequency samples;

5. Requantized encoding frequency samples: According to the updated bit allocation information, re-quantize and encode according to the frequency samples of the ISO11172-3 standard channel subband (2, 13) and channel subband (1, 11) to generate a new sample codeword;

6. Frame encoding stream formatting: According to the ISO11172-3 standard, write all frame information including channel subband new bit allocation and new sample codewords into the encoding stream, and fill the unused 16 bits with 0.

The validity of the present invention is verified from three aspects of file change, sound quality and format maintenance and processing speed. In the link of testing sound quality and format maintenance and processing speed, in addition to the method of the present invention, a kind of reduction scheme will be introduced, that is, remove the 4th and 5th steps of the embodiment (the channel sub-bands that carry out bit allocation adjustment) Inverse quantization decoding of sample codewords and requantization coding steps), that is, the changed channel subbands are not requantized and encoded after bit allocation adjustment, but are directly organized into frames with atomic band sample codewords, so that only the channel subbands are changed Bit allocation information is not enough to destroy viruses. In the sound quality evaluation process, the Perceptual Evaluation of Audio Quality (PEAQ) method described in the standard ITU-R BS.1387 is adopted, which is the method with the highest correlation with the subjective evaluation results among the current objective evaluation methods of audio quality. This method uses the subjective perception characteristics of the human ear to calculate the masking threshold and distortion threshold of the signal, and then uses the artificial neural network to fuse an evaluation parameter ODG (Object Difference Grade), which is used in this implementation to objectively evaluate the audio quality.

1. File changes

First, compare and analyze the processed files with the original files. According to the previous analysis, the data change will occur in the bit allocation of the frame and the channel subband sample codeword, and the size of the data change is mainly related to the difference between the number of bits released by the reduction of bit allocation and the number of bits consumed by the increase of bit allocation diff_bit related. Since the codewords of the subband samples of each channel are arranged alternately in the encoded stream, once the parameter diff_bit is not 0, it will cause a serious misalignment of the processed bit arrangement, resulting in a large-scale change of the frame data.

Figure 3 shows the data comparison of the 2000th frame listed above before and after processing, in which the background of the changed bytes is displayed with shadows. The number of bits released by reducing the frame bit allocation is 24, and the number of bits consumed by increasing the bit allocation is 12, that is, diff_bit=12. It can be seen that the change in the front part of the frame has 2 bytes, corresponding to the bit allocation information of the two channel subbands that are changed. In the channel subband sample codeword area which accounts for most of the audio frame, the data is almost all changed, which is the effect of serious misalignment of the bit arrangement. For the entire frame, the data change rate reaches 84.77%, and the largest continuous area appears in the front part of the frame, and its length is 95 bytes.

2. Sound quality and format maintenance

The same standard decoder is used to decode the original mp2 file and the file processed by the method of the present invention and its reduction method. Table 1 lists the sound quality comparison of the three audio files after decoding, and the evaluation parameter is ODG. The closer the value of this parameter is to 0, the better the sound quality.

Table 1 original file, the sound quality comparison of the file processed by the present invention and reduction method thereof

Original file After the present invention is processed After reduction method processing -3.326 -3.330 -3.354

It can be seen that the sound quality difference between the audio after processing by the method of the present invention and the original audio is very small, and because the quantization bit number corresponding to the original sample codeword is not consistent with the new bit allocation in the reduction method, the sound quality after it is processed has a slight difference. Decline. In addition, the method of the present invention does not change the frame header of the audio frame of the original file, and keeps the frame size unchanged, so the processed file keeps the original attributes such as the size of the original file, the number of channels, the sampling rate, and the bit rate.

3. Processing speed

Using the method of the present invention and its reduction method to transcode the test audio file, Table 2 lists the results of the corresponding processing time (in seconds), which includes the time used for reading and writing files. The test machine is an Intel(R) i3 processor with a main frequency of 2.53GHz. For the convenience of comparison, Table 2 also lists the processing time for standard encoding of the original uncompressed audio file corresponding to the test file and standard decoding of the original mp2 file.

Table 2 The present invention and its reducing method and the processing time comparison of four kinds of processing of encoding and decoding

encoding decode The present invention deals with Reduced method processing 6.328 4.282 0.406 0.360

It can be seen that the processing speed of the reduction method is the fastest among the processing. Compared with encoding and decoding, the processing speed of the method of the present invention is also very fast, which is 15.59 and 10.55 times of encoding and decoding respectively, and the audio frequency of 4 minutes and 24 seconds is completed in only 0.406 seconds, and the real-time speed of its processing is higher than that of (total audio time / processing time) up to 650.25.

Claims (4)

1. the method for a rapid damage broadcast audio file virus, it is characterized in that, one by one the audio frame in encoding stream is carried out the adjustment of quantization bit position, the audio frame after heavy quantization encoding is re-write in encoding stream, audio frame is carried out to bit adjustment and comprise the following steps:

A. resolve frame head and the sub-band coding information of current audio frame, obtain the position of all passage subbands in current audio frame and distribute and sample code word;

B. in all passage subbands that participate in current audio frame coding, first select 1 passage subband to carry out position distribution and subtract 1 adjustment, then select one or more passage subbands to carry out position and divide with addition of 1 adjustment; The satisfied condition of selecteed passage subband is: carry out position and distribute and reduce the sample code word bits number discharging after a passage subband distribution in place of adjusting reduces and be more than or equal to and carry out position and distribute to increase after the distribution in place of all passage subbands of adjusting increases and need total bit number of the sample code word expending, and carry out position and distribute and reduce the subband numbering of adjusting and number different with the subband that increases adjustment;

C. to distributing the sample code word of all passage subbands of adjusting to decode and inverse quantization through position, generate the frequency samples of respective channel subband, then according to point coordination after adjusting, these frequency samples are carried out to re-quantization and coding, generate new sample code word;

D. by comprising, the new position of passage subband is distributed and all frame informations of new samples code word generate the audio frame after changing by normal structure.

2. a kind of method of rapid damage broadcast audio file virus as claimed in claim 1, is characterized in that, distributes while reducing the passage subband of adjusting selecting to carry out position, and a passage subband that reduces by 1 rear caused signal noise ratio slippage minimum is distributed in the preferential position of selecting.

3. a kind of method of rapid damage broadcast audio file virus as claimed in claim 2, it is characterized in that, in the time selecting to carry out the passage subband of position distribution increase adjustment, a preferential passage subband for the increment maximum of distribution increase by 1 caused signal noise ratio of selecting.

4. a kind of method of rapid damage broadcast audio file virus as claimed in claim 1, is characterized in that, step a also calculates current audio frame unspent bit number in the time of initial coding; In step b, the satisfied condition of selecteed passage subband becomes: carry out position and distribute while reducing the sample code word bits number discharging after the passage subband distribution in place of adjusting reduces with initial coding unspent bit number sum to be more than or equal to carry out position to distribute to increase the total bit number that needs the sample code word expending after the distribution in place of all passage subbands of adjusting increases.