WO2005096273A1 - Enhanced audio encoding/decoding device and method - Google Patents

Abstract

An enhanced audio encoding device is consisted of a psychoacoustical analyzing module, a time-frequency mapping module, a quantization and entropy encoding module, a bit-stream multiplexing module, a signal characteristic analyzing module and a multi-resolution analyzing module, in which the signal characteristic analyzing module is configured to analyze the signal type of the input audio signal, the psychoacoustical analyzing module calculates a masking threshold and a signal-to-masking ratio of the audio signal, and outputs to said quantization and entropy encoding module, the multi-resolution analyzing module is configured to perform an analysis of the multi-resolution based on the signal type, the quantization and entropy encoding module performs a quantization and an entropy encoding of the frequency-domain coefficients under the control of the signal-to-masking ratio, and the bit-stream multiplexing module forms an audio encoding code stream. The device can support the ratio signal whose sampling rate is between 8kHz and 192kHz.

Description

 Enhanced audio codec device and method

 The present invention relates to the field of audio codec technology, and in particular to an enhanced audio codec device and method based on a perceptual model. Background technique

 In order to obtain a high-fidelity digital audio signal, the digital audio signal is audio encoded or audio compressed for storage and transmission. The purpose of encoding an audio signal is to achieve a transparent representation of the audio signal with as few bits as possible, for example, there is little difference between the originally input audio signal and the encoded output audio signal.

 In the early 1980s, the advent of CDs represented the many advantages of digitally representing audio signals, such as high fidelity, large dynamic range, and robustness. However, these advantages are at the expense of high data rates. For example, the sample rate of the CD-quality stereo signal requires a sampling rate of 44.1 kHz, and each sample value is uniformly quantized with 15 bits, so that the uncompressed data rate reaches 1.41 Mb/s. Such a high data rate brings great inconvenience to the transmission and storage of data, especially in the case of multimedia applications and wireless transmission applications, and is limited by bandwidth and cost. In order to maintain high quality audio signals, new network and wireless multimedia digital audio systems are required to reduce the rate of data without compromising the quality of the audio. In response to the above problems, various audio compression technologies have been proposed which can obtain high compression ratio and high fidelity audio signals, and typically have the MPEG-1/-2/-4 technology of the International Organization for Standardization (ISO) EC0 EC, Dolby's AC-2/AC-3 technology, Sony's ATRAC/MiniDisc/SDDS technology, and Lucent's PAC/EPAC/MPAC technology. The following is a detailed description of MPEG-2 AAC technology and Dolby AC-3 technology.

MPEG-1 technology and MPEG-2 BC technology are high-quality audio coding techniques primarily used for mono and stereo audio signals, with the need for multi-channel audio coding that achieves higher coding quality at lower bit rates. Increasingly, since MPEG-2 BC encoding technology emphasizes backward compatibility with MPEG-1 technology, it is impossible to achieve high-quality encoding of five channels at a code rate lower than 540 kbps. In response to this shortcoming, the MPEG- ² AAC technology is proposed, which can achieve higher quality encoding for five-channel signals at a rate of 320 kbps.

Figure 1 shows a block diagram of an MPEG-2 AAC encoder comprising a gain controller 101, a filter bank 102, a time domain noise shaping module 103, an intensity/coupling module 104, a psychoacoustic model, a second order backward self. An adaptive predictor 105, a / difference stereo module 106, a bit allocation and quantization encoding module 107, and a bitstream multiplexing module 108, wherein the bit allocation and quantization encoding module 107 further includes a compression ratio/distortion processing controller, a scale factor module, Non-uniform quantizer and entropy coding module.

The filter bank 102 employs a modified discrete cosine transform (MDCT) whose resolution is signal adaptive, that is, a 2048-point MDCT transform is used for the steady-state signal, and a 256-point MDCT transform is used for the transient signal; thus, for ⁴⁸ The kHz-like signal has a maximum frequency resolution of 23 Hz and a maximum time resolution of 2.6 ms. At the same time, a sine window and a Kaiser-Bessel window can be used in the filter bank 102, and the sinusoidal window is used when the harmonic interval of the input signal is less than 140 Hz, and the Ka iser-Bessel window is used when the strong component interval of the input signal is greater than 220 Hz. .

 After the audio signal passes through the gain controller 101, it enters the filter bank 102, performs filtering according to different signals, and then processes the spectral coefficients output by the filter bank 102 through the time domain noise shaping module 103. The time domain noise shaping technique is in the frequency domain. Linear prediction analysis is performed on the spectral coefficients, and then the quantization noise is controlled in the time domain according to the above analysis, thereby achieving the purpose of controlling the pre-echo.

 The intensity/coupling module 104 is for stereo encoding of signal strength, since for a high frequency band (greater than 2 kHz) the sense of direction of the hearing is related to the change in signal strength (signal envelope), regardless of the waveform of the signal, That is, the constant envelope signal has no influence on the sense of direction of the hearing, so this feature and the related information between multiple channels can be used to combine several channels into one common channel for encoding, which forms an intensity/coupling technique.

 The second-order backward adaptive predictor 105 is used to eliminate redundancy of the steady state signal and improve coding efficiency. The and difference stereo (M/S) module 106 operates for a pair of channels, which are two channels for capturing left and right channels or left and right surround channels, such as a two-channel signal or a multi-channel signal. The M/S module 106 utilizes the correlation between the two channels of the channel pair to achieve the effect of reducing the code rate and improving the coding efficiency. The bit allocation and quantization coding module 107 is implemented by a nested loop process in which the non-uniform quantizer performs lossy coding and the entropy coding module performs lossless coding, which removes redundancy and reduces correlation. The nested loop includes an inner loop and an outer loop, wherein the inner loop adjusts the step size of the non-uniform quantizer until the supplied bits are used up, and the outer loop estimates the encoding of the ^ using the ratio of the quantization noise to the masking threshold. quality. The last encoded signal forms an encoded audio stream output through bitstream multiplexing module 108.

 In the case where the sampling rate is scalable, the input signal simultaneously generates four equal-bandwidth bands in the quad-band polyphase filter bank (PQF), and each band uses MDCT to generate 256 spectral coefficients, for a total of 1024. A gain controller 101 is used in each band. In the decoder, the high frequency PQF band can be ignored to obtain a low sampling rate signal.

Figure 2 shows a block diagram of the corresponding MPEG-2 AAC decoder. The decoder includes a bitstream demultiplexing block 201, a lossless decoding module 202, an inverse quantizer 203, a scale factor module 204, and a / difference stereo (M/S) module 205, a prediction module 206, an intensity/coupling module 207, The time domain noise shaping module 208, the filter bank 209, and the gain ^empty module 210. The encoded audio stream is demultiplexed by the bitstream demultiplexing module 201 to obtain a corresponding data stream and control stream. After the above signal is decoded by the lossless decoding module 202, an integer representation of the scale factor and a signal spectrum are obtained. Quantitative value. The inverse quantizer 203 is a set of non-uniform quantizers implemented by a companding function for converting integer quantized values into reconstruction Fu. Since the scale factor module in the encoder differentiates the current scale factor from the previous scale factor and then uses the Huffman code for the difference value, the scale factor module 204 in the decoder performs Huffman decoding to obtain the corresponding difference value, and then recovers. A true scale factor. The M/S module 205 converts the difference channel to the left and right channels under the control of the side information. Since the second order backward adaptive predictor 105 is used in the encoder to eliminate the redundancy of the steady state signal and improve the coding efficiency, the prediction decoding is performed by the prediction module 206 in the decoder. The intensity/coupling module 207 performs intensity/coupling decoding under the control of the side information, and then outputs it to the time domain noise shaping module 208 for time domain noise shaping decoding, and finally performs comprehensive filtering by the filter bank 209, and the filter bank 209 adopts the reverse direction. Improved Discrete Cosine Transform (IMDCT) technique.

 For the case where the sampling frequency is scalable, the high frequency PQF band can be ignored by the gain control module 210 to obtain a low sampling rate signal.

 MPEG-2 AAC codec technology is suitable for medium and high bit rate audio signals, but the encoding quality of low bit rate or low bit rate audio signals is poor. At the same time, the codec technology involves more codec modules. High complexity is not conducive to real-time implementation.

 Figure 3 shows the structure of the encoder using Dolby AC-3 technology, including the transient signal detection module 301, ? The discrete cosine transform filter MDCT 302, the frequency borrowing envelope/exponential encoding module 303, the mantissa encoding module 304, the forward-backward adaptive sensing model 305, the parameter bit allocation module 306, and the bit stream multiplexing module 307.

 The audio signal is judged to be a steady state signal or a transient signal by the transient signal detecting module 301, and the time domain data is mapped to the frequency domain data by the signal adaptive MDCT filter bank 302, wherein a long window of 512 points is applied to the steady state signal. , a pair of short windows applied to the transient signal.

The spectral envelope/exponential encoding module 303 encodes the exponential portions of the signals in three modes according to the requirements of the code rate and frequency resolution, namely the D15, D25, and D45 encoding modes. The AC-3 technology differentially encodes the spectral envelope in frequency because at most ± 2 increments are required, each increment represents a 6 dB level change, the first DC term is absolute coded, and the remaining indices are differential. coding. In the D15 spectral envelope index coding, each index requires approximately 1.33 bits, and three difference packets are encoded in a 7-bit word length. The D15 coding mode provides fine frequency resolution by sacrificing temporal resolution. Since fine frequency resolution is only required for relatively stationary signals, and the frequency of such signals remains relatively constant over many blocks, D15 is occasionally transmitted for steady state signals, typically every 6 sound blocks ( The spectral envelope of a data frame is transmitted once. When the signal spectrum is unstable, the frequency estimate needs to be updated frequently. Estimates are encoded with a smaller frequency resolution, usually using the D25 and D45 encoding modes. The D25 encoding mode provides suitable frequency resolution and time resolution, and differential encoding is performed every other frequency coefficient. Each index requires approximately 1.15 bits. When the frequency is stable on 2 to 3 blocks and then suddenly changes, the D25 encoding mode can be used. The D45 coding mode is to perform a difference encoding every three frequency coefficients, so that each index needs about 0.58 bits. The D45 encoding mode provides high temporal resolution and low frequency resolution, so it is generally used in the encoding of transient signals.

 The forward-backward adaptive sensing model 305 is used to estimate the masking threshold of each frame of the signal. The forward adaptive part is only applied to the encoder end. Under the limitation of the code rate, a set of the most perceptual model parameters is estimated by iterative loop, and then these parameters are passed to the backward adaptive part to estimate each frame. Mask the threshold. The post-adaptive part is applied to both the encoder side and the decoder side.

 The parameter bit allocation module 306 analyzes the frequency of the audio signal according to the masking criteria to determine the number of bits to assign to each mantissa. The module 306 utilizes a bit pool for global ratio allocation of all channels. When encoding is performed in the mantissa encoding module 304, bits are cyclically extracted from the bit pool and allocated to all channels, and the quantization of the mantissa is adjusted in accordance with the number of available bits. For the purpose of compression coding, the AC-3 encoder also uses high-frequency coupling technology to divide the high-frequency part of the coupled signal into 18 sub-bands according to the critical bandwidth of the human ear, and then select some channels from a certain sub-band. Start coupling. Finally, the AC-3 audio stream is formed by the bit stream multiplexing module 307.

Figure 4 shows a schematic diagram of the process using Dolby AC-3 decoding. First, the bit stream encoded by the ϋ AC-3 encoder is input, and data frame synchronization and error detection are performed on the bit stream. If a data code is detected, error concealment or mute processing is performed. Then the bit stream is unpacked to obtain the main information and the side information, and then the index is decoded. In the exponential decoding, two side information is needed: one is the number of indexed packets; one is the indexing strategy of the mining, such as D15, D25 or D45 mode. The decoded index and bit allocation side information are then bit-matched, indicating the number of bits used for each packed mantissa, resulting in a set of bit allocation pointers, each bit allocation pointer corresponding to an encoded mantissa. The bit allocation pointer indicates the quantizer used for the mantissa and the number of bits occupied by each mantissa in the code stream. The single coded mantissa value is dequantized, converted into a dequantized value, and the mantissa occupying zero bits is complexed to zero by zero, or replaced by a random jitter value under the control of the jitter flag. The decoupling operation is then performed, and the decoupling is to recover the high frequency portion of the coupled channel from the common coupling channel and the coupling factor, including the exponent and the mantissa. For example, when the encoding end adopts 2/0 mode encoding, matrix processing is applied to a certain sub-band, then the sub-band and the difference channel value are converted into left and right channel values by matrix recovery at the decoding end. The dynamic range control of each audio block is included in the code stream to dynamically range the value to change the magnitude of the coefficients, including the exponent and mantissa. The frequency domain coefficients are inverse transformed and converted into time domain samples. Then, the time domain samples are windowed, and the adjacent blocks are overlapped and added to reconstruct the PC1M audio signal. When the number of channels outputted by the decoding is smaller than the number of channels in the encoded bit stream, the audio signal needs to be mixed, and finally the PCM is output. Dolby AC- ³ encoding technology is mainly for high bit rate multi-channel surround sound signals, but when the 5.1 bit encoding bit rate is lower than 384kbps, its encoding effect is poor; and for mono and dual sound Channel stereo coding efficiency is also low.

 In summary, the existing codec technology cannot comprehensively solve the coding and decoding quality from very low code rate, low bit rate to high bit rate audio signal and mono channel and silent channel signals, and the implementation is complicated. Summary of the invention

 The technical problem to be solved by the present invention is to provide an apparatus and method for enhancing audio encoding/decoding to solve the problem of low coding efficiency and poor quality of the lower rate audio signal in the prior art.

 The enhanced audio coding device of the present invention comprises a psychoacoustic analysis module, a time-frequency mapping module, a quantization and entropy coding module, and a bit stream multiplexing module, a signal property analysis module and a multi-resolution analysis module; wherein the signal property analysis module, And performing type analysis on the input audio signal, and outputting to the psychoacoustic analysis module and the time-frequency mapping module, and outputting signal type analysis result information to the bit stream multiplexing module; the psychoacoustic analysis module And calculating a masking threshold and a mask ratio of the audio signal, and outputting to the quantization and entropy encoding module; the time-frequency mapping module, configured to convert the time domain audio signal into a frequency domain coefficient, and output to the multi-resolution The multi-resolution analysis module is configured to perform multi-resolution analysis on the frequency domain coefficients of the fast-changing type signal according to the signal type analysis result output by the signal property analysis module, and output the same to the quantization and entropy coding module. The quantization and entropy coding module, outputted by the psychoacoustic analysis module Under the control of the mask ratio, the frequency domain coefficients are quantized and entropy encoded, and output to the bit stream multiplexing module; the bit stream multiplexing module is configured to multiplex the received data to form an audio code. Code stream.

 The enhanced audio decoding apparatus of the present invention comprises: a bit stream demultiplexing module, an entropy decoding module, an inverse quantizer group, a frequency-time mapping module, and a multi-resolution synthesis module; and the bit stream demultiplexing module is used for compression And demultiplexing the audio data stream, and outputting corresponding data signals and control signals to the entropy decoding module and the multi-resolution synthesis module; the entropy decoding module is configured to decode the foregoing signals, and restore spectrum quantization And outputting to the inverse quantizer group; the inverse quantizer group is configured to reconstruct an inverse quantization spectrum, and outputting to the multi-resolution synthesis module; the multi-resolution synthesis module is configured to perform inverse quantization The resolution is integrated and output to the frequency-time mapping module; the frequency-time mapping module is configured to perform frequency-time mapping on the spectral coefficients, and output a time domain audio signal.

The invention is applicable to high-fidelity compression coding of audio signals of various sampling rates and channel configurations, and can support audio signals with sampling rates between 8 kHz and 192 kHz; all possible channel configurations can be supported; and a wide range of support is supported. Audio encoding/decoding of the target bit rate. DRAWINGS

 Figure 1 is a block diagram of an MPEG-2 AAC encoder;

 Figure 2 is a block diagram of an MPEG-2 AAC decoder;

 Figure 3 is a schematic structural view of an encoder using Dolby AC-3 technology;

 Figure 4 is a schematic diagram of a decoding process using Dolby AC-3 technology;

 Figure 5 is a schematic structural view of an encoding device of the present invention;

 6 is a schematic diagram of a filtering structure using a Harr wavelet-based wavelet transform;

 Figure 7 is a schematic diagram of time-frequency division obtained by using Harr wavelet-based wavelet transform;

 8 is a schematic structural diagram of a decoding apparatus of the present invention;

 Figure 9 is a schematic structural view of Embodiment 1 of the coding apparatus of the present invention;

 FIG. 10 is a schematic structural diagram of Embodiment 1 of a decoding apparatus according to the present invention; FIG.

 Figure 11 is a schematic structural view of Embodiment 2 of the encoding apparatus of the present invention;

 FIG. 12 is a schematic structural diagram of Embodiment 2 of a decoding apparatus according to the present invention; FIG.

 Figure 13 is a schematic structural view of a third embodiment of the encoding apparatus of the present invention;

 FIG. 14 is a schematic structural diagram of Embodiment 3 of a decoding apparatus according to the present invention; FIG.

 Figure 15 is a schematic structural view of Embodiment 4 of the encoding apparatus of the present invention;

 16 is a schematic structural diagram of Embodiment 4 of a decoding apparatus according to the present invention;

 Figure 17 is a schematic structural view of Embodiment 5 of the coding apparatus of the present invention;

 18 is a schematic structural diagram of Embodiment 5 of a decoding apparatus according to the present invention;

 Figure 19 is a schematic structural view of Embodiment 6 of the encoding apparatus of the present invention;

 20 is a schematic structural diagram of Embodiment 6 of a decoding apparatus according to the present invention;

 Figure 21 is a schematic structural view of Embodiment 7 of the coding apparatus of the present invention;

 Figure 22 is a block diagram showing the structure of a seventh embodiment of the decoding apparatus of the present invention. Detailed ways

 Fig. 1 to Fig. 4 are schematic diagrams showing the structure of several encoders of the prior art, which have been introduced in the background, and are not mentioned here.

It should be noted that, in order to explain the present invention conveniently and clearly, the specific embodiment of the following codec device is described in a corresponding manner, but the encoding device and the decoding device are not necessarily limited. As shown in FIG. 5, the audio encoding apparatus provided by the present invention includes a signal property analyzing module 50, a psychoacoustic analyzing module 51, a time-frequency mapping module 52, a multi-resolution analyzing module 53, a quantization and entropy encoding module 54, and bit stream multiplexing. The module 55 is configured to perform type analysis on the input audio signal, output the audio signal to the psychoacoustic analysis module 51 and the time-frequency mapping module 52, and output the signal type analysis result to the bit stream multiplexing module 55. The psychoacoustic analysis module 51 is configured to calculate a masking threshold and a signal mask ratio of the input audio signal, and output to the quantization and entropy encoding module 54. The time-frequency mapping module 52 is configured to convert the time domain audio signal into frequency domain coefficients and output the same. The resolution analysis module 53 is configured to perform multi-resolution analysis on the frequency domain coefficients of the fast-changing type signal according to the signal type analysis result output by the psychoacoustic analysis module 51, and output to the quantization and entropy coding module. 54; the quantization and entropy coding module 54 is used to control the frequency domain system under the control of the mask ratio output by the psychoacoustic analysis module 51. The number is quantized and entropy encoded and output to the bit stream multiplexing module 55; the bit stream multiplexing module 55 is configured to multiplex the received data to form an audio encoded code stream.

 The digital audio signal is subjected to signal type analysis in the signal property analysis module 50, and the type information of the audio signal is output to the bit stream multiplexing module 55; and the audio signal is simultaneously output to the psychoacoustic analysis module 51 and the time-frequency map. In the module 52, on the one hand, the masking threshold and the mask ratio of the frame audio signal are calculated in the psychoacoustic analysis module 51, and then the mask ratio is transmitted as a control signal to the quantization and entropy encoding module 54; The signal is converted into frequency domain coefficients by the time-frequency mapping module 52; the frequency domain coefficients are multi-resolution analysis module 53 in the multi-resolution analysis module 53 to improve the time resolution of the fast-changing signal, and output the result to In the quantization and entropy coding module 54; under the control of the mask ratio output by the psychoacoustic analysis module 51, quantization and entropy coding are performed in the quantization and entropy coding module 54, and the encoded data and control signals are multiplexed in the bit stream. Module 55 multiplexes to form a code stream that enhances the audio coding.

 The respective constituent modules of the above audio encoding device will be described in detail below.

 The signal property analysis module 50 performs signal type analysis on the input audio signal, and outputs the type information of the audio signal to the bit stream multiplexing module 55; and simultaneously outputs the audio signal to the psychoacoustic analysis module 51 and the time-frequency mapping module 52. .

 The signal property analysis module 50 performs front and back masking effect analysis based on the adaptive threshold and the waveform prediction to determine whether the signal type is a slow-changing signal or a fast-changing signal, and if it is a fast-changing type signal, continues to calculate related parameter information of the abrupt component. Such as the location of the mutation signal and the strength of the mutation signal.

The psychoacoustic analysis module 51 is mainly used to calculate the masking threshold, the mask ratio and the perceptual entropy of the input audio signal. The perceptual entropy calculated by the psychoacoustic analysis module 51 can dynamically analyze the number of bits required for the current signal frame to be transparently encoded, thereby adjusting the bit allocation between frames. The psychoacoustic analysis module 51 outputs the mask ratio of each sub-band to the quantized sum The entropy encoding module 54 controls it.

 The time-frequency mapping module 52 is configured to implement the transformation of the audio signal from the time domain signal to the frequency domain coefficient, and is composed of a filter bank, and specifically may be a discrete Fourier transform (DFT) filter bank, a discrete cosine transform (DCT) filter bank, Modified discrete cosine transform (MDCT) filter bank, cosine modulated filter bank, wavelet transform filter bank, etc. The frequency domain coefficients obtained by the time-frequency mapping are output to the quantization and entropy encoding module 54, and quantization and encoding processing is performed.

 For the fast-changing type signal, in order to effectively overcome the pre-echo phenomenon generated in the encoding process and improve the encoding quality, the encoding apparatus of the present invention increases the time resolution of the encoded fast-changing signal by the multi-resolution analyzing module 53. The frequency domain coefficients output by the time-frequency mapping module 52 are input to the multi-resolution analysis module 53. If it is a fast-varying type signal, the frequency domain wavelet transform or the frequency domain modified discrete cosine transform (MDCT) is performed to obtain the frequency domain coefficients. The multi-resolution representation is output to the quantization and entropy encoding module 54. If it is a slowly varying type signal, the frequency domain coefficients are not processed and are directly output to the quantization and entropy encoding module 54.

 The multi-resolution analysis module 53 includes a frequency domain coefficient transform module and a recombination module, wherein the frequency domain coefficient transform module is configured to transform the frequency domain coefficients into time-frequency plane coefficients; the recombination module is configured to reorganize the time-frequency plane coefficients according to certain rules. . The frequency domain coefficient transform module may adopt a frequency domain wavelet transform filter bank, a frequency domain MDCT transform filter bank, or the like.

 The quantization and chirp encoding module 54 further includes a non-linear quantizer group and an encoder, where the quantizer can be a scalar quantizer or a vector quantizer. The vector quantizer is further divided into two categories: memoryless vector quantizer and memory vector quantizer. For a memoryless vector quantizer, each input vector is independently quantized, independent of the previous vectors; a memory vector quantizer considers the previous vector when quantizing a vector, ie, exploits the correlation between vectors. The main memoryless vector quantizers include a full search vector quantizer, a tree search vector quantizer, a multilevel vector quantizer, a gain/waveform vector quantizer, and a separate mean vector quantizer; the main memory vector quantizer includes predictive vector quantization And finite state vector quantizer.

 If a scalar quantizer is employed, the non-linear quantizer group further includes M sub-band quantizers. In each sub-band quantizer, the scale factor is mainly used for quantization, specifically: nonlinearly compressing all the frequency domain coefficients in the M scale factor bands, and then using the scale factor to quantize the frequency domain coefficients of the sub-band, The quantized spectrum represented by the integer is output to the encoder, and the first scale factor in each frame signal is output to the bit stream multiplexing module 55 as a common scale factor, and other scale factors are differentially processed with the previous scale factor and output to the encoder. .

 The scale factor in the above steps is a constantly changing value, which is adjusted according to the bit allocation strategy. The present invention provides a bit allocation strategy with minimal global perceptual distortion, as follows:

First, each sub-band quantizer is initialized, and an appropriate scale factor is selected such that the quantized value of the spectral coefficients in all sub-bands is zero. At this time, the quantization noise of each sub-band is equal to the energy value of each sub-band, and the noise masking ratio of each sub-band is NMR, etc. In its letter mask ratio SMR, the number of bits consumed for quantization is 0, and the number of remaining bits is equal to the number of target bits.

 Secondly, finding the subband with the largest noise masking ratio NMR, if the maximum noise masking ratio R is less than or equal to 1, the scale factor is unchanged, the output allocation result is output, and the bit allocation process ends; otherwise, the scale factor of the corresponding subband quantizer is Decrease by one unit and then calculate the number of bits Δ5,. (β) that the subband needs to increase. If the remaining bits of the subband

B _t ≥AB _i (Q _i ), then confirm the modification of the scale factor, and subtract Δ5,.(3⁄4) from the remaining number of bits A, recalculate the noise mask of the sub-band, and then continue to find the noise. Masking the largest subband of Bili R, repeat the subsequent steps. If the number of remaining bits of the subband is < Δ5, (ρ,), the modification is canceled, the previous scale factor and the remaining number of bits are retained, and finally the allocation result is output, and the bit allocation process ends.

 If a vector quantizer is used, the frequency domain coefficients are composed into a plurality of dimensional vectors and input into the nonlinear quantizer group. For each dimension vector, the flattening factor is used to perform the flattening, that is, the dynamic range of the chirp is reduced, and then the vector quantizer is used. The subjective perceptual distance measure criterion finds the codeword with the smallest distance from the vector to be quantized in the codebook, and transmits the corresponding codeword index to the encoder. The leveling factor is adjusted according to the bit allocation strategy of vector quantization, and the bit allocation of vector quantization is controlled according to the perceived importance between different sub-bands.

 After the above quantization process, the entropy coding technique is used to further remove the quantized coefficients and the statistical redundancy of the side information. Entropy coding is a kind of source coding technology. The basic idea is to give shorter-length codewords to symbols with higher probability of occurrence and longer codewords to symbols with lower probability of occurrence, so that the length of average codewords The shortest. According to Shannon's noiseless coding theorem, if the symbols of the transmitted N source messages are independent, then using the appropriate variable length coding, the average length of the codeword will satisfy H(x) _ 1

≤ n < , where represents the entropy of the source, log ₂ (£>) log ₂ (D) N shows the symbol variable. Since entropy #W is the shortest limit of the average codeword length, the above formula indicates that the average length of the codeword is very close to its lower bound entropy at this time. Therefore, this variable length coding technique becomes "entropy coding". The entropy coding mainly includes methods such as Huffman coding, arithmetic coding or run length coding, and any entropy coding in the present invention may adopt any of the above coding methods.

 After being quantized by the scalar quantizer, the quantized spectrum and the scaled factor after the differential processing are entropy encoded in the encoder, and the code book serial number, the scale factor coded value and the lossless coded quantized spectrum are obtained, and then the code book number is entropy coded. The code book serial number encodes the value, and then outputs the scale factor code value, the code book sequence code value, and the lossless code quantizer 到 to the bit stream multiplexing module 55.

The codeword index obtained by the vector quantizer quantization is subjected to one-dimensional or multi-dimensional entropy coding in the encoder to obtain an encoded value of the codeword index, and then the encoded value of the codeword index is output to the bitstream multiplexing module 55. The encoding method based on the above encoder specifically includes: performing signal type analysis on the input audio signal; calculating a signal mask ratio of the audio signal; performing time-frequency mapping on the audio signal to obtain a frequency domain coefficient of the audio signal; and performing more on the frequency domain coefficient Resolution analysis and quantization and entropy coding; multiplexing the signal type analysis result and the encoded audio code stream to obtain a compressed audio code stream.

 The analysis signal type is determined based on adaptive threshold and waveform prediction for front and back masking effect analysis. The specific steps are: decomposing the input audio data into frames; decomposing the input frame into multiple sub-frames, and finding each sub-frame The local maximum point of the absolute value of the upper PCM data; the subframe peak value is selected in the local maximum point of each subframe; for a certain subframe peak, a plurality of (typically 3) subframe peak predictions in front of the subframe are utilized Typical sample values of a plurality of (typically 4) subframes relative to the forward delay of the subframe; calculating a difference and a ratio of the peak value of the subframe to the predicted typical sample value; if both the predicted difference and the ratio are If it is greater than the set threshold, it is determined that there is a sudden signal in the sub-frame, and the sub-frame has a local maximum peak point of the back-masking pre-echo capability, if between the front end of the sub-frame and the mask peak, 2.5 ms. If there is a sub-frame with a sufficiently small peak, it is judged that the frame signal belongs to the fast-changing type signal; if the predicted difference value and the ratio are not greater than the set threshold, the above steps are repeated until it is determined The frame signal is a fast-changing type signal or reaches the last sub-frame. If the last sub-frame is not determined to be a fast-changing type signal, the frame signal belongs to a slowly varying type signal.

 There are many methods for time-frequency transform of time-domain audio signals, such as discrete Fourier transform (DFT), discrete cosine transform (DCT), modified discrete cosine transform (MDCT), cosine-modulated filter bank, wavelet transform, and so on. The following describes the process of time-frequency mapping by taking the modified discrete cosine transform MDCT and cosine modulation filtering as an example.

 For the case of time-frequency transform using the modified discrete cosine transform (MDCT), first select the time domain signals of the previous frame and the current frame samples, and then window the time domain signals of the two frames of the two frames, and then The MDCT transform is performed on the windowed signal to obtain a frequency domain coefficient.

则 MDCT变换为： X{k) = ∑ x(n)h_k ( ) 0≤k≤M - l , 其中： 为窗函数; x (n)^} MDCT 变换的输入时域信号； X MDCT变换的输出频域信号。 The impulse response of the MDCT analysis filter is:

Then the MDCT transform is: X{k) = ∑ x(n)h _k ( ) 0≤k≤M - l , where: is the window function; x (n)^} the input time domain signal of the MDCT transform; X MDCT transform Output frequency domain signal.

 In order to satisfy the condition of complete signal reconstruction, the window function of the MDCT transform must satisfy the following two conditions:

w(2M - l - n) = w(n) and w ² (n) + w ² (n + M) = l . In practice, the S ine window can be selected as the window function. Of course, it is also possible to use a bi-orthogonal transformation to specify The analysis filter and synthesis filter modify the above limitations on the window function.

 For the case of time-frequency transform using cosine modulation filtering, first select the time domain signal of the previous frame sample and the M frame of the current frame, and then perform windowing operation on the time domain signals of two samples of the two frames, and then A cosine modulation transform is performed on the windowed signal to obtain a frequency domain coefficient.

 The impulse response of the traditional cosine modulation filtering technique is

K (") = 2p _a (") + 0.5)(" - 3⁄4 + 3⁄4

n=QX"', N _h -1

n = 0, l, ---, N _f - 1 where 0 ≤ < Μ - 1, 0 ≤ η < 2 ΚΜ -1, is an integer greater than zero, 1) * ^. Assume that the analysis response window (analytical prototype filter) of the dice with cosine-modulated filter bank has a shock response length of Ν _α , and the integrated window (integrated prototype filter) p _s ( ) has an impulse response length of N _s . When the analysis window and the synthesis window are equal, that is, P. (n~) = p _s (n), RN _a = N _s , the cosine-modulated filter bank represented by the above two equations is an orthogonal filter bank, where matrix H and ( [H] _nk = h _k ( n), [F] _nJ[ = ΛΟ)) is an orthogonal transformation matrix. To obtain a linear phase filter bank, it is further specified that the symmetric window satisfies ρ _α (2ΚΜ— \− ) = _Ρα ( ). In order to ensure the complete reconfiguration of orthogonal and biorthogonal systems, the window function also needs to meet certain conditions, see the literature "Multirate Systems and Filter Banks", PP Vaidynathan, Prentice Hal 1, Englewood Cliffs, NJ, 1993.

 Calculating the masking value and the mask ratio of the resampled signal includes the following steps:

 The first step is to map the signal from time domain to frequency domain. The fast Fourier transform and the hanning window technique can be used to convert the time domain data into frequency domain coefficients l], expressed by the amplitude r] and the phase

X[k]^r[k]e ^jm , then the energy of each sub-band is the sum of the energy of all the spectral lines in the sub-band, that is, the sum >, respectively representing the upper and lower boundaries of the sub-band b.

The second step is to determine the pitch and non-tonal components in the signal. The tonality of the signal is estimated by inter-predicting each spectral line. The Euclidean distance between the predicted and true values of each squall line is mapped to an unpredictable measure, and the highly predictive spectral components are considered to be tones. Very strong, and low-predictive spectral components are considered to be noise-like. The amplitude r _pred and the phase φ _{ρκά of the} predicted value can be expressed by the following formula: r _pred \ =, [k] + {r _t _ [k] - r _t _ ₂ [k]) where ί represents the coefficient of the current frame; -1 indicates the coefficient of the previous frame; Bu 2 indicates the coefficient of the first two frames.

Then, the formula for calculating the unpredictable measure is:

Among them, the Euclidean distance i3⁄4t (; is calculated by the following formula: di

Therefore, the unpredictability of each subband is the weighted sum of the energy of all the spectral lines in the subband to its unpredictability. Subband energy and unpredictability c[b] are convoluted separately with the extension function

Calculate, obtain subband energy extension and subband unpredictability extension ^: ό], mask The extension function for subband 6 is expressed as sU, ]. In order to eliminate the influence of the extension function on the energy transformation, the sub-band unpredictability extension needs to be normalized, and the normalized result is expressed as [δ] = ^ by [b]. Similarly, to eliminate the effect of the extension function on the subband energy, define the normalized energy extension 5; | ] is [ ]:=^, where the normalization factor is: n[b]

 Bmax

n[b]=∑∑s[i,b], ό„„ is the number of subbands divided by the frame signal. According to the normalized unpredictability expansion, the pitch of the subband can be calculated: t[b] = -0.299 - 0.43 log _e (c _s [b]) , and 0 ≤ t[6] ≤ l. When [6]=1, it indicates that the subband signal is pure tone; when [6]=0, it indicates that the subband signal is white noise.

The third step is to calculate the Signal-to-Noise Ratio (SNR) required for each sub-band. Set the value of the noise masking tone (Noise-Masking-Tone) of all sub-bands to 5dB, and the value of Tone-Masking-Noise (TMN) to 18dB. When perceived, the required signal-to-noise ratio for each sub-band ₁ 51⁄2 [0] is ₁ 3⁄4, [6] = 18 ] + 6(1- ]). The fourth step is to calculate the masking threshold of each subband and the perceptual entropy of the signal. According to the normalized signal energy of each sub-band obtained by the foregoing steps and the required signal-to-noise ratio SNR, the noise energy threshold of each sub-band is calculated as «[b] = [6] 10 - fiber ⁰ . In order to avoid the influence of pre-echo, the noise energy threshold [6] of the current frame is compared with the noise energy threshold n _prev [b] of the previous frame. 'The masking threshold of the signal is = m(n[b], 2n _prev [b]) , This ensures that the masking threshold does not deviate due to high energy impacts at the near end of the analysis window.

Further, considering the influence of the static masking threshold ^ rW, the masking threshold for selecting the final signal is greater than the value of the static masking threshold and the above calculated masking threshold, ie "[6] = max(n[b], Qsthr[b]). Then use the formula under the mouth to calculate the perceptual entropy, that is, 1⁄2 where cbiv represents each

The number of lines included in the band.

 Step 5: Calculate the signal-to-mask ratio (Signa to-Mask Ratio, SMR for short) of each sub-band signal. The mask ratio of each subband is SMi[b] = for SMR [b].

 Multi-resolution analysis of the frequency domain coefficients is then performed. The multi-resolution analysis module 53 performs time-frequency salt re-organization on the input frequency domain data, and improves the time resolution of the frequency domain data at the cost of reducing the frequency precision, thereby automatically adapting to the time-frequency characteristics of the fast-changing type signal, The effect of the pre-echo is suppressed, and the form of the filter bank in the time-frequency mapping module 52 need not be adjusted.

 The multi-resolution analysis includes two steps of frequency domain coefficient transform and recombination, wherein frequency domain coefficients are transformed into time-frequency plane coefficients by frequency domain coefficient transform; time-frequency plane coefficients are grouped according to certain rules by recombination.

 The process of multi-resolution analysis is illustrated by taking the frequency domain wavelet transform and the frequency domain MDCT transform as examples.

 1) Frequency domain wavelet transform

 Assuming the time series x(f), i = 0,1,···, 2 -1 , the frequency domain coefficients obtained after time-frequency mapping are X(k), 1=0,

1, ..., Ml _a frequency domain wavelet or wavelet packet transform wavelet base can be fixed or adaptive.

 Let's take the simplest Harr wavelet-based wavelet transform as an example to illustrate the multi-resolution analysis of frequency domain coefficients.

The scale factor of the Harr wavelet base is -^], and Figure ⁶ shows the use of Harr.

1 1

^ Schematic diagram of the filtering structure of the wavelet transform, where H. Indicates low pass filtering (filter coefficient is ),

V2 V2 H, which means high-pass filtering (filter coefficient is), and "I 2" means 1 time of downsampling operation. For the frequency domain

 V2 1

The middle and low frequency parts of the number :), = 0, · · ·, ^ Do not perform wavelet transform, perform Harr wavelet transform on the high frequency part of the frequency domain coefficient, and obtain coefficients ₂ (k), X _{3 of} different time-frequency intervals. (k), X k), X ₅ (k) . X ₆ (k) ^ X ₇ (k) , the corresponding time-frequency plane division is shown in Fig. 7. Different wavelet bases can be selected, and different wavelet transform structures can be selected for processing, and other similar time-frequency plane partitions are obtained. Therefore, it is possible to adjust the time-frequency plane division of the signal analysis arbitrarily according to the needs, and to meet the analysis requirements of different time and frequency resolutions.

 The time-frequency plane coefficients are reorganized according to certain rules in the recombination module. For example: the time-frequency plane coefficients can be organized in the frequency direction, the coefficients in each frequency band are organized in the time direction, and then the well-organized coefficients are followed by the sub-window. , the order of the scale factor bands.

 2) Frequency domain MDCT transform

 Let the frequency domain data of the input frequency domain MDCT transform filter bank be X(l), 1= 0, 1 , ..., N-1, and perform MCT transform of M point in sequence for the N point frequency domain data, so that time The frequency accuracy of the frequency domain data is reduced, and the time accuracy is correspondingly improved. Different frequency-domain MDCT transforms can be used in different frequency domain ranges to obtain different time-frequency plane divisions, that is, different time and frequency precisions. The recombination module reorganizes the time-frequency domain data outputted by the frequency domain MDCT transform filter bank. A recombination method is to first organize the time-frequency plane coefficients in the frequency direction, and the coefficients in each frequency band are organized in the time direction, and then The organized coefficients are arranged in the order of the sub-window and the scale factor.

 Quantization and entropy coding further include two steps of nonlinear quantization and entropy coding, where the quantization can be scalar quantization or vector quantization.

 The scalar quantization includes the following steps: nonlinearly compressing the frequency domain coefficients in all scale factor bands; and then quantizing the frequency domain coefficients of the subbands by using the scale factor of each subband to obtain a quantized spectrum represented by an integer; The first scale factor in each frame of the signal is used as a common scale factor; other scale factors are differentially processed from their previous scale factor.

 Vector quantization includes the following steps: constituting a plurality of multi-dimensional vector signals by frequency domain coefficients; panning for each dimension vector according to a flattening factor; ■ finding a code having the smallest distance from a vector to be quantized in a codebook according to a subjective perceptual distance measure criterion Word, get its codeword index.

 The entropy coding step comprises: entropy coding the quantized spectrum and the differentially processed scale factor to obtain a codebook serial number, a scale factor coding value, and a lossless coding quantization language; entropy coding the codebook sequence number to obtain a codebook serial number coding value.

Or: Perform one-dimensional or multi-dimensional entropy coding on the codeword index to obtain the coded value of the codeword index. The above entropy coding method can adopt any of the existing methods such as Huffman coding, arithmetic coding or run length coding.

 After quantization and entropy coding processing, the encoded audio code stream is obtained, and the code stream is multiplexed together with the common scale factor and signal type analysis result to obtain a compressed audio code stream.

Figure 8 is a block diagram showing the structure of an audio decoding device of the present invention. The audio decoding apparatus includes a bit stream by demultiplexing module 60, entropy decoding module 61, an inverse quantizer group ^62, the multi-resolution integration module 63 and a frequency - time map module 64. After the compressed audio code stream is demultiplexed by the bit stream demultiplexing module 60, corresponding data signals and control signals are obtained, which are output to the entropy decoding module 61 and the multi-resolution synthesis module 63; the data signal and the control signal are in the entropy decoding module. In 61, the decoding process is performed, 'the quantized value of the spectrum is recovered. The above quantized values are reconstructed in the inverse quantizer group 62 to obtain an inverse quantized spectrum. The inverse quantized words are output to the multi-resolution synthesis module 63, and after multi-resolution synthesis, are output to the frequency-time mapping module 64, and then The frequency-time mapping yields an audio signal in the time domain.

 The bit stream demultiplexing module 60 decomposes the compressed audio stream to obtain corresponding data signals and control signals, and provides corresponding decoding information for other modules. After the compressed audio data stream is demultiplexed, the signal outputted to the entropy decoding module 61 includes a common scale factor, a scale factor coded value, a codebook sequence number coded value, and a lossless coded quantized spectrum, or an encoded value of the codeword index; The type information is sent to the multi-resolution synthesis module 63.

 If a scalar quantizer is employed in the quantization and entropy encoding module 54 in the encoding device, in the decoding device, the entropy decoding module 61 receives the common scale factor, scale factor encoded value, and code output by the bitstream demultiplexing module 60. The book serial number coded value and the lossless coded quantized spectrum are then subjected to codebook sequence number decoding, spectral coefficient decoding and scale factor decoding, reconstructed quantization, and output the integer representation of the scale factor and the quantization of the spectrum to the inverse quantizer group 62. value. The decoding method employed by the entropy decoding module 61 corresponds to an entropy-encoded encoding method in the encoding device, such as Huffman decoding, arithmetic decoding, or run-length decoding. ■

 After receiving the quantized value of the spectrum and the integer representation of the scale factor, the inverse quantizer group 62 inversely quantizes the quantized value of the spectrum into a non-scaled reconstruction (inverse quantization spectrum), and outputs the inverse quantization borrowing to the multi-resolution synthesis module 63. . The inverse quantizer group 62 may be a uniform quantizer group or a non-uniform quantizer group implemented by a companding function. In the encoding apparatus, the quantizer group employs a scalar quantizer, and the inverse quantizer group 62 also employs a scalar inverse quantizer in the decoding apparatus. In the scalar inverse quantity controller, the spectral quantized values are first nonlinearly expanded, and then each scale factor is used to obtain all the spectral coefficients (inverse quantized spectrum) in the corresponding scale factor bands.

If the vector quantizer is employed in the quantization and chirp encoding module 54, in the decoding apparatus, the entropy decoding module 61 receives the encoded value of the codeword index output by the bitstream multiplexing module 60, and uses the encoded value of the codeword index. The entropy decoding method corresponding to the entropy coding method at the time of encoding is decoded to obtain a corresponding codeword index. The codeword index is output to the inverse quantizer group 62, and the quantized value (inverse quantized spectrum) is obtained by querying the codebook, and output to the multi-resolution synthesis module 63. The inverse quantizer group 62 employs an inverse vector quantizer. After the inverse quantization spectrum is integrated by the multi-resolution, the time domain audio signal is obtained by the mapping process of the frequency-time mapping module 64. The frequency-time mapping module 64 may be an inverse discrete cosine transform (IDCT) filter bank, an inverse discrete Fourier transform (IDFT) filter bank, an inverse modified discrete cosine transform (IMDCT) filter bank, an inverse wavelet transform filter bank, and a cosine Modulation filter bank, etc.

 The decoding method based on the above decoder includes: demultiplexing the compressed audio code stream to obtain data information and control information; performing entropy decoding on the information to obtain a quantized value of the i-precision; performing inverse quantization processing on the quantized value of the spectrum, The inverse quantization is obtained; after the inverse quantization spectrum is multi-resolution integrated, frequency-time mapping is performed to obtain a time domain audio signal.

 If the demultiplexed information includes the codebook serial number encoding value, the common scale factor, the scale factor encoding value, and the lossless encoding quantization spectrum, it indicates that the ixiang coefficient is quantized by the scalar quantization technique in the encoding device, and then the decoding is performed. The steps include: decoding the code number of the code book to obtain the code book number of all the scale factor bands; decoding the quantized coefficients of all the scale factor bands according to the code book corresponding to the code book serial number; decoding the scale factors of all the scale factor bands, reconstructing Quantitative spectrum. The entropy decoding method adopted in the above process corresponds to an entropy coding method in the coding method, such as a run length decoding method, a Huffman decoding method, an arithmetic decoding method, and the like.

 In the following, the process of entropy decoding is illustrated by using a run-length decoding method to decode a codebook sequence number, a Huffman decoding method to decode a quantized coefficient, and a Huffman decoding method to decode a scale factor.

 First, the codebook number of all scale factor bands is obtained by the process decoding method, and the decoded codebook sequence number is an integer of a certain interval. If the interval is set to [0, 11], then only the valid range is within the valid range. That is, the codebook number between 0 and 11 corresponds to the corresponding spectral coefficient Huffman codebook. For all zero sub-bands, you can select a code book serial number corresponding to a typical optional 0 serial number.

 After decoding the codebook number of each scale factor band, the spectral coefficient Huffman codebook corresponding to the codebook number is used to decode the quantized coefficients of all the scales. If the codebook number of a scale factor band is within the valid range, and the implementation is, for example, between 1 and 11, then the codebook number corresponds to a spectral coefficient codebook, and the quantized coefficient of the scale factor band is decoded from the quantized spectrum using the codebook. The codeword index is then unpacked from the codeword index to obtain the quantized coefficients. If the codebook number of the scale factor band is not between 1 and 11, the codebook number does not correspond to any spectral coefficient codebook, and the quantized coefficient of the subband is not decoded, and the quantized coefficients of the subband are all set to zero.

The scale factor is used to reconstruct the i-value based on the inverse quantized spectral coefficients. If the codebook number of the scale factor is within the valid range, each codebook number corresponds to a scale factor. When decoding the above scale factor, first reading the code stream occupied by the first scale factor, and then performing Huffman decoding on other scale factors, and sequentially obtaining the difference between each scale factor and the previous scale factor, The difference is added to the previous scale factor value to obtain each scale factor. in case The quantized coefficients of the current subband are all zero, then the scale factor of the subband does not need to be decoded.

 After the above entropy decoding process, an integer representation of the spectral quantized value and the scale factor is obtained, and then the quantized value of the spectrum is inversely quantized to obtain an inverse quantized spectrum. The inverse quantization process includes: nonlinearly expanding the quantized values of the spectra; and obtaining all spectral coefficients (inverse quantized spectra) in the corresponding scale factor bands according to each scale factor.

 If the demultiplexed information includes the coded value of the codeword index, it indicates that the coding means quantizes the spectral coefficients by using a vector quantization technique, and then the step of decoding comprises: adopting an entropy corresponding to the entropy coding method in the coding device The decoding method decodes the encoded value of the codeword index to obtain a codeword index. The codeword index is then inverse quantized to obtain an inverse quantized spectrum.

 For the inverse quantization spectrum, if it is a fast-varying type signal, multi-resolution analysis is performed on the frequency domain coefficients, and then the multi-resolution representation of the frequency domain coefficients is quantized and entropy encoded; if it is not a fast-changing type signal, the frequency is directly The domain coefficients are quantized and entropy encoded.

 Multi-resolution synthesis can be performed by frequency domain wavelet transform or frequency domain MDCT transform. The frequency domain wavelet synthesis method includes: first recombining the above-mentioned time-frequency plane coefficients according to a certain rule; then performing wavelet transform on the frequency domain coefficients to obtain a time-frequency plane coefficient. The MDCT transform method includes: first recombining the above-mentioned time-frequency plane coefficients according to a certain rule, and then performing MDCT transform on the frequency domain coefficients several times to obtain a time-frequency plane coefficient. The method of recombining may include: firstly, the time-frequency plane coefficients are organized in the frequency direction, the coefficients in each frequency band are organized in the time direction, and then the organized coefficients are arranged in the order of the sub-window and the scale factor sub-band.

 The method of performing frequency-time mapping processing on frequency domain coefficients corresponds to the time-frequency mapping processing method in the encoding method, and may use inverse discrete cosine transform (IDCT), inverse discrete Fourier transform (IDFT), inverse modified discrete cosine transform ( IMDCT), inverse wavelet transform and other methods are completed.

 The inverse-corrected discrete cosine transform IMDCT is taken as an example to illustrate the frequency-time mapping process. The frequency-time mapping process consists of three steps: IMDCT transformation, time domain windowing, and time domain superposition.

 First, the IMDCT transform is performed on the pre-predicted spectrum or the inverse quantized spectrum to obtain the transformed time domain signal x,. The expression of the IMDCT transform is: χ,.„ , where η represents the sample number, and

'

0 ≤ n < N , indicating the number of time domain samples, the value is 2048, " ₀ = (Λ/2 + 1) / 2 ; / indicates the frame number; indicates the spectrum number.

Secondly, the time domain signal obtained by the IMDCT transform is windowed in the time domain. To satisfy the full reconstruction condition, the window function (n) must satisfy the following two conditions: w(2M -\ - n) = w(n) and w ² (n) + w ² (" + = 1. Typical window functions are S i ne windows, Ka i se r- Bes se l windows, and the like. The present invention employs a fixed window function whose window function is:

0. . . N -l ; represents the kth coefficient of the window function, with w (k) = w (2*^l-i); represents the number of samples of the encoded frame, and the value is 7^1 024. Alternatively, the double positive history transform can be used to modify the above-mentioned restrictions on the window function using a specific analysis filter and synthesis filter.

, 其中 表示帧序号, 表示样本 序号， 有 0≤"≤ 且 Ν的取值为 2 048。 Finally, the windowed time domain signal is superimposed to obtain a time domain audio signal. Specifically, the first Nil samples of the signal obtained after the windowing operation and the last Nil samples of the previous frame signal are overlapped and added to obtain Nil output time domain audio samples, timeSam _in = preSa _in +

, where the frame number is indicated, indicating the sample number, with 0 ≤ " ≤ and Ν is 2 048.

 2

 Figure 9 is a schematic illustration of a first embodiment of an encoding apparatus of the present invention. This embodiment adds a frequency domain linear prediction and vector quantization module 56 on the basis of FIG. 5, between the output of the multiresolution analysis module 53 and the input of the quantization and entropy coding module 5, and outputs the residual sequence to the quantized sum. The entropy encoding module 54 simultaneously outputs the quantized codeword index as side information to the bitstream multiplexing module 55.

 Since the frequency domain coefficients are obtained by multi-resolution analysis, the time-frequency coefficients with specific time-frequency plane divisions are obtained, so the frequency-domain linear prediction and vector quantization module 56 needs to perform linear prediction on the frequency domain coefficients in each time period. Multi-level vector quantization.

 The frequency signal output from the multi-resolution analysis module 53 is transmitted to the frequency domain linear prediction and vector quantization module 56. After multi-resolution analysis of the frequency domain coefficients, ^" the frequency domain coefficients on each time period are standard. Linear predictive analysis; if the predicted gain satisfies a given condition, the frequency domain coefficients are linearly predicted error filtered, and the obtained predictive coefficients are converted into line spectral frequency coefficients LSF (Line Spec t rum Frequency ), and then the best distortion is used. The metric search calculates the codeword index of each codebook, and transmits the code index as side information to the bit stream multiplexing module 55, and the residual sequence obtained through the prediction analysis is output to the quantity ^^ and the entropy coding module. 54.

 The frequency domain linear prediction and vector quantization module 56 is composed of a linear predictive analyzer, a linear predictive filter, a converter, and a vector quantizer. The frequency domain coefficient is input into the line 'f prediction analyzer for prediction analysis, and the prediction gain and the prediction coefficient are obtained. The frequency domain coefficients satisfying certain conditions are filtered out to the linear prediction filter to obtain a residual sequence; The difference sequence is directly output to the quantization and entropy coding mode 54, and the prediction coefficient is converted into a line spectral frequency coefficient LSF by the converter, and then the LSF parameter is sent to the vector quantizer for multi-level vector quantization, and the quantized signal is transmitted to In the bit stream multiplexing module 55.

Frequency domain linear prediction of the audio signal can effectively suppress the pre-echo and obtain a larger coding gain. Fake Let the real signal have its square Hi lber t envelope e y> as: _e (t) = F - ¹ fc( ) . C*( - ^, where

CO is a one-sided i-perform corresponding to the positive frequency component of the signal, ie the Hi lber t envelope of the signal is related to the autocorrelation function of the signal spectrum. The relationship between the power 谙 density function of the signal and the autocorrelation function of its time domain waveform is: PSD(f) , so the squared Hi lber t envelope of the signal in the time domain and the power of the signal in the frequency domain

The borrowing density function is mutually dual. From the above, the partial bandpass signal in each certain frequency range, if its Hi lbert envelope remains constant, then the autocorrelation of the adjacent value will remain constant, which means that the spectral coefficient sequence is relative to the frequency. It is a steady-state sequence so that the predictive coding technique can be used to process the speech values, and a common set of prediction coefficients is used to effectively represent the signal.

 The encoding method based on the encoding apparatus shown in FIG. 9 is basically the same as the encoding method based on the encoding apparatus shown in FIG. 5, except that the following steps are added: When the frequency domain coefficients are subjected to multi-resolution analysis, for each time period The frequency domain coefficient is subjected to standard linear prediction analysis to obtain prediction gain and prediction coefficient; whether the prediction gain exceeds the set threshold value, and if it exceeds, frequency domain linear prediction error filtering is performed on the frequency domain coefficient according to the prediction coefficient to obtain a residual The prediction coefficient is converted into a line pair frequency coefficient, and the line potential is subjected to multi-level vector quantization processing to obtain side information; the residual sequence is quantized and entropy encoded; if the prediction gain does not exceed the set threshold, The frequency domain coefficients are then quantized and entropy encoded. ·

 After the multi-resolution analysis of the frequency domain coefficients, the standard linear prediction analysis of the frequency domain coefficients on each time period is first performed, including calculating the autocorrelation matrix and recursively performing the Levinson-Durb in algorithm to obtain the prediction gain and the prediction coefficient. Then, it is judged whether the calculated prediction gain exceeds a preset threshold, and if it exceeds, linear prediction error filtering is performed on the frequency domain coefficients according to the prediction coefficient; otherwise, the frequency domain coefficients are not processed, and the next step is performed to quantize the frequency domain coefficients. And entropy coding.

 Linear prediction can be divided into forward prediction and backward prediction. Forward prediction refers to predicting the current value by using the value before a certain moment, while backward prediction refers to predicting the current value by using the value after a certain moment. In the following, the linear prediction error filtering is described as an example. The transfer function of the linear prediction error filter is = 1 -∑^.2-'', where α, represents the prediction coefficient, and ρ is the prediction order. After the time-frequency transform, the frequency domain coefficient YU is filtered to obtain the prediction error.

E (k) , also known as the residual sequence, satisfies the relationship between the two E k) = X(k) · A(z) = X(k) - J a _t X{k - i).

 =1

Thus, after linear prediction error filtering, the frequency domain coefficient X (k) of the time-frequency transform output can be represented by the residual sequence E ί3⁄4 and a set of prediction coefficients. And then converting the set of prediction coefficients a _t into a line 'i Pu frequency coefficient LSF and Perform multi-level vector quantization, vector quantization selects the best distortion metric (such as nearest neighbor criterion), and searches and calculates the codeword index of each codebook, so as to determine the code corresponding to the prediction coefficient, and use the codeword index as Side information output. At the same time, the residual sequence is quantized and entropy encoded. According to the linear prediction analysis coding principle, the dynamic range of the residual sequence of the spectral coefficients is smaller than the dynamic range of the original coefficients, so that fewer bits can be allocated in the quantization, or an improved coding gain can be obtained for the same number of bits. .

 10 is a schematic diagram of Embodiment 1 of a decoding apparatus. The decoding apparatus adds an inverse frequency domain linear prediction and vector quantization module 65 based on the decoding apparatus shown in FIG. 8, and the inverse frequency domain cropping prediction and vector quantization module 65 is located between the output of the inverse quantizer group 62 and the input of the multiresolution synthesis module 63, and the bitstream demultiplexing module 60 outputs inverse frequency domain linear predictive vector quantization control information thereto for inverse quantization f (residual The inverse quantization process and the inverse linear prediction filter-wave are performed to obtain a spectrum before prediction, and output to the multi-resolution synthesis module 63.

 In the encoder, frequency domain linear predictive vector quantization techniques are used to p-pre-echo and obtain a larger coding gain. Therefore, in the decoder, the inverse frequency domain linear predictive vector quantization control information output by the inverse quantization spectrum and bit stream demultiplexing module 60 is input to the inverse frequency domain linear prediction and vector quantization module 65 to recover the spectrum before the linear prediction.

 The inverse frequency domain linear prediction and vector quantization module 65 includes an inverse vector quantizer, an inverse transformer, and an inverse linear predictor, wherein the inverse vector quantizer is used to inverse quantize the codeword index to the J line pair frequency coefficient (LSF). The inverse converter is used to inversely convert the line spectral frequency coefficient (LSF) into a prediction coefficient; the inverse linear prediction filter is used to inversely filter the inverse quantized word by the prediction coefficient, obtain the spectrum before prediction, and output to the multi-resolution Synthesis module 63.

 The decoding method based on the decoding device shown in FIG. 10 is basically the same as the decoding method based on the decoding device shown in FIG. 8, except that the following steps are added: After obtaining the inverse quantization word, it is determined whether or not the inverse quantization spectrum is included in the control information. Inverse frequency domain linear predictive vector quantization information, if it is, then inverse vector quantization process is performed to obtain prediction coefficients, and linear prediction synthesis is performed on the inverse quantization spectrum according to the prediction coefficients to obtain a pre-test spectrum; Resolution synthesis.

 After obtaining the inverse quantized language, determining, according to the control information, whether the frame signal is subjected to frequency domain linear predictive vector quantization, and if so, obtaining a codeword index of the predicted coefficient vector quantization from the control information; and then obtaining the quantized according to the codeword index The line frequency coefficient (LSF) is used to calculate the prediction coefficient; then the inverse quantization spectrum is used for linear prediction synthesis to obtain the spectrum before prediction. The transfer function used in the linear prediction error filtering process Α ω^ '. Α(ζ) = 1 - ^ ,.ζ— '' , where: is pre

 (=1

Measuring coefficient; for predicting the order. Therefore, the residual sequence and the pre-predicted spectrum 0" are satisfied: Xik) = E(k) . -i- = EW + X a _t X(k― i). Thus, the residual sequence and the calculated prediction coefficient α, after frequency domain linear prediction synthesis, can obtain the pre-predictive language.

X(k), the pre-predicted spectrum 0 > is subjected to frequency-time mapping processing.

 If the control information indicates that the signal frame has not undergone frequency domain linear predictive vector quantization, the inverse frequency domain linear predictive vector quantization process is not performed, and the inverse quantization 镨 is directly subjected to frequency-time mapping processing.

Figure 11 is a block diagram showing the structure of a second embodiment of the encoding apparatus of the present invention. This embodiment adds a difference stereo (M/S) encoding module 57 on the basis of FIG. 5 between the output of the multiresolution analysis module 53 and the input of the quantization and entropy encoding module 54. For multi-channel signal, a psychoacoustic analysis module 51 in addition to a monaural audio signal is calculated masking threshold value, and further calculates a masking threshold difference channel output to quantization and entropy encoding module ^54. The and difference stereo encoding module 57 may also be located between the quantizer group and the encoder in the quantization and entropy encoding module 54.

 And the difference stereo encoding module 57 is to use the correlation between the two channels of the channel pair, and the frequency domain coefficient/residual sequence of the left and right channels is equivalent to the frequency domain coefficient/residual sequence of the difference channel, In this way, the effect of reducing the code rate and improving the coding efficiency is achieved, so that it is only applicable to multi-channel signals of the same signal type. If it is a mono signal or a multi-channel signal of which the signal type is not uniform, the difference and the stereo encoding process are not performed.

 The encoding method based on the encoding apparatus shown in FIG. 11 is the same as the encoding method based on the encoding apparatus shown in FIG. 5, and the difference is that the following steps are added: before the quantization and encoding processing of the frequency domain coefficients are performed, Whether the audio signal is a multi-channel signal, if it is a multi-channel signal, it is determined whether the signal types of the left and right channel signals are consistent, and if the signal types are the same, it is determined whether the scale factor bands corresponding to the two channels are satisfied and The difference stereo coding condition, if it is satisfied, is subjected to the sum and the difference stereo coding to obtain the frequency domain coefficients of the difference channel; if not, the difference and the stereo coding are not performed; if the mono signal or the signal type is inconsistent For multi-channel signals, the frequency domain coefficients are not processed.

 And the difference stereo coding can be applied before quantization and before entropy coding, that is, after quantization of the frequency domain coefficients, whether the audio signal is a multi-channel signal, and if it is a multi-channel signal, Then, it is judged whether the signal types of the left and right channel signals are consistent. If the signal types are the same, it is judged whether the scale factor bands corresponding to the two channels satisfy the difference and the stereo encoding conditions, and if they are satisfied, the stereo coding is performed. If not, the stereo encoding process is not performed; if it is a mono signal or a multi-channel signal with inconsistent signal types, the frequency domain coefficients are not processed and the differential stereo encoding process is performed.

There are many methods for judging whether the scale factor band can be performed and differential stereo coding. The judgment method adopted by the present invention is: by K-L transform. The specific judgment process is as follows: If the spectral coefficient of the left channel scale factor band is l (k), the spectral coefficient of the scale factor band corresponding to the right channel is r (k), and its correlation matrix

 1

 C„. =丄∑r(W * r(k); is the number of spectral lines of the scale factor band. The -Z transformation is performed on the correlation matrix C to obtain

 0, cos a - sm a

 RCR' = A where R

 0 λ... sin a cos a

旋转角度 a满足 tan(2iz) 就是和差立体声编码模式。 因此

The rotation angle a satisfies tan(2iz) and the difference stereo coding mode. therefore

When the absolute value of the rotation angle a deviates from ττ/4, such as 3 /16 < |^ < 5 /16 , the corresponding scale factor band can be subjected to and differential stereo coding. .

If the sum and difference stereo coding are applied before the quantization process, the frequency domain coefficients of the left and right channels in the scale factor band are replaced by the frequency domain coefficients of the linear transform and the difference channel:

Where Μ denotes the channel frequency domain coefficient; S denotes the difference channel frequency domain coefficient; L denotes the left channel frequency domain coefficient; R denotes the right channel frequency domain coefficient.

If the sum and difference stereo coding are applied after quantization, the quantized frequency domain coefficients of the left and right channels at the scale due to the thousands of bands are replaced by the frequency domain coefficients of the linear transform and the difference channel:

Where: / represents the quantized sum channel frequency domain coefficients; represents the quantized difference channel frequency domain coefficients; represents the quantized left channel frequency domain coefficients; A represents the quantized right channel frequency domain coefficients.

 By placing the sum and difference stereo coding after the quantization process, not only the correlation of the left and right channels can be effectively removed, but also since the quantization is performed, lossless coding can be achieved.

Figure 12 is a schematic diagram of a second embodiment of a decoding apparatus. The decoding apparatus adds a difference and difference stereo decoding module 66, based on the decoding apparatus shown in FIG. 8, between the output of the inverse quantizer group 62 and the input of the multi-dividend ratio synthesizing module 63, and receives the bit stream demultiplexing. The signal type analysis result and the difference stereo control signal output by the module 60 are used for The inverse quantized spectrum of the difference channel and the difference channel are converted into inverse quantization pans of the left and right channels according to the above control information.

 In the sum and difference stereo control signals, there is a flag to indicate whether the current channel pair needs to be and the difference stereo decoding. If necessary, there is also a flag on each scale factor band indicating whether the corresponding scale factor band needs and is poor. The stereo decoding, and difference stereo decoding module 66 determines whether inverse quantization 侮 and differential stereo decoding are required in certain scale factor bands based on the flag bits of the scale factor band. If the sum and the difference stereo coding are performed in the encoding device, the inverse quantized speech must be subjected to the difference stereo decoding in the decoding device.

 The sum and difference stereo decoding module 66 may also be located between the output of the demodulation module 61 and the input of the inverse quantizer group 62, and receive the sum and difference stereo control signals and signal type analysis outputs output by the bitstream demultiplexing module 60.

 The decoding method based on the decoding apparatus shown in FIG. 12 is basically the same as the decoding method based on the decoding apparatus shown in FIG. 8, except that the following steps are added: after the inverse quantization spectrum is obtained, if the signal type analysis result indicates that the letter type is consistent And determining whether it is necessary to perform inverse stereo decoding on the inverse quantized signal according to the difference stereo control signal; if necessary, determining whether the scale factor band requires and difference stereo decoding according to the flag bit on each scale factor band, ^. If necessary, the inverse quantized spectrum of the difference channel in the scale factor band is converted into the inverse quantized spectrum of the left and right channels, and then subjected to subsequent processing; if the signal type is inconsistent or does not need to perform and differential stereo decoding, then The quantized spectrum is processed without further processing.

 And the difference stereo decoding can also be performed after the entropy decoding process and before the inverse quantization process, that is, after the quantized value of the spectrum, if the signal type analysis result indicates that the signal types are consistent, it is determined according to the difference and the stereo control signal whether it is necessary to The quantized value of the spectrum is subjected to and the difference stereo decoding; if necessary, the flag bit on each scale factor band is used to determine whether the scale factor band requires and the difference stereo decoding, and if necessary, the sum of the scale factor bands The quantized value of the channel is converted into the quantized value of the spectrum of the left and right channels, and then subjected to subsequent processing; if the signal type is inconsistent or does not need to be subjected to and the difference stereo decoding, the quantized value of the spectrum is not processed, and subsequent processing is directly performed.

 If the sum and difference stereo decoding are after entropy decoding and before inverse quantization, then the frequency domain coefficients of the left and right channels in the scale factor band are obtained by the following operations through the frequency domain coefficients of the difference channel: , where: represents the quantized

和声道频域系数； S表示量化后的差声道频域系数； Z表示量化后的左声道频域系数； f表 示量化后的右声道频域系数。

And the channel frequency domain coefficient; S represents the quantized difference channel frequency domain coefficient; Z represents the quantized left channel frequency domain coefficient; f represents the quantized right channel frequency domain coefficient.

 If the inverse stereo decoding is after inverse quantization, the frequency coefficients of the left and right channels after inverse quantization in the subband are obtained according to the following matrix operation and the frequency domain coefficients of the difference channel, where: Table and channel frequency domain

Coefficient; s represents the difference channel frequency domain coefficient; 7 represents the left channel frequency domain coefficient; "represents the right channel frequency domain coefficient. Fig. 13 is a view showing the configuration of a third embodiment of the encoding apparatus of the present invention. This embodiment, on the basis of Fig. 9, adds a difference and stereo encoding module 57 between the output of the frequency domain linear prediction and vector quantization module 56 and the input of the quantization and entropy encoding module 54, the psychoacoustic analysis module 51 The masking values of the difference channels are output to a 4: quantization and entropy encoding module 54.

 The sum and difference stereo coding module 57 may also be located between the quantizer group and the scalar in the quantization and entropy coding module 54, and receive the signal type analysis result output by the psychoacoustic analysis module 51.

 In the present embodiment, the function and working principle of the and difference stereo coding module 57 are the same as those in FIG. 11, and details are not described herein again.

 The encoding method based on the encoding apparatus shown in FIG. 13 is basically the same as the encoding method based on the encoding apparatus shown in FIG. 9, except that the following steps are added: Before the quantization and entropy encoding processing of the frequency domain coefficients, the audio signal is judged. Whether it is a multi-channel signal, if it is a multi-channel signal, it is judged whether the signal types of the left and right channel signals are consistent. If the signal types are the same, it is judged whether the scale factor band satisfies the coding condition, and if it is satisfied, the scale is The factor band performs and the difference stereo encoding; if not, the sum and difference stereo encoding processing is not performed; if it is a mono signal or a multi-channel signal of which the signal type is inconsistent, the difference stereo encoding processing is not performed.

 And the difference stereo coding can be applied before quantization and before entropy coding, that is, after quantization of the frequency domain coefficients, whether the audio signal is a multi-channel signal, and if it is a multi-channel signal, And determining whether the signal types of the left and right channel signals are consistent. If the signal types are consistent, determining whether the scale factor band satisfies the coding condition, and if so, performing the difference stereo coding on the scale factor band; if not, not Performing and difference stereo encoding processing; if it is a mono signal or a multi-channel signal of which the signal type is inconsistent, the pairing and the difference stereo encoding processing are performed.

 Figure 14 is a block diagram showing a third embodiment of the decoding apparatus. The decoding apparatus adds a sum and difference stereo decoding module 66 on the basis of the decoding apparatus shown in Fig. 10, between the output of the inverse quantizer group 62 and the input of the inverse frequency domain linear prediction and vector quantization module 65, the bit stream The demultiplexing module 60 outputs and the difference stereo control signals thereto.

 The sum and difference stereo decoding module 66 may also be located between the output of the entropy decoding module 61 and the input of the inverse quantizer group 62, and receive the sum and difference stereo control signals output by the bit stream demultiplexing module 60.

 In this embodiment, the function and working principle of the and difference stereo decoding module 66 are the same as those in FIG. 10, and details are not described herein again.

The decoding method based on the decoding device shown in FIG. 14 is basically the same as the decoding method based on the decoding device shown in FIG. 10, and the difference is that the following steps are added: after the inverse quantization pan is obtained, if the signal type analysis result indicates that the signal type is consistent, Then, according to the difference stereo control signal, it is judged whether it is necessary to perform inverse stereo quantization and differential stereo decoding; if necessary, Determining whether the scale factor band requires and difference stereo decoding according to the flag bit on each scale factor band, if necessary, converting the inverse quantization spectrum of the difference channel in the scale factor band into the inverse quantization spectrum of the left and right aisles Then, the subsequent processing is performed; if the signal types are inconsistent or the difference stereo decoding is not required, the inverse quantization spectrum is not processed, and the subsequent processing is directly performed.

 And the difference stereo decoding can also be performed before the inverse quantization process, that is: after obtaining the quantized value of 镨, if the signal type analysis result indicates that the signal type is consistent, it is determined according to the difference and the stereo control ft number whether the quantized value of the spectrum needs to be performed. And difference stereo decoding; if necessary, determining whether the scale factor band requires and difference stereo decoding according to the flag bit of each scale factor band, and if necessary, converting the quantized value of the spectrum of the scale factor band and the difference channel The quantized values of the spectrum of the left and right channels are subjected to subsequent processing; if the signal type is consistent or no difference stereo decoding is required, the quantized values of the spectrum are not processed, and the subsequent processing is directly performed'

 Fig. 15 shows a schematic representation of a fourth embodiment of the encoding device of the present invention. In this embodiment, based on the encoding apparatus shown in FIG. 5, a resampling module 590 and a band extending module 591 are added, wherein the resampling module 590 resamples the input audio signal, changes the sampling rate of the audio signal, and then changes The sampling rate audio signal is output to the signal property analysis module 50; the band extension module 591 is configured to analyze the input audio signal over the entire frequency band, and extract the spectral envelope of the high frequency portion and its relationship with the low frequency portion. , and 3⁄4" out to the bit stream multiplexing module 55.

The resampling module 590 is configured to resample the input audio signal, and the t sampling includes both upsampling and downsampling. The following sampling is used as an example to illustrate resampling. In the present embodiment, the resampling block 590 includes a low pass filter and a down sampler, wherein the low pass filter is used to limit the frequency band of the audio signal, eliminating aliasing that may be caused by the sample. The input audio signal is low-pass filtered and downsampled. Assume that the input audio signal is "s (n), the impulse response is 1φ). The output of the low-pass filter is v (nj, then there is 1) = ^(» (« - for M times The sampled sequence is x (n), then x( ). Thus, the re-sampled audio signal x (n)

The sampling rate is reduced by a factor of 1 compared to the original input audio signal.

 After the original input audio signal is input to the band extension module 591, the analysis is performed on the entire frequency band, and the spectral envelope of the high-frequency portion and its characteristics related to the low-frequency portion are extracted, and output as the frequency extension control information to the bit stream multiplexing. Module 55.

The basic principle of frequency band expansion is: For most audio signals, the characteristics of the high frequency part have a strong correlation with the characteristics of the low frequency part, so the high frequency part of the audio signal can be effectively reconstructed by the low frequency part, thus, The high frequency portion of the audio signal may not be transmitted. In order to ensure that the high frequency part can be reconstructed correctly, only the compressed audio stream is compressed. It is sufficient to transmit a small number of band extension control signals.

 The band expansion module 591 includes a parameter extraction module and a language envelope extraction module, and the input signal enters the parameter extraction module to extract parameters indicating spectral characteristics of the input signal in different time-frequency regions, and then in the spectral envelope extraction module, at a certain time The frequency resolution estimates the spectral envelope of the high frequency portion of the signal. To ensure that the time-frequency resolution is best suited to the characteristics of the current input signal, the time-frequency resolution of the spectral envelope is freely selectable. The parameters of the input signal spectrum characteristics and the speech envelope of the high frequency portion are output as a band extension control signal to the bit stream multiplexing module 55 for multiplexing.

 The bitstream multiplexing module 55 receives the code stream including the common scale factor, the scale factor coded value, the codebook sequence number coded value, and the lossless coded quantized spectrum or the coded value of the codeword index and the band extension output by the quantization and entropy coding module 54. After the band extension control signal output by the module 591 is multiplexed, a compressed audio data stream is obtained.

 The encoding method based on the encoding apparatus shown in FIG. 15 specifically includes: analyzing an input audio signal over the entire frequency band, extracting a high spectral envelope and a signal spectral characteristic parameter as a frequency band extension control signal; resampling the input audio signal and a signal Type analysis; calculating the signal-to-mask ratio of the resampled signal; performing time-frequency mapping on the resampled signal to obtain a frequency domain coefficient of the audio signal; performing quantization and entropy coding on the frequency domain coefficient; and extending the frequency band control signal and coding The audio stream is multiplexed to obtain a compressed audio stream. The resampling process includes two steps: limiting the frequency band of the audio signal; and down-sampling the audio signal of the limited band.

 FIG. 16 is a schematic structural diagram of Embodiment 4 of the decoding apparatus. The embodiment is based on the decoding apparatus shown in FIG. 8. A band extension module 68 is added, and the band extension control information and frequency output by the bit stream demultiplexing module 60 are received. - The time domain audio signal of the low frequency band output by the time mapping module 64 is reconstructed by frequency shifting and high frequency adjustment to output a wideband audio signal.

 The decoding method based on the decoding apparatus shown in FIG. 16 is basically the same as the decoding method based on the decoding apparatus shown in FIG. 8, except that the following steps are added: After obtaining the time domain audio signal, the control information and the time domain audio are expanded according to the frequency band. The signal reconstructs the high frequency portion of the audio signal to obtain a wideband audio signal.

 Figures 17, 1 and 21 are fifth to seventh embodiments of the encoding apparatus, based on the encoding apparatus shown in Figures 11, 9, and 13, respectively, with the addition of the resampling module 590 and the band extending module 591. The connection relationship, functions, and principles of the two modules are the same as those in FIG. 15 and will not be described here.

 18, 20 and 22 are fifth to seventh embodiments of the decoding apparatus, respectively, based on the decoding apparatus shown in Figs. 12, 10 and 14, respectively, a band extension module 68 is added, and a bit stream is received. The band extension control information output by the demultiplexing module 60 and the time domain audio signal of the low frequency band output by the frequency-time mapping module 64 reconstruct the high frequency signal portion by spectrum shifting and high frequency adjustment, and output a wideband audio signal.

In the seven embodiments of the foregoing encoding apparatus, the gain control module may further be included, and the signal property analysis module is received. 50 output audio signal, controlling the dynamic range of the fast-changing type signal, eliminating pre-echo in the audio, and outputting the output to the time-frequency mapping module 52 and the psychoacoustic analysis module 51, and simultaneously adjusting the gain to the bit stream Use module 55.

 The gain control module only controls the fast-change type signal according to the signal type of the audio signal, and does not process and directly output the slowly-changing type signal. For fast-change type signals, the gain control module adjusts the time-domain energy envelope of the signal, and increases the gain value of the signal before the fast-changing point, so that the time-domain signal degrees before and after the fast-changing point are relatively close; then the time domain energy is adjusted. The time domain signal of the envelope is output to the time-frequency mapping module 52, and the increased adjustment amount is output to the bit stream multiplexing module 55.

 The encoding method based on the encoding device is basically the same as the encoding method based on the above encoding device, with the difference that the following steps are added: Gain control is performed on the signal subjected to signal type analysis.

 In the seven embodiments of the foregoing encoding apparatus, an inverse gain control module fe may be further included, and after receiving the output of the frequency-time mapping module 64, the signal type and the gain adjustment amount output by the bit stream demultiplexing module 60 are received. Information, used to adjust the gain of the time domain signal, to control the pre-echo. After the inverse gain control module is connected to the reconstructed time domain signal outputted by the frequency-time mapping module 64, the fast-changing type signal is controlled, and the slow-changing type is not processed. For the fast-change type signal, the inverse gain control module 4N3⁄4 gain adjustment amount information adjusts the energy envelope of the reconstruction or the signal, reduces the amplitude value of the signal before the fast-changing point, and returns the energy envelope to the original front low and high State, such that the magnitude of the quantization noise before the fast change point is correspondingly reduced with the amplitude value of the signal, thereby controlling the pre-return.

 The decoding method based on the decoding device is basically the same as the decoding method based on the above decoding device, with the difference that the following steps are added: inverse gain control is performed on the reconstructed time domain signal.

 It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and the present invention will be described in detail with reference to the preferred embodiments, and those skilled in the art should understand that the technical solutions of the present invention can be Modifications or equivalents are intended to be included within the scope of the appended claims.

Claims

Rights request An enhanced audio coding apparatus, comprising a psychoacoustic analysis module, a time-frequency mapping module, a quantization and entropy coding module, and a bitstream multiplexing module, further comprising a signal property analysis module and a multi-resolution analysis module; The signal property analysis module is configured to perform type analysis on the input audio signal, and output the signal to the psychoacoustic analysis module and the time-frequency mapping module, and output the type analysis result of the audio signal to the lost stream. Multiplexing module  The psychoacoustic analysis module is configured to calculate a masking threshold and a signal mask ratio of the audio signal, and output to the quantization and entropy encoding module;  The time-frequency mapping module is configured to convert a time domain audio signal into a frequency domain coefficient, and output the data to: a resolution analysis module;  The multi-resolution analysis module is configured to perform multi-resolution analysis on the frequency domain coefficients of the fast-changing type signal according to the signal class analysis result output by the signal property analysis module, and output the same to the quantization and entropy coding module;  The quantization and entropy coding module is configured to quantize and entropy encode frequency domain coefficients under control of a mask ratio output by the psychoacoustic analysis module, and output the same to the bit stream multiplexing module;  The bit stream multiplexing module is configured to multiplex the received data to form an audio code u.  2. The enhanced audio encoding apparatus according to claim 1, wherein the multi-resolution analysis module comprises a frequency domain coefficient transform module and a recombination module, wherein the frequency domain coefficient transform module is configured to transform a frequency domain coefficient The time-frequency plane coefficient is used; the recombination module is configured to recombine the time-frequency plane coefficients according to a certain rule; wherein the frequency domain coefficient transform module is a frequency domain wavelet transform filter bank or a frequency domain MDCT transform filter bank.  3. The enhanced audio encoding apparatus according to claim 1, further comprising a frequency domain linear prediction and vector quantization module, located at an output of said multiresolution analysis module and said input of said quantized and entropy encoded blocks The frequency domain linear prediction and vector quantization module is specifically composed of a linear prediction analyzer, a linear prediction filter, a converter, and a vector quantizer;  The linear predictive analyzer is configured to perform predictive analysis on frequency domain coefficients, obtain prediction gain and page measurement coefficients, and output frequency domain coefficients satisfying a certain condition to the linear prediction filter; The domain coefficients are directly output to the quantization and entropy coding module;  The linear prediction filter is configured to filter the frequency domain coefficients to obtain a residual sequence, and the ridge residual sequence is output to the quantization and entropy coding module, and the prediction coefficients are output to the converter; The converter is configured to convert a prediction coefficient into a line spectrum pair frequency coefficient; The vector quantizer is configured to perform multi-level vector quantization on the line error pair frequency coefficient, and the quantized related side information is transmitted to the bit stream multiplexing module.  The enhanced audio encoding apparatus according to any one of claims 1 to 3, further comprising a sum and difference stereo encoding module, the output of the frequency domain linear prediction and vector quantization module, and the quantization and entropy coding. Between the inputs of the module, or between the quantizer group and the encoder in the quantization and entropy coding module; the signal property analysis module outputs a signal type analysis result thereto; the sum and difference stereo coding module is used to The residual sequence/frequency domain coefficients of the left and right channels are converted to residual sequence/frequency domain coefficients of the difference channel.  The enhanced audio encoding apparatus according to any one of claims 1 to 4, further comprising a resampling module and a band extending module;  The resampling module is configured to resample the input audio signal, change the sampling rate of the audio signal, and output the audio signal after changing the sampling rate to the psychoacoustic analysis module and the signal property analysis module; a low pass filter and a down sampler; wherein the low pass filter is configured to limit a frequency band of the audio signal, and the down sampler is configured to downsample the audio signal of the limited frequency band to reduce a sampling rate of the signal;  The frequency band expansion module is configured to analyze the input audio signal over the entire frequency band, extract a spectral envelope of the high frequency portion, and characterize the correlation between the low and high frequency chirps, and output the data to the bit stream multiplexing The module includes: a parameter extraction module and a spectral envelope extraction module; the parameter extraction module is configured to extract a parameter indicating an input signal characteristic of the input signal in different time-frequency regions; The time-frequency resolution estimates the spectral envelope of the high-frequency portion of the signal, and then outputs the parameters of the spectral characteristics of the input signal and the spectral envelope of the high-frequency portion to the bitstream multiplexing module.  6. An enhanced audio coding method, comprising the steps of:  Step 1: Perform type analysis on the input audio signal, and use the signal type analysis result as part of the multiplexing information. Step 2: Perform time-frequency mapping on the type-analyzed signal to obtain a frequency domain coefficient of the audio signal; meanwhile, calculate an audio signal Letter cover ratio;  Step 3: If it is a fast variable type signal, perform multi-resolution analysis on the frequency domain coefficient; if it is not a fast change type signal, go to step 4;  Step 4: Quantizing and entropy coding the frequency domain coefficients under the control of the mask ratio;  Step 5: multiplexing the encoded audio signal to obtain a compressed audio code stream. The enhanced audio coding method according to claim 6, wherein the quantization in the fourth step is scalar quantization, and specifically includes: performing nonlinear compression on frequency domain coefficients in all scale factor bands; and using each subband The scale factor quantizes the frequency domain coefficients of the subband to obtain a quantized spectrum expressed by an integer; selects one scale factor in each frame signal as a common scale factor; and other scale factors are differentially processed with the previous scale factor; The entropy coding comprises: entropy coding the quantized spectrum and the differentially processed scale factor, obtaining a codebook sequence "?", a scale factor coded value, and a quantized "i" lossless coded value; entropy coding the codebook sequence number, Get the code book serial number code H.  The enhanced audio coding method according to claim 6, wherein the step three-multi-rate analysis comprises: performing MDCT transform on the frequency domain coefficients to obtain a time-frequency plane coefficient; and determining, according to the time-frequency plane coefficient, The reorganization method includes: firstly, the time-frequency plane coefficients are organized in the frequency direction, the coefficients in each frequency are organized in the time direction, and then the organized coefficients are arranged in the order of the sub-window and the scale factor band. .  The enhanced audio coding method according to any one of claims 6-8, characterized in that, between the step-3 and the step 4, the method further comprises: performing standard linear prediction analysis on the frequency domain coefficients to obtain a prediction gain. And a prediction coefficient; determining whether the prediction gain exceeds a set value; if it exceeds, performing frequency linear prediction error filtering on the frequency domain coefficient according to the prediction coefficient, obtaining a linear prediction residual sequence of the frequency domain coefficient; converting the prediction coefficient into a line Word pair frequency coefficient, and multi-level vector quantization processing on the frequency coefficient to obtain side information; quantize and entropy encode the residual sequence; if the prediction gain does not exceed the set value, then the frequency domain The coefficients are quantized and entropy encoded.  The enhanced audio encoding method according to any one of claims 6-9, wherein the step 4 further comprises: quantizing the frequency domain coefficients; determining whether the audio signal is a multi-channel signal, if it is multi-channel The signal determines whether the signal types of the left and right channel signals are consistent. If the signal types are the same, it is determined whether the scale factor bands corresponding to the two channels satisfy the difference and the stereo coding conditions, and if satisfied, the scale factor band is The spectral system in the middle and the differential stereo coding obtain the frequency domain coefficients of the difference channel; if not, the lineage in the scale factor band does not perform the difference stereo coding; if it is a mono signal or the signal type is inconsistent The multi-channel signal, the frequency coefficient is not processed; the frequency domain coefficients are entropy encoded;  The method for judging whether the scale factor band satisfies the coding condition is: K-L transformation, specifically: calculating a correlation matrix of spectral coefficients of left and right channel scale factor bands; performing KL transformation on the correlation matrix; if the rotation angle α is absolutely straight Deviation; when r/4 is small, such as 3 /16 < |^ < 5 /16, the corresponding scale factor band can be subjected to and differential stereo coding; the sum and difference stereo coding is: , where: represents the quantized harmony Channel frequency domain coefficient;

The quantized difference channel frequency domain coefficient is displayed; the quantized left channel frequency domain coefficient is represented; and the quantized right sound frequency domain coefficient is represented.  The enhanced audio encoding method according to any one of claims 6 to 10, characterized in that, before the step one, further comprising a repeating step and a frequency band expanding step; In the repeating step, the input audio signal is re-sampled to change the sampling rate of the audio signal; The frequency band expansion step analyzes the input audio signal over the entire frequency band, extracting its high spectral envelope and signal spectral characteristic parameters as part of signal multiplexing.  12. An enhanced audio decoding apparatus, comprising: a bit stream demultiplexing module, an entropy decoding module, an inverse quantizer group, and a frequency-time mapping module, further comprising: a multi-resolution synthesis module;  The bitstream demultiplexing module is configured to demultiplex the compressed audio data stream, and output corresponding data signals and control signals to the entropy decoding module and the multi-resolution synthesis module;  The entropy decoding module is configured to perform decoding processing on the foregoing signal, recover the wrong quantized value, and output the result to the inverse quantizer group;  The inverse quantizer group is configured to reconstruct an inverse quantized language and output to the multi-resolution synthesis module;  The multi-resolution synthesis module is configured to perform multi-resolution synthesis on the inverse quantization spectrum and output to the frequency-time mapping module;  The frequency-time mapping module is configured to perform frequency-time mapping on the spectral coefficients and output a time domain audio signal.  The enhanced audio decoding device according to claim 12, wherein the multi-resolution synthesis module comprises: a coefficient recombination module and a coefficient transformation module; the coefficient transformation module is a frequency domain inverse wavelet transform filter bank or Frequency domain inverse modified discrete cosine transform filter bank.  14. The enhanced audio decoding apparatus according to claim 12 or 13, further comprising an inverse frequency domain linear prediction and vector quantization module, the output of the inverse quantizer group and the multi-resolution synthesis module The inverse frequency domain linear prediction and vector quantization module specifically includes an inverse vector quantizer, an inverse transformer, and an inverse linear prediction filter; the inverse vector quantizer is used for inverse quantization of the codeword index to obtain a line language The inverse frequency converter is configured to inversely convert the line spectrum to the frequency coefficient into a prediction coefficient; and the inverse linear prediction filter is configured to inversely filter the inverse quantization spectrum according to the prediction coefficient to obtain a spectrum before prediction.  The enhanced audio decoding device according to any one of claims 12-14, further comprising a sum and difference stereo decoding module, located after the inverse quantizer group or at an output of the entropy decoding module Between the inputs of the inverse quantizer group, receiving the sum and difference stereo control signals output by the bit stream demultiplexing module, for converting the quantized values of the inverse quantized spectrum/spectrum of the difference channel and the differential channel according to the sum and difference stereo control information The quantized value of the inverse quantized/spectrum of the left and right channels.  16. An enhanced audio decoding method, comprising the steps of:  Step 1: Demultiplexing the compressed audio data stream to obtain data information and control information;  Step 2: Entropy decoding the above information to obtain a quantized value of the spectrum; Step 3: performing inverse quantization processing on the quantized value of the spectrum to obtain an inverse quantization spectrum; Step 4: performing multi-resolution synthesis on the inverse quantization spectrum;  Step 5: Perform frequency-time mapping to obtain a time domain audio signal.  The enhanced audio decoding method according to claim 16, wherein the step four-resolution integration step is specifically: arranging the inverse quantized coefficients according to the order of the sub-window and the scale factor band, and then following the frequency order. The recombination is performed, and then multiple inverse modified cosine transforms are performed on the recombined coefficients to obtain an inverse quantized spectrum before multiresolution analysis.  The enhanced audio decoding method according to claim 16, wherein the step 5 may further comprise: performing inverse modified discrete cosine transform to obtain a transformed time domain signal; and transforming the time domain signal at time The domain performs windowing processing; superimposing the windowed time domain signal to obtain a time domain audio signal; wherein the window function in the windowing processing is:  ^(M=cos(/7i/2*((i+0.5)/ -0.94*8ίη(2*/7// *(1+0. )) I {2*pi))) , where ir = 0 ... ^l; ίΧ ^ denotes the kth coefficient of the window function, with w(k)=w(2*7^1_ ); represents the number of samples of the encoded frame.  The enhanced audio decoding method according to claim 17 or 18, further comprising: determining whether the inverse quantization spectrum is included in the control information needs to undergo inverse frequency domain linearity between the step 3 and the fourth step The information of the predicted vector quantization, if included, is subjected to inverse vector quantization processing to obtain a prediction coefficient, and the prediction coefficient is used for linear prediction synthesis of the inverse quantization spectrum to obtain a spectrum before prediction; and the spectrum before prediction is subjected to frequency-time mapping; The inverse vector quantization process further includes: obtaining a codeword index of the prediction coefficient vector quantization from the control information; and obtaining a quantized line frequency coefficient according to the codeword index, and calculating the prediction coefficient.  The enhanced audio decoding method according to any one of claims 16 to 19, wherein, between the second step and the third step, the method further comprises: if the signal type analysis result indicates that the signal types are consistent, the basis and the difference are The stereo control signal determines whether inverse quantization and differential stereo decoding are required; if necessary, determining whether the scale factor band requires and difference stereo decoding according to the flag bit on each scale factor band, and if necessary, the scale factor The inverse quantization of the band and the difference channel is converted into the inverse quantization spectrum of the left and right channels, and the process proceeds to step 3; if the signal type is inconsistent or does not need to perform the difference and the stereo decoding, the inverse quantization spectrum is not processed, Go to step 3; wherein the sum and difference stereo decoding are: , where: represents the quantized sum channel frequency domain

The number represents the quantized difference channel frequency domain coefficient; represents the quantized left channel frequency domain coefficient; represents the quantized right channel frequency domain coefficient.