CN101256773A - Vector Quantization Method and Device for Frequency Parameters of Immittance Spectrum - Google Patents

Abstract

The invention discloses a vector quantization method of the impedance spectrum frequency (ISF) parameter and a vector quantization device of the ISF parameter. The invention also discloses an ISF parameter vector quantization encoding end method, an ISF parameter vector quantization decoding end method, an ISF parameter vector quantization device, an ISF parameter vector quantization encoding end device and an ISF Parameter vector quantization of the decoding end device. The above method or device uses the prediction coefficients of inter-frame predictors with non-equal coefficients in each dimension to calculate the prediction vector of ISF parameters. The application of the present invention can equalize the quantization performance when there are lost frames and no lost frames in the wideband speech coding and decoding process, and improve the quality of speech synthesis.

Description

Vector Quantization Method and Device for Frequency Parameters of Immittance Spectrum

technical field

The invention relates to the field of speech signal processing, in particular to a vector quantization method and device for Immittance Spectral Frequencies (ISF) parameters.

Background technique

In the field of speech coding and decoding, linear prediction (Linear Predictive, LP) parameters are used to characterize the channel synthesis filter. If you want to obtain high-quality synthetic speech, you must use efficient quantization technology to provide the LP coefficients of the channel synthesis filter for quantization, so that the average spectral distortion caused by the quantization of LP coefficients is less than 1dB, so as to achieve high-quality speech codec and Transparent quantization of LP parameters in speech synthesis systems. Due to the large dynamic range of LP parameters, the LP parameters are usually converted into other parameters that are completely equivalent in mathematics before quantization. ISF parameters are an effective way to express LP parameters. Compared with another equivalent parameter of the LP parameter, the Linear Spectral Frequency (LSF) parameter, the ISF parameter reduces the computational complexity while ensuring the accuracy.

With the development of high-speed data services in wireless communication systems, wideband speech (50-7000Hz) coding has been more and more widely adopted, and research on wideband LP parameter quantization has become more and more in-depth. For broadband speech, 16th-order LP parameters are usually required to better represent the spectral envelope of speech. Therefore, converting LP parameters into ISF parameters for vector quantization results in greater computational complexity and storage complexity, and consumes more The number of bits can achieve the effect of transparent quantization without losing frames. To solve this problem, a one-step interpolation predictive vector quantization method and a moving average (MA) predictive vector quantization method with equal coefficients in each dimension used by the International Telecommunication Union Telecommunication Standardization Group (ITU-T) G.722.2 have emerged. Among them, the one-step interpolation predictive vector quantization method can realize transparent quantization without missing frames, but in the case of missing frames, it often affects the correctness of ISF parameters in the decoding process, and directly affects the quality of synthesized speech. The MA predictive vector quantization method used in ITU-TG.722.2 can recover the ISF parameters better in the case of lost frames, and the error propagation is small, but it needs more bits in the case of no lost frames. It achieves the same transparent quantization effect as the one-step interpolation predictive vector quantization in the case of no frame loss, and the computational complexity is increased. It can be seen that neither the existing one-step interpolation predictive vector quantization nor the MA predictive vector quantization used in ITU-TG. Voice quality.

Contents of the invention

In order to solve the above problems, an embodiment of the present invention provides a vector quantization method for ISF parameters.

The embodiment of the present invention also provides a vector quantization coding end method of ISF parameters.

The embodiment of the present invention also provides a decoding end method for vector quantization of ISF parameters.

The embodiment of the present invention also provides a vector quantization device for ISF parameters.

The embodiment of the present invention also provides an ISF parameter vector quantization coding end device.

The embodiment of the present invention also provides a vector quantization decoding end device for ISF parameters.

A method for vector quantization of frequency parameters of the immittance spectrum, the method comprising:

Encoding process:

Subtract the ISF parameter mean value vector from the ISF parameter vector of the current frame to obtain the unbiased ISF parameter vector of the current frame;

Multiply the prediction coefficients of the non-equal coefficient inter-frame predictors of each dimension with the unbiased ISF parameter vector after quantization of the previous frame to obtain the prediction vector of the ISF parameter of the current frame;

Subtract the unbiased ISF parameter vector of the current frame from the predicted vector of the ISF parameter of the current frame to obtain the predicted residual vector of the ISF parameter of the current frame;

Carry out vector quantization to the current frame ISF parameter prediction residual vector, obtain the current frame quantized ISF parameter prediction residual vector and its codebook index value, and write the codebook index value into the code stream;

Adding the ISF parameter prediction residual vector after the current frame quantization to the prediction vector of the current frame ISF parameter to obtain the unbiased ISF parameter vector after the current frame quantization is used as the calculation of the prediction vector of the next frame ISF parameter;

Adding the unbiased ISF parameter vector after the quantization of the current frame and the ISF parameter mean vector to obtain the ISF parameter vector after the quantization of the current frame;

Decoding process:

According to the codebook index value parsed from the code stream, query the corresponding current frame quantized ISF parameter prediction residual vector in the residual codebook;

Multiply the prediction coefficients of the non-equal coefficient inter-frame predictors of each dimension with the unbiased ISF parameter vector after quantization of the previous frame to obtain the prediction vector of the ISF parameter of the current frame;

Adding the quantized current frame ISF prediction residual vector and the prediction vector of the current frame ISF parameter to obtain the unbiased ISF parameter vector after the current frame quantization is used as the calculation of the prediction vector of the next frame ISF parameter;

The quantized unbiased ISF parameter vector of the current frame is added to the ISF parameter mean value vector to obtain the quantized ISF parameter vector of the current frame.

A vector quantization coding end method of an immittance spectrum frequency parameter, the method comprising:

Using the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension, calculate the prediction vector of the ISF parameter of the current frame;

Subtract the unbiased ISF parameter vector of the current frame from the predicted vector of the ISF parameter of the current frame, and calculate the predicted residual vector of the ISF parameter of the current frame;

Perform split vector quantization on the ISF parameter prediction residual vector of the current frame.

A vector quantization decoding end method of an immittance spectrum frequency parameter, the method comprising:

Using the residual codebook to obtain the ISF parameter prediction residual vector after quantization of the current frame;

Using the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension, calculate the prediction vector of the ISF parameter of the current frame;

The quantized ISF prediction residual vector of the current frame is added to the prediction vector of the ISF parameter of the current frame to calculate the unbiased ISF parameter of the current frame after quantization.

A vector quantization device for frequency parameters of immittance spectrum, the device includes: an encoding end unit and a decoding end unit;

The encoding end unit stores the prediction coefficients, residual codebook and ISF parameter mean value vector of each dimension non-equal coefficient inter-frame predictor; calculates the unbiased ISF parameter vector of the current frame; converts each dimension non-equal coefficient inter-frame predictor The prediction coefficient of the previous frame is multiplied by the unbiased ISF parameter vector after quantization of the previous frame to calculate the prediction vector of the ISF parameter of the current frame; the unbiased ISF parameter vector of the current frame is subtracted from the ISF parameter prediction vector of the current frame to obtain the ISF parameter of the current frame Prediction residual vector: Carry out vector quantization on the ISF prediction residual vector of the current frame, obtain the ISF parameter prediction residual vector and its codebook index value after quantization of the current frame, write the codebook index value into the code stream and send it to the decoding end Unit; the ISF parameter prediction residual vector after current frame quantization is added with the prediction vector of current frame ISF parameter, obtains the unbiased ISF parameter vector after current frame quantization, is used as the calculation of the prediction vector of next frame ISF parameter; Adding the unbiased ISF parameter vector after the quantization of the current frame and the ISF parameter mean vector to obtain the ISF parameter vector after the quantization of the current frame;

The decoding end unit stores the prediction coefficients, residual codebook and ISF parameter mean value vector of each dimension non-equal coefficient inter-frame predictor, and parses the codebook index value of the current frame from the code stream sent by the encoding end unit; using The current frame codebook index value queries the corresponding current frame quantized ISF prediction residual vector in the residual codebook; use the prediction coefficient of each dimension non-equal coefficient inter-frame predictor and the unbiased ISF parameter after quantization of the previous frame Multiply vectors to calculate the predicted vector of ISF parameters of the current frame; use the quantized ISF prediction residual vector of the current frame to add the predicted vector of ISF parameters of the current frame to calculate the unbiased ISF parameter vector of the current frame after quantization, which is used as the next Calculation of the predicted vector of the ISF parameter of the frame; the ISF parameter vector of the current frame is calculated by adding the quantized unbiased ISF parameter vector of the current frame and the mean value vector of the ISF parameter.

A vector quantization encoding end device for immittance spectrum frequency parameters, the device includes: an adder, an inter-frame predictor with non-equal coefficients in each dimension, and a vector quantization module;

The adder is used to subtract the unbiased ISF parameter vector of the current frame from the predicted vector of the ISF parameter of the current frame provided by the non-equal coefficient inter-frame predictor of each dimension, and calculate the ISF prediction residual vector of the current frame and provide it to the vector quantization module ;

The inter-frame predictors with unequal coefficients in each dimension are used to store the prediction coefficients of the inter-frame predictors with unequal coefficients in each dimension; and calculate the prediction vector of the ISF parameter of the current frame using the prediction coefficients of the inter-frame predictors with unequal coefficients in each dimension and provided to the adder;

The vector quantization module is configured to perform split vector quantization on the ISF parameter prediction residual vector of the current frame provided by the adder.

A device for vector quantization and decoding of immittance spectrum frequency parameters, the device includes: inter-frame predictors with non-equal coefficients in each dimension, a vector quantization decoding module and an unbiased ISF parameter vector reconstruction module;

The vector quantization decoding module provides the unbiased ISF parameter vector reconstruction module by analyzing the quantized ISF parameter prediction residual vector of the current frame according to the code stream;

The inter-frame predictors with non-equal coefficients in each dimension store the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension; use the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension to calculate the prediction vector of the current frame ISF parameter and provide it to Unbiased ISF parameter vector reconstruction module;

The unbiased ISF parameter vector reconstruction module adds the quantized current frame ISF parameter prediction residual vector provided by the vector quantization decoding module to the current frame ISF parameter prediction vector provided by each dimension non-equal inter-frame coefficient predictor , to calculate the unbiased ISF parameter vector after quantization of the current frame.

The above vector quantization method and device for ISF parameters use the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension to calculate the prediction vector of ISF parameters in each frame. By using the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension, better Eliminate the inter-frame correlation of ISF parameters, reduce the dynamic range of ISF parameters, balance the quantization performance in the case of missing frames and no missing frames, and ensure the quality of synthesized speech in both cases of missing frames and no missing frames quality.

Description of drawings

Fig. 1 is the flow chart of the encoding process of the ISF parameter vector quantization method provided by the embodiment of the present invention;

Fig. 2 is the flow chart of the decoding process of the ISF parameter vector quantization method provided by the embodiment of the present invention;

FIG. 3 is a schematic structural diagram of an ISF parameter vector quantization device provided by an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of an ISF parameter vector quantization encoding end device provided by an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of an ISF parameter vector quantization decoding end device provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

第n帧z(n)的预测矢量为p(n)，第n帧ISF参数预测残差矢量为r(n)，第n帧量化后的ISF参数预测残差矢量为第n帧量化后的五个ISF残差码书的码矢量索引值为k₁(n)、k₂(n)、k₃(n)、k₄(n)和k₅(n)。下面结合上述设置参量进行详细描述。Firstly, various parameters that need to be used in the description of the embodiment of the present invention are set. The vector quantization of ISF parameters in the embodiment of the present invention is performed in units of frames. Let the unbiased ISF parameter vector of the nth frame be z(n), and the unbiased ISF parameter vector of the nth frame after quantization is

The prediction vector of the nth frame z(n) is p(n), the ISF parameter prediction residual vector of the nth frame is r(n), and the quantized ISF parameter prediction residual vector of the nth frame is The code vector index values of the five ISF residual codebooks quantized in the nth frame are k ₁ (n), k ₂ (n), k ₃ (n), k ₄ (n) and k ₅ (n). The following describes in detail in conjunction with the above setting parameters.

Fig. 1 shows the encoding end process in the ISF parameter vector quantization method provided by the embodiment of the present invention, in the vector quantization of ISF parameters in each frame, the process includes:

Step 100: Calculate the current frame unbiased ISF parameter vector.

In this step, the unbiased ISF parameter vector is composed of several components, and the dimension of the ISF parameter vector is equal to the order of the ISF parameter. The specific calculation method is as follows:

Let the i-th component of the ISF parameter vector of the nth frame be ω _i (n), i=0, 1, ..., m-1, where m represents the order of the ISF parameter. In order to reduce the dynamic range of calculation and quantization, set the mean value ω _i of the i-th ISF parameter vector, i=0, 1, ..., m-1, the mean value can be obtained through pre-training, the specific training method is in this field Known to the skilled person. Subtract ω _i from ω _i (n) to obtain the i-th component of the unbiased ISF parameter vector of the nth frame as:

z _i (n)=ω _i (n)-ω _i , i=0, 1,..., m-1

Each component of the unbiased ISF parameter vector of the current frame is calculated to obtain the unbiased ISF parameter vector of the current frame.

Step 101: Calculate the predictive vector of ISF parameters of the current frame.

In this step, the current frame is assumed to be the nth frame, and the i-th component in the ISF parameter prediction vector of the nth frame is set to p _i (n), the calculation formula of this component value is:

${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ;</span>$

为第n-1帧量化后的无偏ISF参数矢量中的第i分量，如果n＝1即当前帧为第一帧，则使用预置的初始值0来计算。上述公式中的α_i为各维非等系数帧间预测器的预测系数，可以通过平方预测误差之和最小的方法，从训练语音序列中估计出来，并在确定之后保持恒定，这里所说的训练语音序列为从LP参数转化为ISF参数形成的序列，具体转化的方法为本领域技术人员公知的内容。α_i具体的确定方法为：in,

It is the i-th component in the unbiased ISF parameter vector after the quantization of the n-1th frame. If n=1, that is, the current frame is the first frame, use the preset The initial value of 0 to calculate. α _i in the above formula is the prediction coefficient of the inter-frame predictor with non-equal coefficients in each dimension, which can be estimated from the training speech sequence by the method of minimizing the sum of the squared prediction errors, and remains constant after being determined. Here, The training speech sequence is a sequence formed by converting LP parameters into ISF parameters, and the specific conversion method is well known to those skilled in the art. The specific determination method of α _i is:

Let the sum of the squared prediction errors of the i-th component in the ISF parameter prediction residual vector be:

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; .</span>$

Wherein, N _f is the total number of frames of the training sequence, r _i (n) is the i-th component value of the ISF parameter prediction residual vector of the current frame, this parameter and its related formula are proposed by the embodiment of the present invention, and will be followed in this process The steps will be described in detail, here is the training formula for α _i obtained by using the r _i (n) correlation formula r _i (n)=z _i (n)−p _i (n).

${\mathrm{<span\; class="notranslate">\; E.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{}<span\; class="notranslate">\; ]</span>$

According to the method of minimum error, set δE _i /δα _i = 0 to get:

${\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ;</span>$

其中z_i(n-1)为第n-1帧无偏ISF参数矢量的第i个分量。For the sake of convenience during calculation, z _i (n-1) is used instead of

Where z _i (n-1) is the i-th component of the unbiased ISF parameter vector of frame n-1.

According to empirical values, in order to balance the quantization performance of ISF parameters when there are no missing frames and when there are missing frames, in this step, some processing is performed on the original values of the prediction coefficients of the trained inter-frame predictors with non-equal coefficients in each dimension. The specific method is to multiply the original value by 0.5 as the prediction coefficient of the inter-frame predictor with non-equal coefficients in each dimension used in the process.

Each component in the ISF parameter prediction vector of the current frame is calculated to obtain the ISF parameter prediction vector of the current frame.

Step 102: Subtract the current frame unbiased ISF parameter vector from the current frame ISF parameter prediction vector to obtain the current frame ISF parameter prediction residual vector.

In this step, according to the current frame unbiased ISF parameter vector obtained in step 100 and the current frame ISF parameter prediction vector obtained in step 101, the current frame ISF parameter prediction residual vector is calculated. Assuming that the current frame is the nth frame, the i-th component of the ISF parameter prediction residual vector of the current frame is r _i (n), and the calculation formula is r _i (n)=z _i (n)-p _i (n).

Each component of the current frame ISF prediction stagger vector is calculated to obtain the prediction residual vector of the current frame ISF parameters.

Step 103: Perform split vector quantization on the ISF parameter prediction residual vector of the current frame.

In this step, the current frame ISF parameter prediction residual vector r(n) obtained in step 102 is split into five sub-vectors, set as r ⁽¹⁾ (n), r ⁽²⁾ (n), r ⁽³⁾ (n), r ⁽⁴⁾ (n), and r ⁽⁵⁾ (n). Where r ⁽¹⁾ (n), r ⁽²⁾ (n), r ⁽³⁾ (n), r ⁽⁴⁾ (n) are all 3-dimensional vectors, r ⁽⁵⁾ (n) is a 4-dimensional vector, Perform vector quantization on the five sub-vectors using the number of bits in Table 1 respectively.

In the embodiment of the present invention, splitting the quantized ISF prediction residual vector into five sub-vectors is a preferred implementation manner. Since in wideband speech coding, the spectral envelope of the speech can be better represented only when the LP coefficient is set to 16th order, the order of the ISF parameter is set to m=16. In addition to the split vector quantization method, other vector quantization methods can also be used for quantization.

Table I

The square of the quantization error is used as the distortion measure in the quantization process, so the quantization process is to find the code vector index number that minimizes the value of the following formula in the residual codebook.

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; m</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; .</span>$

Wherein, m and n represent the element numbers of the first and last elements in the ISF parameter prediction residual sub-vector in the entire ISF prediction residual vector. The quantization of the five sub-vectors corresponds to different residual codebooks, and the index numbers of the five code vectors searched according to the above formula are: k ₁ (n), k ₂ (n), k ₃ (n), k ₄ (n) and k ₅ (n).

The quantization of the five sub-vectors in this step uses 5 different ISF residual codebooks respectively, wherein the residual codebook 1 contains 1024 3-dimensional code vectors, and the residual codebook 2 contains 1024 3-dimensional code vectors. Book 3 contains 512 3-dimensional code vectors, residual code book 4 contains 512 3-dimensional code vectors, and residual code book 5 contains 256 4-dimensional code vectors. The training of the above residual codebooks can be obtained by using the traditional LBG algorithm. The training data is a multilingual database including Chinese, British English, American English, Finnish, Japanese and French. The length is about 2 hours. The rate is 16kHz and the precision is 16-bit linear pulse code modulation (PCM). The specific training steps of the residual codebook are as follows:

First, from the ISF training vector ω(n), use the following formula to calculate the i-th component of the mean value vector of the ISF parameter: $\stackrel{\&OverBar;}{{\mathrm{\&omega;}}_i}=\frac{1}{N_f}\underset{\mathrm{no}=1}{\overset{N_f}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_i\left(\mathrm{no}\right),i=\mathrm{0,1},...,m-1$

Second, calculate the unbiased ISF parameter vector of all training vectors;

Thirdly, calculate the prediction vector p(n) of the unbiased ISF parameter vector by applying the formula in step 101 .

Fourth, apply the formula in step 102 to calculate the ISF parameter prediction residual vector sequence {r _i (n)}, i=0, 1,..., m-1, n=1, 2,..., N _f , each An ISF parameter residual vector is split into five sub-vectors.

N₁为第一个子矢量的码书大小，上角标代表迭代次数。Fifth, the training sequence for the first sub-vector split $\left\{{\hat{r}}_{i_1}\left(\mathrm{no}\right)\right\},\mathrm{no}=\mathrm{1,2},......,N_f,$ Generating initial codebook by splitting method

N ₁ is the codebook size of the first sub-vector, and the superscript represents the number of iterations.

Sixth, the LBG iterative algorithm is used to re-divide the cell and obtain the new centroid R _i ⁽ⁿ⁾ , i=0, 1, ... N ₁ -1. The specific iterative algorithm is well known to those skilled in the art.

Seventh, calculate the average distortion D and relative distortion if relatively distorted $\widetilde{\mathrm{D.}}<\mathrm{\&epsiv;}$ Then continue, where ε is a set minimum error value; otherwise, go to the sixth step.

即为第一个子矢量的最终码书。Eighth, the code book at this time

That is, the final codebook of the first sub-vector.

和 $\left\{{\hat{r}}_{i_5}\left(n\right)\right\},n=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N_f$ 同样进行第五至第八步骤，生成对应后四个子矢量的最终码书

和

其中N₂、N₃、N₄和N₅分别为第二、三、四、五个子矢量的码书大小。against

and $\left\{{\hat{r}}_{i_5}\left(\mathrm{no}\right)\right\},\mathrm{no}=\mathrm{1,2},\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;,N_f$ Also perform the fifth to eighth steps to generate the final codebook corresponding to the last four sub-vectors

and

Where N ₂ , N ₃ , N ₄ and N ₅ are the codebook sizes of the second, third, fourth and fifth sub-vectors respectively.

The residual codebook trained according to the above first to eighth steps can be directly used in the process of the embodiment of the present invention.

Step 104: Perform binary encoding on the codebook index value obtained in step 103 and write it into the code stream.

Step 105: Reconstruct the quantized ISF parameter prediction residual vector of the current frame according to the codebook index value.

In this step, according to the searched code vector index values k ₁ (n), k ₂ (n), k ₃ (n), k ₄ (n) and k ₅ (n), the current frame after quantization is reconstructed The ISF prediction residual vector is:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1,2</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 3,4,5</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 6,7,8</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 9,10,11</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 12,13,14,15</span>}\end{array}\right.<span\; class="notranslate">\; .</span>$

Step 106: Calculate the quantized unbiased ISF parameter subvector of the current frame.

In this step, the current frame ISF prediction residual vector reconstructed in step 105 and the formula r(n)=z(n)-p(n) defined in step 102 are used to calculate the unbiased The ISF parameter sub-vectors are:

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1,2</span>}<span\; class="notranslate">\; ;</span>$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 3,4</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; ;</span>$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 6,7</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; ;</span>$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 9,10</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; ;</span>$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 12</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 14,15</span>}<span\; class="notranslate">\; ;</span>$

All these quantized unbiased ISF parameter sub-vectors constitute a quantized unbiased ISF parameter vector. In the quantization of the next frame, the ISF parameter prediction vector in step 101 will be calculated using the unbiased ISF parameter vector calculated in the current frame.

Step 107: Calculate the quantized ISF parameter vector of the current frame.

In this step, the quantized ISF parameter vector of the current frame is calculated by using the current frame ISF parameter calculated in step 106 and the stored mean value of the ISF parameter vector.

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1,2</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 3,4,5</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 6,7,8</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 9,10,11</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 12,13,14,15</span>}\end{array}\right.<span\; class="notranslate">\; .</span>$

After the above steps 100 to 107, the encoding process of the ISF parameter vector quantization method provided by the embodiment of the present invention ends.

Fig. 2 shows the flow of the decoding end in the ISF parameter vector quantization method provided by the embodiment of the present invention. In the vector quantization of ISF parameters in each frame, the flow includes:

Step 200: Parse the codebook index value from the code stream, and query the quantized ISF prediction residual vector of the current frame corresponding to the codebook index value from the residual codebook.

In this step, the quantized ISF prediction residual sub-vectors of the current frame are queried in the residual codebook according to the codebook index value, which are denoted as ${\hat{r}}_i^{k_1\left(\mathrm{no}\right)},i=\mathrm{0,1,2},$ ${\hat{r}}_i^{k_2\left(\mathrm{no}\right)},i=\mathrm{3,4,5},$ ${\hat{r}}_i^{k_3\left(\mathrm{no}\right)},i=\mathrm{6,7,8},$ ${\hat{r}}_i^{k_4\left(\mathrm{no}\right)},i=\mathrm{9,10,11}$ and ${\hat{r}}_i^{k_5\left(\mathrm{no}\right)},i=\mathrm{12,13,14,15}.$ Merge the five quantized ISF prediction residual sub-vectors to obtain the quantized ISF parameter prediction residual vector of the current frame as:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1,2</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 3,4,5</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 6,7,8</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 9,10,11</span>}\\ {\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 12,13,14,15</span>}\end{array}\right.<span\; class="notranslate">\; ;</span>$

Step 201: Calculate the ISF parameter prediction vector of the current frame by using the prediction coefficients of the non-equal-coefficient inter-frame predictors of each dimension.

Let the current frame be the nth frame, and the ith component of the predictive vector of the ISF parameter of the nth frame is p _i (n), the method of calculating the predictive vector of the ISF parameter of the current frame is:

${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ;</span>$

为第n-1帧量化后的无偏ISF参数矢量的第i个分量，如果当前帧为第一帧，则使用预置的

初始值0带入公式计算。将上述计算的所有分量组合起来构成当前帧ISF参数的预测矢量。Among them, m is the order of ISF parameters,

It is the i-th component of the unbiased ISF parameter vector quantized for the n-1th frame. If the current frame is the first frame, use the preset

The initial value 0 is brought into the formula calculation. All the components calculated above are combined to form the predictive vector of ISF parameters of the current frame.

Step 202: Calculate the quantized unbiased ISF parameter vector of the current frame.

In this step, using the quantized ISF parameter prediction residual vector of the current frame obtained in step 200 and the ISF parameter prediction vector obtained in step 201, the unbiased ISF parameter sub-vectors after quantization of the current frame are calculated as follows:

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1,2</span>}$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 3,4,5</span>}$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 8</span>}$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 11</span>}$

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; r</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 12</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 14,15</span>}$

The five quantized unbiased ISF parameter sub-vectors are combined to obtain the quantized unbiased ISF parameter vector of the current frame. The quantized ISF parameter vector obtained in each frame of quantization will be used in the calculation of the ISF parameter prediction vector in the next frame.

Step 203: Calculate the quantized ISF parameter vector of the current frame.

In this step, the quantized unbiased ISF parameter vector of the current frame calculated in step 202 is added to the ISF parameter mean vector to obtain the quantized ISF parameter vector of the current frame, and the specific calculation formula is as follows:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0,1,2</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 3,4,5</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 6,7,8</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 9,10,11</span>}\\ \stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; z</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; 5</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 12,13,14,15</span>}\end{array}\right.<span\; class="notranslate">\; .</span>$

After steps 200 to 203, the decoding process of the ISF parameter vector quantization method provided by the embodiment of the present invention ends.

The above steps 100 to 107, and steps 200 to 203 together constitute the complete process of the ISF parameter vector quantization method provided by the embodiment of the present invention, wherein steps 100 to 107 can also be used as the ISF parameter vector quantization provided by the embodiment of the present invention A preferred implementation mode of the coding end method, steps 200 to 203 can also be used as a preferred implementation mode of the ISF parameter vector quantization decoding end method provided by the embodiment of the present invention.

Finally, the device for vector quantization of ISF parameters provided by the embodiment of the present invention is introduced in detail. Figure 3 shows a schematic structural diagram of an ISF parameter vector quantization device provided by an embodiment of the present invention. The device includes an encoding end unit and a decoding end unit, wherein the encoding end unit includes: a residual codebook unit, an adder, a split vector quantization encoding module, non-equal coefficient inter-frame predictor of each dimension, split vector quantization decoding module, unbiased ISF parameter vector reconstruction module and ISF parameter vector reconstruction module. The decoding end unit includes: a residual codebook unit, a split vector quantization decoding module, an unbiased ISF parameter vector reconstruction module, an inter-frame predictor with non-equal coefficients in each dimension, and an ISF parameter vector reconstruction module. The calculations involved in each component in the device can use the formulas and calculation methods defined in the ISF parameter vector quantization in the embodiment of the present invention, as a preferred implementation mode of the ISF parameter vector quantization device in the embodiment of the present invention.

Firstly, each component of the encoder unit is introduced. The encoder unit performs vector quantization of ISF parameters for each frame in turn. The following descriptions are made in units of frames, assuming that the current quantization is the nth frame.

Each dimensional unequal coefficient inter-frame predictor is used to store the prediction coefficients of each dimensional unequal coefficient inter-frame predictor, and the prediction coefficients of each dimensional unequal coefficient inter-frame predictor are provided by the unbiased ISF parameter vector reconstruction module The unbiased ISF parameter vectors quantized in the previous frame are multiplied to calculate the predictive vector of ISF parameters in the current frame and provide it to the split vector quantization decoding module and adder. The prediction coefficients of the above-mentioned non-equal inter-frame coefficient predictors of each dimension are obtained by pre-training, and the training method is the same as that described in the ISF parameter vector quantization method provided by the embodiment of the present invention. remains constant in the predictor for use by other components in the encoder side unit.

The adder is used to subtract the unbiased ISF parameter vector of the current frame provided by the unbiased ISF parameter vector generation module from the predicted vector of the ISF parameter of the current frame provided by the non-equal coefficient inter-frame predictor of each dimension, and calculate the ISF prediction residual of the current frame The difference vector is provided to the split vector quantization encoding module.

The residual codebook unit stores five residual codebooks, that is, residual codebooks 1-5, and provides queries for the split vector quantization encoding module and the split vector quantization decoding module. The specific training method of the residual codebook stored in the residual codebook unit is the same as that described in the ISF parameter vector quantization method provided in the embodiment of the present invention.

The split vector quantization encoding module is used to predict the residual vector according to the current frame ISF parameters provided by the adder, query the current frame codebook index value from the residual codebook unit and write it into the code stream, and provide the current frame codebook index value to Split vector quantization decoding module.

The split vector quantization decoding module is used to receive the current frame codebook index value provided by the split vector quantization unit, search for the quantized ISF parameter prediction residual vector corresponding to the current frame codebook index value in the residual codebook unit, and provide it to Unbiased ISF parameter vector reconstruction module.

The unbiased ISF parameter vector reconstruction module is used to compare the current frame quantized ISF parameter prediction residual vector provided by the split vector quantization decoding module with the current frame ISF parameter prediction vector provided by the non-equal coefficient inter-frame predictor of each dimension Add, calculate the quantized unbiased ISF parameter vector of the current frame and cache it, and provide the unbiased ISF parameter vector of the current frame quantized to the non-equal coefficient inter-frame predictor of each dimension in the next frame of the current frame to be used as the ISF of the next frame For the calculation of the predictive vector of the parameter, the quantized unbiased ISF parameter vector of the current frame is provided to the ISF parameter vector reconstruction module in the current frame.

The unbiased ISF parameter generation module is used to store the ISF parameter mean vector, subtract the current frame ISF parameter vector from the stored ISF parameter mean vector, calculate the current frame unbiased ISF parameter vector and provide it to the adder.

The above encoding end unit can also be used as a preferred implementation of the ISF parameter vector quantization encoding end device provided by the embodiment of the present invention. FIG. 4 shows the structure of the ISF parameter vector quantization encoding end device provided by the embodiment of the present invention.

Next, introduce each component in the decoder unit. The decoder unit performs ISF parameter vector quantization on each frame in turn, and the following descriptions are made in units of frames.

Each dimensional unequal coefficient inter-frame predictor is used to store the prediction coefficients of each dimensional unequal coefficient inter-frame predictor; the prediction coefficients of each dimensional unequal coefficient inter-frame predictor and the unbiased ISF parameter vector reconstruction module provide The quantized unbiased ISF parameter vectors of the previous frame are multiplied to calculate the predicted vector of ISF parameters of the current frame and provided to the unbiased ISF parameter vector reconstruction module. The prediction coefficients of the above-mentioned non-equal coefficient inter-frame predictors of each dimension are obtained by pre-training, and the training method is the same as that described in the ISF parameter vector quantization method of the embodiment of the present invention, and are stored in each dimension non-equal coefficient inter-frame predictor after quantization starts keep constant.

The unbiased ISF parameter vector reconstruction module adds the quantized current frame ISF parameter prediction residual vector provided by the split vector quantization decoding module to the current frame ISF parameter prediction vector provided by each dimension non-equal coefficient inter-frame predictor, Calculate the unbiased ISF parameter vector after quantization of the current frame and cache it, provide the vector to the non-equal coefficient inter-frame predictor in each dimension in the next frame of the current frame, and provide the vector to the ISF parameter vector reconstruction module in the current frame .

The residual codebook unit stores five residual codebooks, that is, residual codebooks 1-5, and provides queries for the split vector quantization decoding module.

The split vector quantization decoding module parses the codebook index value from the code stream sent from the encoder unit, and queries the quantized ISF parameters of the current frame in the residual codebook unit to predict the residual vector and provide it to the unbiased ISF parameter Vector reconstruction module.

The ISF parameter vector reconstruction module stores the ISF parameter mean value vector, adds the unbiased ISF parameter vector after current frame quantization provided by the unbiased ISF parameter vector reconstruction module to the ISF parameter mean value vector, and calculates the current frame quantized ISF parameter vector.

The above decoding end unit can also be used as a preferred implementation mode of the ISF parameter vector quantization decoding end device provided by the embodiment of the present invention. FIG. 5 shows the structure of an ISF parameter vector quantization decoder device provided by an embodiment of the present invention.

In the ISF parameter vector quantization device provided by the embodiment of the present invention, the encoding end unit can work according to the encoding process of the ISF parameter vector quantization method provided by the embodiment of the present invention, and the decoding end unit can work according to the ISF parameter vector quantization provided by the embodiment of the present invention The decoding process of the method works, and the device can be applied in a wideband speech codec system.

Applying the ISF parameter vector quantization device provided by the embodiment of the present invention to an actual wideband speech codec system, the following implementation modes can be cited as examples:

At the encoding end of the Algebraic Code Excited Linear Prediction (ACELP) speech encoding and decoding system, the first step is to perform linear prediction analysis on the preprocessed input speech signal, and the analysis frame length is 30ms, of which 20ms (256 samples) from the current frame, 5ms (64 samples) from the previous frame, and 5ms (64 samples) from the next frame. The window function adopts an asymmetric window focused on the fourth subframe of the current frame. The window consists of two parts. The first part is a half Hamming window (256 points), and the second part is a quarter cosine function (128 points). point). In the second step, autocorrelation estimation is performed on the speech after windowing, and the autocorrelation function is multiplied by the hysteresis window to make it have a bandwidth extension of 60 Hz. The third step is to use the Levinson-Durbin recursive algorithm to obtain the 16th-order LP coefficients. The fourth step is to convert the LP coefficients into ISF parameters. In the fifth step, the ISF parameters are quantized in the encoder unit of the ISF parameter vector quantization device provided by the embodiment of the present invention, and the obtained five codebook index values are written into the code stream.

At the decoding end in the wideband ACELP speech codec system, in the first step, after the decoding end unit of the ISF parameter vector quantization device according to the embodiment of the present invention receives the code stream from the encoding end unit, according to the parsed five codebook index values , to restore the quantized ISF parameter vector. In the second step, the restored quantized ISF parameter vectors are converted into Immittance Spectral Pairs (ISP) vectors, the ISP vectors are interpolated to obtain the ISP vectors of four subframes, and converted back to ISF parameters. In the third step, the ISF parameters of the four subframes are converted into LP coefficients, and the decoding is completed.

In the implementation of the above-mentioned application in a specific codec system, the ISF parameter vector quantization device provided by the embodiment of the present invention completes the vector quantization of the ISF parameter converted from the LP parameter, and the vector quantization of the ISF parameter in the embodiment of the present invention is used for quantization The encoding process and decoding process in the method. The effect obtained by the vector quantization method and device for ISF parameters provided by the embodiment of the present invention is illustrated by the following experimental results:

17250 ISF parameter vectors were generated with 345 seconds (17250 frames) of Chinese speech outside the training speech. Experiments show that, according to the internationally accepted spectral distortion calculation method, in the case of no lost frames, the average spectral distortion obtained by quantizing each frame with 46 bits is 0.787173dB, and the percentage of frames with spectral distortion between 2dB and 4dB is only 0.968 %, and the percentage of average spectral distortion greater than 4dB is zero, and its quantization performance has reached the standard of transparent quantization. In the case of a lost frame, only the information of the previous received frame can be used to restore the ISF parameters of this frame, and the continuity of the error can be kept within a few frames. Disgusting sound.

The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. Protection scope, within the spirit and principles of the present invention, any modification, equivalent replacement, improvement, etc., shall be included in the protection scope of the present invention.

Claims (21)

1. A vector quantization method of the immittance spectrum frequency parameter, is characterized in that, the method comprises: Encoding process: Subtract the ISF parameter mean value vector from the ISF parameter vector of the current frame to obtain the unbiased ISF parameter vector of the current frame; Multiply the prediction coefficients of the non-equal coefficient inter-frame predictors of each dimension with the unbiased ISF parameter vector after quantization of the previous frame to obtain the prediction vector of the ISF parameter of the current frame; Subtract the unbiased ISF parameter vector of the current frame from the predicted vector of the ISF parameter of the current frame to obtain the predicted residual vector of the ISF parameter of the current frame; Perform split vector quantization on the current frame ISF parameter prediction residual vector to obtain the current frame quantized ISF parameter prediction residual vector and its codebook index value, and write the codebook index value into the code stream; Adding the ISF parameter prediction residual vector after the current frame quantization to the prediction vector of the current frame ISF parameter to obtain the unbiased ISF parameter vector after the current frame quantization is used as the calculation of the prediction vector of the next frame ISF parameter; Adding the unbiased ISF parameter vector after the quantization of the current frame and the ISF parameter mean vector to obtain the ISF parameter vector after the quantization of the current frame; Decoding process: According to the codebook index value parsed from the code stream, query the corresponding current frame quantized ISF parameter prediction residual vector in the residual codebook; Multiply the prediction coefficients of the non-equal coefficient inter-frame predictors of each dimension with the unbiased ISF parameter vector after quantization of the previous frame to obtain the prediction vector of the ISF parameter of the current frame; Adding the quantized current frame ISF prediction residual vector and the prediction vector of the current frame ISF parameter to obtain the unbiased ISF parameter vector after the current frame quantization is used as the calculation of the prediction vector of the next frame ISF parameter; The quantized unbiased ISF parameter vector of the current frame is added to the ISF parameter mean value vector to obtain the quantized ISF parameter vector of the current frame. 2. A vector quantization coding end method of the immittance spectrum frequency parameter, is characterized in that, the method comprises: Using the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension, calculate the prediction vector of the ISF parameter of the current frame; Subtract the unbiased ISF parameter vector of the current frame from the predicted vector of the ISF parameter of the current frame, and calculate the predicted residual vector of the ISF parameter of the current frame; Perform split vector quantization on the ISF parameter prediction residual vector of the current frame. 3. The method according to claim 2, further comprising: pre-training the prediction coefficients, residual codebook and ISF parameter mean value vector of each dimension non-equal coefficient inter-frame predictor; After the split vector quantization, further include: Adding the quantized current frame ISF parameter prediction residual vector to the current frame ISF parameter prediction vector, reconstructing the current frame quantized unbiased ISF parameter vector, used as the calculation of the next frame ISF parameter prediction vector; The quantized unbiased ISF parameter vector of the current frame is added to the ISF parameter mean value vector to obtain the quantized ISF parameter vector of the current frame. 4. The method according to claim 3, wherein the method for calculating the vector predictor of the current frame ISF parameter is: Multiply the prediction coefficients of the non-equal coefficient inter-frame predictors in each dimension with the quantized unbiased ISF parameters of the previous frame to calculate the prediction vector of the ISF parameters of the current frame. 5. The method according to claim 3, wherein the current frame is the first frame, and the method for calculating the vector predictor of the current frame ISF parameter is: Multiply the prediction coefficients of the non-equal-coefficient inter-frame predictors of each dimension with 0 to calculate the prediction vector of the ISF parameters of the current frame. 6. the method for claim 3, is characterized in that, the method for the prediction coefficient of described training each dimension non-equal coefficient inter-frame predictor is: Make the sum of the squared prediction errors of each component of the ISF parameter prediction residual vector of the current frame the smallest, and multiply the unbiased ISF parameter vector quantized by the previous frame by the prediction coefficient of the non-equal coefficient inter-frame predictor to be solved to represent the current The predicted vector of the ISF parameter of the frame, and the unbiased ISF parameter vector of the current frame minus the predicted vector of the ISF parameter of the current frame is used to represent the predicted residual vector of the ISF parameter of the current frame, and the non-equal coefficient frame of each dimension corresponding to each component is solved The prediction coefficient value of the inter-predictor; Prediction coefficient values of inter-frame predictors with unequal coefficients in each dimension of all components are used as prediction coefficients of inter-frame predictors with unequal coefficients in each dimension trained in advance. 7. The method according to claim 6, wherein said obtaining the values of the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension further comprises: Multiply the prediction coefficient value of the inter-frame predictor with non-equal coefficients of each dimension corresponding to each component by 0.5, and use the multiplied value as the inter-frame prediction of non-equal coefficients of each dimension corresponding to each component used in the quantization process value of the predictor coefficient. 8. The method according to claim 2, wherein the method for carrying out split vector quantization to the current frame ISF parameter prediction residual vector is: The current frame ISF parameter prediction residual vector is split into five sub-vectors, wherein the first 3 components corresponding to the current frame ISF parameter prediction residual vector form the first sub-vector, and the 4th to 6 components form the second sub-vector, The 7th to 9th components form the third subvector, the 10th to 12th components form the fourth subvector, and the 13th to 16th components form the fifth subvector; Each codebook index value corresponding to the five sub-vectors is searched in the residual codebook. 9. A vector quantization decoding end method of the immittance spectrum frequency parameter, is characterized in that, the method comprises: Using the residual codebook to obtain the ISF parameter prediction residual vector after quantization of the current frame; Using the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension, calculate the prediction vector of the ISF parameter of the current frame; The quantized ISF parameter prediction residual vector of the current frame is added to the ISF parameter prediction vector of the current frame to calculate the unbiased ISF parameter vector of the current frame quantization. 10. The method according to claim 9, further comprising: pre-training the prediction coefficients, residual codebook and ISF parameter mean value vector of each dimension non-equal coefficient inter-frame predictor; After the calculation of the unbiased ISF parameter vector after quantization of the current frame, it further includes: The quantized unbiased ISF parameter vector of the current frame is added to the ISF parameter mean value vector to obtain the quantized ISF parameter vector of the current frame. 11. The method according to claim 10, wherein the prediction vector of the ISF parameter of the current frame is calculated using the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension: Multiply the prediction coefficients of the unequal-coefficient inter-frame predictors in each dimension with the quantized unbiased ISF parameter vector of the previous frame to calculate the predictive vector of the ISF parameters of the current frame. 12. The method according to claim 10, wherein the current frame is the first frame, and the predicted vector for calculating the ISF parameter of the current frame is: Multiply the prediction coefficients of the non-equal-coefficient inter-frame predictors of each dimension with 0 to calculate the prediction vector of the ISF parameters of the current frame. 13. The method according to claim 10, wherein the method for the prediction coefficients of the non-equal coefficient inter-frame predictors of each dimension of the training is: Make the sum of the squared prediction errors of each component of the ISF parameter prediction residual vector of the current frame the smallest, and multiply the unbiased ISF parameter vector quantized by the previous frame by the prediction coefficient of the non-equal coefficient inter-frame predictor to be solved to represent the current The predicted vector of the ISF parameter of the frame, and the unbiased ISF parameter vector of the current frame minus the predicted vector of the ISF parameter of the current frame is used to represent the predicted residual vector of the ISF parameter of the current frame, and the non-equal coefficient frame of each dimension corresponding to each component is solved The prediction coefficient value of the inter-predictor; Prediction coefficient values of inter-frame predictors with unequal coefficients in each dimension of all components are used as prediction coefficients of inter-frame predictors with unequal coefficients in each dimension trained in advance. 14. The method according to claim 13, further comprising: Multiply the prediction coefficient value of the inter-frame predictor with non-equal coefficients of each dimension corresponding to each component by 0.5, and use the multiplied value as the inter-frame prediction of non-equal coefficients of each dimension corresponding to each component used in the quantization process value of the predictor coefficient. 15. A vector quantization device of an immittance spectrum frequency parameter, characterized in that the device comprises: an encoding end unit and a decoding end unit; The encoding end unit stores the prediction coefficients, residual codebook and ISF parameter mean value vector of each dimension non-equal coefficient inter-frame predictor; calculates the unbiased ISF parameter vector of the current frame; converts each dimension non-equal coefficient inter-frame predictor The prediction coefficient of the previous frame is multiplied by the unbiased ISF parameter vector after quantization of the previous frame to calculate the prediction vector of the ISF parameter of the current frame; the unbiased ISF parameter vector of the current frame is subtracted from the ISF parameter prediction vector of the current frame to obtain the ISF parameter of the current frame Prediction residual vector: perform vector quantization on the current frame ISF prediction residual vector, obtain the current frame quantized ISF parameter prediction residual vector and its codebook index value, write the codebook index value into the code stream and send it to the decoder unit ; The ISF parameter prediction residual vector after the current frame quantization is added with the prediction vector of the current frame ISF parameter to obtain the unbiased ISF parameter vector after the current frame quantization, which is used as the calculation of the prediction vector of the next frame ISF parameter; The unbiased ISF parameter vector after quantization of the current frame is added to the ISF parameter mean value vector to obtain the ISF parameter vector after quantization of the current frame; The decoding end unit stores the prediction coefficients, residual codebook and ISF parameter mean value vector of each dimension non-equal coefficient inter-frame predictor, and parses the codebook index value of the current frame from the code stream sent by the encoding end unit; using The current frame codebook index value queries the corresponding current frame quantized ISF prediction residual vector in the residual codebook; use the prediction coefficient of each dimension non-equal coefficient inter-frame predictor and the unbiased ISF parameter after quantization of the previous frame Multiply vectors to calculate the predicted vector of ISF parameters of the current frame; use the quantized ISF prediction residual vector of the current frame to add the predicted vector of ISF parameters of the current frame to calculate the unbiased ISF parameter vector of the current frame after quantization, which is used as the next step Calculation of the predicted vector of the ISF parameter of the frame; the ISF parameter vector of the current frame is calculated by adding the quantized unbiased ISF parameter vector of the current frame to the mean value vector of the ISF parameter. 16. A vector quantization encoding end device of the immittance spectrum frequency parameter, characterized in that the device comprises: an adder, an inter-frame predictor with non-equal coefficients in each dimension and a vector quantization module; The adder is used to subtract the unbiased ISF parameter vector of the current frame from the predicted vector of the ISF parameter of the current frame provided by the inter-frame predictor with non-equal coefficients of each dimension, and calculate the ISF parameter prediction residual vector of the current frame and provide it to vector quantization module; The inter-frame predictors with unequal coefficients in each dimension are used to store the prediction coefficients of the inter-frame predictors with unequal coefficients in each dimension; and calculate the prediction vector of the ISF parameter of the current frame using the prediction coefficients of the inter-frame predictors with unequal coefficients in each dimension and provided to the adder; The vector quantization module is configured to perform split vector quantization on the ISF parameter prediction residual vector of the current frame provided by the adder. 17. The device according to claim 16, wherein the vector quantization module comprises: a split vector quantization encoding module, a split vector quantization decoding module and a residual codebook unit; The split vector quantization encoding module is used to predict the residual vector according to the current frame ISF parameters provided by the adder, search the current frame codebook index value from the residual codebook stored in the residual codebook unit, and write the code stream to the The current frame codebook index value is provided to the split vector quantization decoding module; The split vector quantization decoding module is configured to receive the current frame codebook index value provided by the split vector quantization encoding module, and search for the quantized ISF corresponding to the current frame codebook index value in the residual codebook stored in the residual codebook unit Parameter prediction residual vector; The residual codebook unit stores a residual codebook of ISF parameters for query by the split vector quantization encoding module and the split vector quantization decoding module. 18. The device according to claim 17, further comprising: an unbiased ISF parameter vector generation module, an unbiased ISF parameter vector reconstruction module and an ISF parameter vector reconstruction module; The unbiased ISF parameter vector generation module is used to store the ISF parameter mean value vector, subtract the current frame ISF parameter vector from the ISF parameter mean value vector, calculate the current frame unbiased ISF parameter vector and provide it to the adder; The unbiased ISF parameter vector reconstruction module is used for predicting the current frame quantized ISF parameter prediction residual vector provided by the split vector quantization decoding module and the current frame ISF parameter provided by each dimension non-equal coefficient inter-frame predictor Vector addition, calculate and cache the unbiased ISF parameter vector after quantization of the current frame, and provide the unbiased ISF parameter vector after quantization of the current frame to the non-equal coefficient inter-frame predictor of each dimension in the next frame of the current frame for use as the next frame The calculation of the prediction vector of the ISF parameter of the frame provides the ISF parameter vector reconstruction module with the unbiased ISF parameter vector after the quantization of the current frame in the current frame; The ISF parameter vector reconstruction module is used to store the ISF parameter mean value vector, and adds the unbiased ISF parameter after quantization of the current frame provided by the unbiased ISF parameter vector reconstruction module to the ISF parameter mean value vector, and calculates the current frame after quantization ISF parameter vector; The calculation of the prediction vector of the current frame ISF parameter by the non-equal coefficient inter-frame predictor of each dimension is specifically: quantizing the prediction coefficient of each dimension non-equal coefficient inter-frame predictor and the previous frame provided by the unbiased ISF parameter vector reconstruction module The final unbiased ISF parameter vector is multiplied to calculate the prediction vector of the ISF parameter of the current frame. 19. A vector quantization decoding terminal device of immittance spectrum frequency parameters, characterized in that the device comprises: each dimension non-equal coefficient inter-frame predictor, vector quantization decoding module and unbiased ISF parameter vector reconstruction module; The vector quantization decoding module provides the unbiased ISF parameter vector reconstruction module by analyzing the quantized ISF parameter prediction residual vector of the current frame according to the code stream; The inter-frame predictors with non-equal coefficients in each dimension store the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension; use the prediction coefficients of the inter-frame predictors with non-equal coefficients in each dimension to calculate the prediction vector of the current frame ISF parameter and provide it to Unbiased ISF parameter vector reconstruction module; The unbiased ISF parameter vector reconstruction module adds the quantized current frame ISF parameter prediction residual vector provided by the vector quantization decoding module to the current frame ISF parameter prediction vector provided by each dimension non-equal inter-frame coefficient predictor , to calculate the unbiased ISF parameter vector after quantization of the current frame. 20. The device according to claim 19, wherein the device further comprises: an ISF parameter vector reconstruction module; The unbiased ISF parameter vector reconstruction module is further used to: buffer the unbiased ISF parameter vector after the quantization of the current frame, and provide the unbiased ISF parameter vector after the quantization of the current frame to each dimension non-equal parameter vector in the next frame of the current frame. The coefficient inter-frame predictor provides the unbiased ISF parameter vector after quantization of the current frame to the ISF parameter vector reconstruction module in the current frame; The ISF parameter vector reconstruction module is used to store the ISF parameter mean value vector, and the unbiased ISF parameter vector after the current frame quantization provided by the unbiased ISF parameter vector reconstruction module is added to the ISF parameter mean value vector to calculate the current frame quantization After ISF parameters vector. 21. The device according to claim 19, wherein the vector quantization decoding module comprises: a residual codebook unit and a split vector quantization decoding module; The residual codebook unit stores a residual codebook of ISF parameters for query by the split vector quantization decoding module; The split vector quantization decoding module is configured to query the corresponding current frame quantized ISF parameter prediction residual vector in the residual codebook stored in the residual codebook unit according to the codebook index value parsed from the code stream, The quantized ISF parameter prediction residual vector of the current frame is provided to the unbiased ISF parameter vector reconstruction module.

Images