An adaptive audio information hiding method

Abstract

The present invention proposes an adaptive audio information hiding method, comprising: step 1, at the transmitting end, after the original audio signal is divided into multiple frames of short-time signals, the block type to which the signal belongs and its masking threshold in the entire frequency band are determined; a window function is selected for windowing processing and then time-frequency conversion is performed to obtain a frequency domain signal, and the embedding strength of each frequency point of the frequency domain signal is determined in combination with the masking threshold curve obtained by the psychoacoustic model, and the information is embedded in each frequency point of the signal to complete information embedding and obtain audio containing hidden information; step 2, the audio containing hidden information is transmitted to the receiving end through a channel; step 3, at the receiving end, the embedded information in the received audio is extracted. The method introduces an AAC encoder at the transmitting end to analyze the signal, and the analysis of the carrier signal characteristics is more detailed, so that the information embedding is more adapted to the characteristics of the signal itself, the modification range of the original audio is smaller, the sound quality loss is smaller, and it has better imperceptibility.

Description

Technical Field

[0001] The present invention relates to the technical field of covert communication, and in particular to an audio information adaptive hiding method. Background Art

[0002] Audio information hiding technology uses audio as a carrier, utilizes the characteristics of the human auditory system, and the redundancy of the audio signal itself, to hide information in the carrier signal in an imperceptible way. In recent years, the transmission and release of audio content on the Internet has brought about a series of problems and application needs, which can be generally divided into two parts: the copyright protection of audio digital content and the covert communication problem in special scenarios. Both of these research issues belong to the research scope of audio information hiding. Using information hiding technology to embed copyright and other authentication information into the audio carrier can achieve copyright protection of digital content. This type of application scenario mainly emphasizes that the embedded information cannot be perceived by people and does not affect the original value of the carrier, and the security requirements for the technology are generally low; covert communication is mostly used to transmit secret information. The secret information to be transmitted is embedded in the audio carrier and sent to the receiver. The receiver extracts the secret information from the received signal. This situation requires not only that the embedded information cannot be perceived, but also that the algorithm has good anti-interference and anti-attack capabilities. Traditional encryption methods often rely on the security of the key to ensure the security of the system, which is easily destroyed by interceptors. With the development and popularization of Internet technology, its application defects in digital content management and protection are gradually exposed. Compared with traditional encryption technology, information hiding has an extra layer of protection, hiding the existence of information, further improving the security of information, and has become a research hotspot in the field of information security.

[0003] Most of the existing frequency-domain based low-perceptible information embedding methods use a point-by-point embedding method, which directly multiplies the amplitude or power of each frequency point by the signal-to-mask ratio (SMR) of its corresponding critical frequency band as the embedding strength of each frequency point.

[0004] For example, the paper A Spread Spectrum Audio Watermarking System with High Perceptual Quality proposed an audio information hiding algorithm based on spread spectrum theory. The transmitter uses DFT to convert the time domain signal to the frequency domain, determines the embedding strength of each critical frequency band according to the psychoacoustic model, and embeds the spread information bits into the audio carrier. The receiver uses the autocorrelation characteristics of the spread spectrum sequence to correlate the received signal with the spread spectrum sequence to despread the signal and extract the hidden information.

[0005] The prior art has the following technical problems in actual application:

[0006] The point-by-point embedding method determines the embedding strength according to the amplitude of each frequency point and the SMR of its corresponding critical frequency band. However, since the SMR is calculated in units of critical frequency bands in the psychoacoustic model, it cannot fully represent the maximum allowable distortion of a single frequency point. Therefore, although this method meets the overall requirement that the embedding energy in each frequency band is lower than the masking threshold, it is still possible that the embedding of a single frequency point exceeds its actual masking threshold, thereby causing serious damage to the audio quality. In addition, when converting time to frequency, this method has a fixed division length for the time domain signal, so it has the same frequency domain resolution and time domain resolution for stationary signals and transient signals. The adaptability of information embedding to the carrier signal is poor, and the concealment of the algorithm still needs to be improved. In addition, due to the spectral symmetry of the DFT transform, the actual transform domain coefficients available for embedding information are half the length of the time domain signal. When the hidden capacity is high, the robustness of the algorithm will be greatly reduced.

[0007] Based on the above problems existing in the prior art, the present invention provides an audio information adaptive hiding method. Summary of the invention

[0008] The present invention provides an audio information adaptive hiding method.

[0009] The present invention adopts the following technical solutions:

[0010] An audio information adaptive hiding method, comprising:

[0011] Step 1, at the transmitting end, after the original audio signal is divided into multiple frames of short-time signals, one path enters the psychoacoustic model to determine the block type to which the signal belongs and its masking threshold in the entire frequency band; the other path enters the filter module, and a window function is selected according to the block type determined by the psychoacoustic model for windowing processing and then time-frequency conversion is performed to obtain a frequency domain signal, wherein the information to be embedded is divided into hidden information and synchronization information, wherein the hidden information is the information to be embedded, which is converted into binary representation after encoding, and the synchronization information is used for synchronization of the receiving end, and the information bits are spread spectrum into a group of spread spectrum codewords, and the embedding strength of each frequency point of the frequency domain signal is determined in combination with the masking threshold curve obtained by the psychoacoustic model, and embedded at each frequency point of the signal to obtain the frequency domain signal after the embedded information is obtained, and the time domain audio signal is inversely transformed back through the filter module to complete the information embedding and obtain the audio containing hidden information;

[0012] Step 2, the audio containing the hidden information is transmitted through the channel to the receiving end;

[0013] Step 3, at the receiving end, the audio containing hidden information is framed to form multiple frames of short-time signals, one of which enters the psychoacoustic model to determine the block type to which the signal belongs; the other enters the filter module, and a window function is selected according to the block type determined by the psychoacoustic model for windowing processing and then time-frequency conversion is performed to obtain a frequency domain signal with embedded information, which is despread by correlation operation with the same spread spectrum sequence p as the transmitting end, and the hidden information bits obtained by despreading are decoded using a decoding method corresponding to the encoding at the transmitting end, so that the embedded information in the received audio can be extracted.

[0014] Furthermore, in step 1, the information bit "1" to be embedded is mapped to p using the spreading sequence p, and "-1" is mapped to -p.

[0015] Furthermore, in step 1, the psychoacoustic model module in the AAC audio coding is used to analyze the characteristics of the audio carrier signal, and the signal is divided into long blocks and short blocks according to its own characteristics. Each frame of the audio signal is divided into corresponding multiple scale factor bands according to the sampling rate and the long block and short block types; the minimum masking ratio of the frame signal in each scale factor band is calculated according to the frequency domain masking characteristics as a basis for embedding hidden information.

[0016] Furthermore, in step 1, the filter module is a time-frequency transform processing module for AAC audio coding, and adopts an improved discrete cosine transform MDCT, which uses a time domain aliasing elimination TDAC technology. The calculation formulas of the improved discrete cosine transform MDCT and the inverse modified discrete cosine transform IMDCT are as follows:

[0017]

[0018]

[0019] In the above formula 1-2, x(n) is the original signal, X(k) is the MDCT frequency value, N is the number of transformation points, and n and k are the time domain and frequency domain signal indexes respectively.

[0020] Furthermore, the embedding strength of the signal at each frequency point is determined by referring to the non-uniform quantization principle in AAC audio coding, including:

[0021] Step 1.11, calculate the maximum allowable quantization noise in the current scale factor band according to the signal frequency value and the masking ratio in the current scale factor band calculated by the psychoacoustic model:

[0022]

[0023]

[0024] In the above formula 3-4, sfb represents the scale factor band index, low(sfb) and high(sfb) are the start and end indices of the spectrum line in sfb, respectively, and Enrg sfb is the total energy in the scale factor frequency band sfb, maxNoise sfb is the maximum allowed quantization noise in the scale factor band sfb, X(k) is the MDCT frequency value at index k, SMR sfb is the masking ratio within the scale factor band sfb;

[0025] Step 1.12, initialize the quantization step size to ensure that the current quantization noise is less than the maximum allowable distortion;

[0026] Step 1.13, non-uniformly quantize each frequency point within the current scale factor band, and the quantization formula is:

[0027]

[0028] In the above formula 5, X(k) is the signal frequency value, sgn(·) is the sign function, +1 represents a non-negative number, -1 represents a negative number, int(·) is the rounding function, stepsize is the quantization step size, and X quant (k) is the quantized value;

[0029] Step 1.14, perform inverse quantization according to the quantized value of each frequency point:

[0030]

[0031] Step 1.15, calculate the quantization noise in the scale factor band;

[0032] Step 1.16, if the current quantization noise is less than the maximum allowable quantization noise, save the current quantization step size, continue to increase the quantization step size, and repeat steps 1.13 to 1.15. Otherwise, the previously saved quantization step size is the optimal quantization step size within the current scale factor frequency band;

[0033] Step 1.17, determine the embedding strength of each frequency point according to the optimal quantization step size, and the calculation formula is:

[0034] δ(k)=|X' invquant (k)-X(k)|……(7),

[0035] In the above formula 7, X i ' nvquant (k) is the inverse quantized value quantized according to the optimal quantization step size, and δ(k) is the embedding strength corresponding to each frequency point.

[0036] Furthermore, in step 1, information embedding includes: the transmitting end converts the information bits after channel coding into a group of information bits with opposite polarities and embeds them into the previous frame and the next frame of the carrier signal, which is expressed as:

[0037] Y1=sgn(X1)·(|X1|+b·p·δ1)......(8),

[0038] Y2=sgn(X2)·(|X2|-b·p·δ2)......(9),

[0039] In the above formulas 8-9, X1 and X2 represent the spectral coefficients of the previous frame and the next frame signals respectively, b is the information bit to be embedded, p is the spread spectrum sequence, δ1 and δ2 are the embedding strengths corresponding to each frequency point of the previous frame and the next frame, and Y1 and Y2 are the frequency values ​​of the previous frame and the next frame signals after the information is embedded.

[0040] Furthermore, in step 3, the calculation formula of the detection decision amount r of the correlation operation is as follows:

[0041] r=(Y2'-Y1')·p

[0042] =(|X2|-|X1|)·p+b·p·(δ2+δ1)·p+(n2-n1)·p......(10),

[0043] In the above formula 10, Y′1 and Y′2 are the frequency values ​​of the previous and next audio frames received after being mixed with noise n1 and n2 through the channel transmission, and p is the spread spectrum sequence.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. The method for adaptively hiding audio information of the present invention analyzes the signal by introducing an AAC encoder in the transmitting end, adaptively switches the window function and performs time-frequency conversion according to the characteristics of the carrier signal, and dynamically adjusts the embedding strength by non-uniform quantization according to the masking threshold calculated by the psychoacoustic model. Compared with the prior art that directly determines the embedding strength of the signal at each frequency point according to the masking threshold calculated by the psychoacoustic model, the method has a more detailed analysis of the characteristics of the carrier signal, so that the embedding of information is more adapted to the characteristics of the signal itself, the modification of the original audio is smaller, the sound quality loss is smaller, and the method has better imperceptibility;

[0046] 2. The audio information adaptive hiding method described in the present invention combines spread spectrum theory with AAC audio coding. The information to be embedded is embedded in the frequency domain signal after spread spectrum processing. The information to be embedded is expanded to a very wide frequency band, so that the anti-interference ability of the method is improved. AAC audio coding uses MDCT for time-frequency transformation. Compared with DFT, more spectral coefficients of information can be embedded. Therefore, a higher hiding capacity can be achieved without reducing robustness. In addition, by utilizing non-uniform embedding to obtain the embedding strength that guarantees audio quality and maximizes the embedding energy, large signals are embedded with less information and small signals are embedded with more information. It can also ensure that the absolute embedding energy of large signals is greater than that of small signals, and has better anti-noise ability, effectively resolving the conflict between robustness and hiding capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a processing flow chart of the sending end in an embodiment of the present invention;

[0048] Figure 2 is a processing flow chart of the receiving end in an embodiment of the present invention;

[0049] Figure 3 is a flow chart of the calculation of the psychoacoustic model in an embodiment of the present invention;

[0050] Figure 4 is a schematic diagram of switching between different window types in an embodiment of the present invention;

[0051] Figure 5 A flowchart of determining embedding strength in an embodiment of the present invention;

[0052] Figure 6 is a comparison diagram of ODG value changes under different SWRs in an embodiment of the present invention;

[0053] Figure 7 4 is a comparison diagram of the relationship between the bit error rate and the signal-to-noise ratio in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0055] Example

[0056] In this embodiment, the audio information hiding system is composed of three parts: the transmitter, the channel, and the receiver;

[0057] like Figure 1As shown in the figure, it is a structural diagram of the transmitting end. The original audio signal is divided into multiple frames of short-time signals through framing. One path enters the psychoacoustic model to determine the window function selected by the filter module and the masking threshold of each subframe signal in the entire frequency band; the other path enters the filter module, and the time-frequency conversion is completed after windowing processing according to the result of the psychoacoustic model to obtain the signal frequency value, and then the embedding strength of the signal at each frequency point is determined in combination with the masking threshold curve. Among them, the information to be embedded is divided into synchronization information and codeword information. The synchronization information is used for synchronization operation during reception, and the codeword information is the data information to be transmitted. The information bits are spread into a set of spread spectrum codewords, and the spectral coefficients of the signal are modified according to the determined embedding strength and concealed information. Finally, it is inversely transformed back to the time domain audio signal through the filter module to complete the information embedding.

[0058] like Figure 2 As shown in the figure, it is a structural diagram of the receiving end. After the audio containing hidden information is transmitted through the channel and enters the receiving system, similar to the initial processing at the transmitting end, the received signal needs to be framed and time-frequency transformed to obtain the frequency domain signal, and then the correlation operation is performed with the same spread spectrum sequence p as the transmitting end to perform despreading. The despreading result is input to the decoding end for desynchronization or codeword information;

[0059] In the above embodiment, frequency domain spread spectrum technology is adopted, and the information bit "1" to be embedded is mapped to p and "-1" is mapped to -p using the spread spectrum sequence, wherein the length of the spread spectrum sequence is selected related to the hidden capacity of the algorithm, and the length of the spread spectrum sequence p is selected related to the hidden capacity of the algorithm. The shorter the spread spectrum sequence, the higher the hidden capacity.

[0060] In the above embodiment, the psychoacoustic model adopts the psychoacoustic model based on ISO / IEC 13818. The psychoacoustic model has two functions. One is to determine the block type of each frame of audio signal and output it to the filter bank as an input parameter of time-frequency transformation or frequency-time transformation. According to the difference between steady-state signal and transient signal, the adaptive window switching technology is used to dynamically divide the audio signal into long blocks and short blocks. The long block signal has a long duration, high frequency resolution and high coding efficiency, while the short block has a short duration and has a higher time domain resolution, which can effectively suppress the generation of pre-echo. The second is to divide each frame of audio signal into corresponding multiple scale factor bands according to its sampling rate and long and short block types, and calculate the minimum masking ratio of the frame signal in each scale factor band in combination with the frequency domain masking characteristics as an important basis for embedding hidden information. The calculation process of the psychoacoustic model is as follows: Figure 3 shown.

[0061] In the above embodiment, the main task of the filter module is to take a time domain or frequency domain signal of appropriate length, and perform windowing processing with an appropriate window function, and finally perform time-frequency conversion (before embedding) or frequency-time conversion (after embedding). Considering the influence of the window function on the frequency response of the filter, the window functions used in this module are sine window and Kaiser-Bessel window (KBD window). When the intervals between the spectral components of the signal are small, the use of the sine window can better separate the adjacent frequency components, and the Kaiser-Bessel window is suitable for signals with wide intervals between the spectral components. The switching of long and short blocks in this module is mainly manifested by the switching of window functions. The switching standard is determined according to the calculation results of the psychoacoustic model. In addition, in order to smoothly transition between long windows and short windows, a transition block is introduced in the middle. The specific switching rules between windows are as follows: Figure 4 shown.

[0062] In the above embodiment, the transformation method mainly used by the filter module is the modified discrete cosine transform (MDCT), which is a linear orthogonal overlapped transform and uses a technology called time domain aliasing cancellation (TDAC). All blocks overlap each other by 50%, and aliasing can be completely eliminated in theory. The specific MDCT and IMDCT calculation formulas are as follows:

[0063]

[0064]

[0065] Where x(n) is the original signal, X(k) is the MDCT frequency value, N is the number of transformation points, and n and k are the time domain and frequency domain signal indexes respectively;

[0066] In the above embodiment, in order to solve the conflict between the concealment and robustness of the system, the algorithm of the present technical solution uses the non-uniform quantization module in the AAC encoder to dynamically adjust the embedding strength. The specific flow chart is as follows: Figure 5 As shown, the embedding strength of each frame signal is calculated in units of scale factor frequency band, and the specific process is:

[0067] 1. Calculate the maximum allowable quantization noise in the current scale factor band based on the signal frequency value and the masking ratio in the current scale factor band calculated by the psychoacoustic model:

[0068]

[0069]

[0070] Among them, sfb represents the scale factor band index, low(sfb) and high(sfb) are the start and end indices of the spectrum line in sfb, respectively, and Enrg sfbis the total energy in the scale factor frequency band sfb, maxNoise sfb is the maximum allowed quantization noise in the scale factor band sfb, X(k) is the MDCT frequency value at index k, SMR sfb is the masking ratio within the scale factor band sfb;

[0071] 2. Initialize the quantization step size to ensure that the current quantization noise is less than the maximum allowable distortion;

[0072] 3. Perform non-uniform quantization on each frequency point within the current scale factor band. The quantization formula is:

[0073]

[0074] Among them, X(k) is the signal frequency value, sgn(·) is the sign function, +1 represents a non-negative number, -1 represents a negative number, int(·) is the rounding function, stepsize is the quantization step size, X quant (k) is the quantized value;

[0075] 4. Dequantize according to the quantized value of each frequency point:

[0076]

[0077] 5. Calculate the quantization noise within the scale factor band;

[0078] 6. If the current quantization noise is less than the maximum allowable quantization noise, save the current quantization step size, continue to increase the quantization step size, and repeat steps 3-5. Otherwise, the previously saved quantization step size is the optimal quantization step size within the current scale factor frequency band.

[0079] 7. Determine the embedding strength of each frequency point according to the optimal quantization step size. The calculation formula is:

[0080] δ(k)=|X' invquant (k)-X(k)|,

[0081] Among them, X' invquant (k) is the inverse quantization value quantized according to the optimal quantization step size, δ(k) is the embedding strength corresponding to each frequency point, if there are some frequencies quantized to 0, the non-embedding method is adopted to ensure the audio quality after embedding;

[0082] In the above embodiment, in the blind detection system, the receiving end cannot know the original audio signal a priori, and the correlation between the original audio and the spread spectrum sequence cannot be known. Therefore, when performing the despreading correlation operation, the carrier audio itself also becomes an interference item. Considering the short-term stationary characteristics of the audio signal, a differential embedding method is used when embedding information to suppress the interference of the carrier audio itself to ensure the decoding performance of the receiving end. That is, the transmitting end converts the information bits after channel coding into a group of information bits with opposite polarities and embeds them into the two frames before and after the carrier signal, which can be expressed as:

[0083] Y1=sgn(X1)·(|X1|+b·p·δ1),

[0084] Y2=sgn(X2)·(|X2|-b·p·δ2),

[0085] Among them, X1 and X2 represent the spectrum coefficients of the previous and next frame signals respectively, b is the information bit to be embedded, p is the spread spectrum sequence, δ1 and δ2 are the embedding strengths corresponding to the frequency points of the previous and next frames, and Y1 and Y2 are the frequency values ​​of the previous and next frame signals after the information is embedded;

[0086] In the above embodiment, the receiving end divides the received audio signal into frames and performs time-frequency conversion to obtain the frequency value of each frame signal. Since the transmitting end adopts the differential embedding method, the receiving end can extract the embedded information in the received audio by subtracting the frequency values ​​of the previous and next frame signals, performing correlation operation with the spread spectrum sequence, despreading, and then decoding. The calculation formula of the detection decision amount r of the correlation operation is as follows:

[0087] r=(Y2'-Y1')·p

[0088] =(|X2|-|X1|)·p+b·p·(δ2+δ1)·p+(n2-n1)·p,

[0089] Among them, Y′1 and Y′2 are the frequency values ​​of the audio frame signals received after being transmitted through the channel and mixed with noise n1 and n2, and p is the spread spectrum sequence;

[0090] Since the frame length is very small for the entire signal, it can be considered that the two frames of signal have short-term stability, that is, X1≈X2. Similarly, the noise in the transmission process can also be considered to be approximately equal, n1≈n2. Then, the detection decision amount r can be approximated as:

[0091] r≈b·p·(δ2+δ1)·p,

[0092] Therefore, the embedded information bit can be extracted according to the polarity of the decision value r. If r is positive, the information bit b embedded in the current audio frame is judged to be "+1", otherwise it is "-1".

[0093] In order to verify the superior technical effect of the audio information adaptive hiding method, the existing technology is used to compare with the audio information adaptive hiding method. The existing technology adopts the information hiding algorithm of the paper A Spread Spectrum Audio Watermarking System with High Perceptual Quality (hereinafter referred to as: Algorithm 1);

[0094] The performance of the two algorithms is compared from two perspectives. The first is the imperceptibility of the embedded audio quality, which is measured by the Objective Difference Grade (ODG); the second is the decoding performance, which uses the bit error rate to measure the system's ability to resist noise attacks. The simulation results and analysis are as follows:

[0095] Table 1 Relationship between ODG value and sound quality

[0096] ODG Sound quality describe 0 excellent Unnoticeable changes -1 good Noticeable but not obtrusive -2 generally Slightly harsh -3 Poor Harsh -4 Difference Very harsh

[0097] As shown in Table 1, the larger the ODG value, the better the sound quality and the higher the imperceptibility. Figure 6 The ODG values ​​of the two information hiding methods measured by the perceptual evaluation of audio quality (PEAQ) method at different signal-watermark ratios (SWR) are expressed. By comparing Table 1, it can be seen that when the SWR is similar, the ODG value of the audio information adaptive hiding method is the highest, and when the SWR is large, the ODG difference between the two methods is more obvious. The algorithm of the audio information adaptive hiding method uses the AAC encoding method to analyze the characteristics of the carrier signal, and uses non-uniform quantization to dynamically adjust the embedding strength of each frequency point, so as to realize the adaptive embedding of the hidden information, so that the audio quality after embedding is more imperceptible;

[0098] In order to objectively and impartially evaluate the algorithm's ability to resist common signal processing and malicious attacks, the robustness test should be carried out under the condition of fixed imperceptibility and hidden capacity. By adjusting the algorithm-related parameters, the SWR of the two algorithms is approximately the same, about 15dB, and the hidden capacity is 172.27bps. The audio of the two algorithms after embedding information is processed with noise to test their decoding capabilities in a noisy environment. Figure 7 Expresses the relationship between the bit error rate at the receiving end and the signal-to-noise ratio under different noise environments;

[0099] It can be seen that under the same embedding strength, as the signal-to-noise ratio increases, the bit error rates of the two algorithms show a downward trend. That is to say, as the intensity of Gaussian white noise increases, the decoding performance of the algorithm decreases accordingly. However, the bit error rate of the audio information adaptive hiding method is lower than that of Algorithm 1 under any signal-to-noise ratio strength. Due to the spectral symmetry of the DFT transform, the spectrum coefficients that can actually be used to embed information in Algorithm 1 are half the length of the time domain signal. At the same hiding capacity, only a shorter spread spectrum sequence can be selected. The audio information adaptive hiding method uses MDCT for time-frequency transform, and each transform has a 50% overlap with the previous frame signal. Therefore, more spectrum coefficients can be embedded in the information. Under the same imperceptibility and hiding capacity, the method proposed in this embodiment has a stronger ability to resist noise interference.

[0100] The present invention is not limited by the above embodiments. The above embodiments and descriptions are only for explaining the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention may be subject to various changes and improvements, which are within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for adaptively hiding audio information, characterized in that: include: Step 1: At the transmitting end, the original audio signal is divided into multiple short-time frames and then enters the psychoacoustic model to determine the block type to which the signal belongs and its masking threshold in the entire frequency band; The other path enters the filter module, selects the window function according to the block type determined by the psychoacoustic model for windowing processing, and then performs time-frequency conversion to obtain the frequency domain signal, where the information to be embedded is divided into hidden information and synchronization information. The hidden information is the information that is actually to be embedded, which is converted into binary representation after encoding, and the synchronization information is used for synchronization of the receiving end. The information bits are spread spectrum into a set of spread spectrum codewords, and the embedding strength of each frequency point of the frequency domain signal is determined by combining the masking threshold curve obtained by the psychoacoustic model. The frequency domain signal after the embedded information is obtained is embedded at each frequency point of the signal, and the time domain audio signal is inversely transformed through the filter module to complete the information embedding and obtain. For audio containing hidden information, the psychoacoustic model module analyzes the characteristics of the audio carrier signal, divides it into long blocks and short blocks according to the characteristics of the signal itself, and divides each frame of audio signal into corresponding multiple scale factor bands according to the sampling rate and the types of long blocks and short blocks; the minimum masking ratio of the frame signal in each scale factor band is calculated according to the frequency domain masking characteristics as the basis for embedding hidden information. The filter module uses the improved discrete cosine transform MDCT, which uses the time domain aliasing elimination TDAC technology. Among them, the calculation formulas of the improved discrete cosine transform MDCT and the inverse modified discrete cosine transform IMDCT are as follows: In the above equations (1)-(2), x(n) is the original signal, X(k) is the MDCT frequency value, N is the number of transformation points, and n and k are the time domain and frequency domain signal indexes respectively; Step 2, the audio containing the hidden information is transmitted through the channel to the receiving end; Step 3, at the receiving end, the audio containing hidden information is framed to form multiple frames of short-time signals, one of which enters the psychoacoustic model to determine the block type to which the signal belongs; the other enters the filter module, and a window function is selected according to the block type determined by the psychoacoustic model for windowing processing and then time-frequency conversion is performed to obtain a frequency domain signal with embedded information, which is despread by correlation operation with the same spread spectrum sequence p as the transmitting end, and the hidden information bits obtained by despreading are decoded using a decoding method corresponding to the encoding at the transmitting end, so that the embedded information in the received audio can be extracted.

2. The method for adaptively hiding audio information according to claim 1, characterized in that: In step 1, the information bit "1" to be embedded is mapped to p and "-1" is mapped to -p using the spreading sequence p.

3. The method for adaptively hiding audio information according to claim 1, characterized in that: Determining the signal embedding strength at each frequency point includes: Step 1.11, calculate the maximum allowable quantization noise in the current scale factor band according to the signal frequency value and the masking ratio in the current scale factor band calculated by the psychoacoustic model: In the above formulas (3)-(4), sfb represents the scale factor band index, low(sfb) and high(sfb) are the start and end indices of the spectrum line in sfb, respectively, and Enrg sfb is the total energy in the scale factor frequency band sfb, maxNoise sfb is the maximum allowed quantization noise in the scale factor band sfb, X(k) is the MDCT frequency value at index k, SMR sfb is the masking ratio within the scale factor band sfb; Step 1.12, initialize the quantization step size to ensure that the current quantization noise is less than the maximum allowable distortion; Step 1.13, non-uniformly quantize each frequency point within the current scale factor band, and the quantization formula is: In the above formula (5), X(k) is the signal frequency value, sgn(·) is the sign function, +1 represents a non-negative number, -1 represents a negative number, int(·) is the rounding function, stepsize is the quantization step size, and X quant (k) is the quantized value; Step 1.14, perform inverse quantization according to the quantized value of each frequency point: Step 1.15, calculate the quantization noise in the scale factor band; Step 1.16, if the current quantization noise is less than the maximum allowable quantization noise, save the current quantization step size, continue to increase the quantization step size, and repeat steps 1.13 to 1.

15. Otherwise, the previously saved quantization step size is the optimal quantization step size within the current scale factor frequency band; Step 1.17, determine the embedding strength of each frequency point according to the optimal quantization step size, and the calculation formula is: δ(k)=|X′ invquant (k)-X(k)|……(7), In the above formula (7), X′ invquant (k) is the inverse quantized value quantized according to the optimal quantization step size, and δ(k) is the embedding strength corresponding to each frequency point.

4. The method for adaptively hiding audio information according to claim 1, characterized in that: In step 1, information embedding includes: the transmitting end converts the information bits after channel coding into a group of information bits with opposite polarities and embeds them into the previous frame and the next frame of the carrier signal, which is expressed as: Y1=sgn(X1)·(|X1|+b·p·δ1)......(8), Y2=sgn(X2)·(|X2|-b·p·δ2)......(9), In the above equations (8)-(9), X1 and X2 represent the spectral coefficients of the previous frame and the next frame signals respectively, b is the information bit to be embedded, p is the spread spectrum sequence, δ1 and δ2 are the embedding strengths corresponding to the frequency points of the previous frame and the next frame, and Y1 and Y2 are the frequency values ​​of the previous frame and the next frame signals after the information is embedded.

5. The method for adaptively hiding audio information according to claim 4, characterized in that: In step 3, the detection decision value r of the correlation operation b The calculation formula is as follows: In the above formula (10), Y'1 and Y'2 are the frequency values ​​of the previous and next audio frames received after being mixed with noise n1 and n2 through the channel transmission, and p is the spread spectrum sequence.

Images