CN100477531C - Encoding method for compression encoding of multi-channel digital audio signal - Google Patents

Abstract

This invention relates to a device and a method for compressing code and decode to multiple channel digital audio signals, in which, the code device includes: a frame length selector, a sub-band resolve and filter set, a transient detector, a proportion factor evaluator, a bit distributor used in distributing the bit resources decided by the target code rate to the sub-band sections adaptively, a sub-quantizer and a multiple multiplexer. The compression rate of the device and the method is high and the encoded audio signals can be edited synchronously with video signals and endure over ten times of tandem encode so as to meet the needs of distribution and transmission of programs and reduce the processes and devices greatly.

Description

Encoding method for compression encoding of multi-channel digital audio signal

This application is a divisional application submitted on August 21, 2002, with application number 02130245.6.

technical field

The present invention relates to a digital audio signal encoding/decoding device and its method, more precisely, to a multi-channel digital audio signal compression encoding/decoding device and method.

Background technique

Multi-channel (including stereo) digital audio compression coding technology has been widely used in VCD, SVCD, DVD, satellite TV, digital TV, and Internet (Internet) and other fields. The main problem it needs to solve is that the code rate for expressing multi-channel digital audio signals is very high, but the channel capacity that can be used to propagate or store it is very limited. For example, using PCM to express 5.1-channel surround sound with 48kHz sampling rate and 24 bits per sample requires a code rate of 6912kbps (kilobits per second), and applications with limited channel capacity such as digital TV can be allocated to digital The code rate of the audio signal is generally 384kbps, and the code rates that can be assigned to the digital audio signal are generally 384kbps, 768kbps, and 1536kbps even for applications such as DVD with relatively loose channel capacity. Here, digital audio compression coding technology needs to provide up to 18 times the compression ratio.

The research and development of digital audio compression coding technology can be traced back to the early 1970s. After 30 years of development, the currently widely used technical framework has basically been finalized as follows: frame length selector, frequency or subband resolver, transient detector, linear predictor, bit allocator, quantizer, entropy encoder, and multiplexers.

For example, MPEG 2AAC [Reference 1] and MPEG 4AAC [Reference 2] techniques divide the input audio PCM signal stream into 1024 samples per frame, and then perform transient detection on each frame signal. If no transient response is found in the samples of this frame, then (optionally) perform long-term linear prediction, then perform frequency decomposition of 1024 sub-bands, and then (optionally) perform short-term linear prediction on each sub-generation signal. If a transient response is found, the 1024 samples of this frame are further divided into 8 subframes, each frame has 128 samples, and then the frequency decomposition of 128 subbands is performed, and the positions of those subframes where the transient response is located are sent to multiplexer. Then, a global bit allocator based on the human auditory model is implemented, nonlinear scalar quantization is performed on the subband signal, and Huffman (Huffman) encoding is performed on the quantization index. Finally, the multiplexer packs the auxiliary information generated by the above steps and the Huffman code expressing each sub-band sample into a complete frame-based compressed code stream. The advantage of AAC is its high compression efficiency. However, the encoder and decoder are complicated, and the sound quality of the decoded audio signal is not completely transparent.

For another example, the frame length selector of DTS's multi-channel audio encoder [References 3, 4 and 5] can select a frame length from 256, 512, 1024, 2048, and 4096 according to the sampling rate and code rate, and press This frame length divides the incoming audio PCM signal stream into frames. Then do frequency decomposition of 32 sub-bands, and then do sub-band encoding for each sub-band signal. Subband coding includes transient detection, linear prediction, global bit allocator based on human auditory model, scalar/vector quantization, and Huffman coding. Finally, the multiplexer packs the auxiliary information generated by the above steps and the quantization index or Huffman code expressing each sub-band sample into a complete frame-based compressed code stream. The advantage of DTS is that the sound quality of the decoded audio signal is good, and it is considered completely transparent by many people at high bit rates (such as 1536kbps). But it doesn't compress efficiently.

With the commercial broadcast of digital TV in Europe and North America in recent years, the distribution of multi-channel audio programs as TV audio has become an urgent problem to be solved. One of the major issues involved here is that current television and recording studio facilities only support stereo. Changing them to multi-channel means replacing almost all equipment related to audio. This problem can be avoided by compressing multi-channel programs to a bit rate that stereo can support. Compressing multi-channel programs is also beneficial to the transmission and sharing of programs between various TV stations and recording studios. But the compressed audio stream introduces a frame structure. If the length of the audio frame is not equal to that of the video frame, when the video stream is clipped on the frame boundary, it will be cut into the audio frame, thereby destroying the structure of the audio frame and causing a decoder error. In addition, in the production and distribution process of multi-channel programs, it is often necessary to perform multiple encoding and decoding (Tandem Coding) operations. This requires that the compression technology must be able to withstand at least ten times of column coding without hearing distortion.

Dolby E is an audio compression coding technology designed for the above applications [Ref 6]. Its frame length is fixed at 1792, but it uses the method of sampling rate conversion to make the time occupied by a frame of Dolby E data stream equal to the frame length of various common video frame frequencies (NTSC, PAL, and film). To achieve the purpose of editing synchronously with them. At the same time, it uses a high bit rate to ensure that it can withstand more than ten times of tandem encoding without hearing distortion. However, Dolby E's compression efficiency is not high, and it is not suitable as a compression technology for transmitting multi-channel programs to end users (such as televisions). Therefore, after the TV station uses Dolby E to produce a good program, it has to be decoded into PCM, and then encoded into a code stream of high-compression-efficiency encoding technology such as AC-3 [Reference 7] or MPEG before it can be transmitted. Figure 1 shows the process of distributing and transmitting audio programs by TV stations using Dolby E as the compression coding technology for program distribution and AC-3 as the compression coding technology for program transmission. It can be seen from this that there are several difficulties in this TV station method: 1) The distortion of the audio signal is large: the sampling rate conversion of Dolby E itself introduces distortion, and the transfer of the code stream from the Dolby E format to the AC-3 format Coding (Transcoding) introduces new distortion. 2) The programs that have been transmitted are difficult to reuse: as shown in Figure 1, if the programs that have been transmitted are to be reused, they must be decoded into PCM and encoded by Dolby E before they can be switched with other programs (such as advertisements, etc.) . It has to go through the transfer encoding process from Dolby E decoding to AC-3 encoding again when transmitting. Since the code rate for the final transmission is generally not high, the sound quality of the programs that have been transmitted after the above series of transfer encoding (AC-3 decoding -> Dolby E encoding -> Dolby E decoding -> AC-3 encoding) is very low. Difficult or impossible to guarantee. 3) Since the sampling rate of Dolby E's input and output has no simple frequency multiplication relationship, its encoding and decoding are very complicated and expensive.

Contents of the invention

The first aspect of the present invention provides a high-efficiency and high-fidelity encoder for compressing and encoding multi-channel (including mono) audio signals and its encoding method. When the audio signal is used as the accompanying sound of the video signal, the method can not only meet the requirements of distributing multi-channel digital audio programs, but also meet the requirements of transmitting multi-channel digital audio programs with medium and low bit rates (high compression efficiency). That is, it implements the combined functions of Dolby E and other transmission compression coding technologies such as AC-3.

The encoder in this aspect of the present invention includes: 1) frame length selector, for selecting the audio frame according to the sampling rate of the audio signal, the code rate, and the video frame frequency (when the multi-channel audio signal is used as the accompanying sound of the video signal) long; 2) sub-band decomposition filter bank, used to decompose the input audio signal frame by frame into a plurality of sub-band signals; 3) transient detector, used to divide the input sub-band signal into transient segment and steady state segment 4) bit allocator, for distributing bit resources determined by the target code rate to each sub-band segment; 5) sub-band quantizer, for quantizing the sub-band signal in units of segments; 6 ) multiplexer (MUX), used to multiplex and package the sub-band quantization indices and related auxiliary information into a complete code stream in units of frames.

The encoding method of the present invention in this aspect comprises: 1) according to the sampling rate of audio signal, code rate, and video frame frequency (when multi-channel audio signal is used as the accompanying sound of video signal) select audio frame length; 2) by subband The analysis filter bank decomposes the input audio signal frame by frame into multiple sub-band signals; 3) divides each sub-band signal into a transient segment and a steady-state segment; 4) allocates bit resources determined by the target code rate to each Sub-band segment; 5) Quantize the sub-band signal in units of segments; 6) Multiplex and package the quantization index of the sub-band signal and related auxiliary information into a complete code stream in units of frames.

The second aspect of the present invention provides a decoder for decoding an audio stream encoded by the above encoder according to the above encoding method and a decoding method thereof. Wherein the decoder includes: 1) demultiplexer (DEMUX), which is used to demultiplex the quantization index of the sub-band signal and related parameters such as audio frame length, sub-band segment boundary, and bit Auxiliary information such as allocation; 2) subband inverse quantizer, used to reconstruct subband signals from quantization indices of subband signals in units of segments according to related auxiliary information; 3) subband synthesis filter bank, used to reconstruct subband signals audio signal.

The decoding method of the present invention in this aspect includes: 1) demultiplexing the quantization index of the sub-band signal and related auxiliary information from the code stream of the above-mentioned audio encoding; 3) Reconstruct the audio signal from the subband signal with the subband synthesis filter bank.

When the multi-channel audio signal is used as the accompanying sound of the video signal, the audio frame length selected by the present invention has a simple multiple relationship with the video frame length: or the video frame length is a positive integer multiple of the audio frame length, or the audio frame length is the video frame length Integer multiples of long. In this way, the code stream generated by the present invention can be clipped synchronously with the video program.

Furthermore, when the multi-channel audio signal is used as the accompanying sound of the video signal, the number of sub-bands of the sub-band decomposition filter bank selected by the present invention is a common factor of different audio frame lengths corresponding to different video frame rates (PAL and movies) . In this way, when the frame rate of the video program changes, the encoder and decoder of the present invention only need to change the multiple of the audio frame length and the common factor, rather than the number of subbands, to maintain the ratio between the audio frame length and the video frame length. The aforesaid multiple relationship is to maintain the synchronous editing ability with the video program. Since the number of sub-bands does not change, the delay lines of the encoder and decoder do not need to be changed, so the encoder and decoder can adapt to the change of the frame rate of the video program without resetting.

The above limitations on the number of subbands and the frame length do not prevent the present invention from having a very short frame length to provide a low-delay working mode.

The filters used by the sub-band decomposer and synthesizer of the present invention are relatively long to ensure smooth transition of the decoded audio signal after editing the data stream compressed and coded by the present invention at the frame boundary.

The transient detector of the encoding system of the present invention splits the input subband signal into transient and stationary segments. It has a very high time resolution to reduce the forward echo (Pre-echo) effect often encountered in audio compression coding. Each subsequent component of the present invention processes the input audio signal in units of segments.

The present invention utilizes the correlation of audio signals across channels on the same subband (frequency) to make long-term and short-term predictions. In this way, the periodicity of the audio signal and the correlation between channels can be fully utilized, and a low-order predictor can be used to reduce the amount of computation.

The bit allocator of the present invention makes very limited use of the human auditory model for the purpose of simplifying bit allocation. In this way, the computational complexity of bit allocation can be greatly reduced, and only one parameter can be used to express the result of bit allocation. After receiving this parameter, the decoder can very simply reconstruct exactly the bit allocation used by the encoder from it. This saves the bit resource used by other techniques to transmit the bit allocation. Because of this savings, the improvement in coding efficiency with this approach is likely to be comparable to that achieved with a very complex model of the human ear.

In terms of quantization, the strategy adopted in this patent is to use vector quantization when there are few quantization levels, and use scalar quantization when there are many quantization levels, so as to achieve the purpose of optimizing compression efficiency, decoding complexity and the demand for memory units.

Description of drawings

Figure 1: Distribution and transmission scheme of digital TV programs using Dolby's audio compression technology.

Figure 2: The encoder of the first embodiment.

Figure 3: Schematic diagram of transient state detection in the present invention.

Figure 4: Encoder of the second embodiment.

Figure 5: Threshold in Quite of the human ear in a quiet environment.

Figure 6: Encoder of a third embodiment.

Figure 7: Encoder of a fourth embodiment.

Figure 8: Quantization of linear prediction and prediction error signals for this channel.

Figure 9: Cross-channel linear prediction and quantization of the prediction error signal.

Figure 10: Use of local channel and cross-channel linear prediction, and quantization of the prediction error signal.

Figure 11: Encoding flowchart.

Figure 12: Decoding flowchart.

Figure 13: Decoder.

Figure 14: Reconstruction process of subband samples when using linear prediction for this channel.

Figure 15: The reconstruction process of subband samples when using cross-channel linear prediction.

Figure 16: Reconstruction process of sub-band samples with local and cross-channel linear prediction.

Fig. 17: Schematic diagram of the distribution and transmission of digital TV programs using the encoder and decoder of the present invention.

Detailed ways

The encoding and decoding method of the present invention is almost completely separated and identical to the processing of each channel, so the following description of the present invention is based on one channel, unless otherwise specified.

coding

[Example 1]

As shown in Figure 2, the encoder of the present invention consists of a frame length selector 1, a sub-band decomposition filter bank 2, a transient detector 3, a bit allocator 4, a sub-band quantizer 5, and a multiplexer 6 Composition, used to compress and encode the input audio signal. The audio signal may be an audio signal of a digital television. The following describes the working principle of the encoder through a detailed description of each component.

Frame length selector 1:

The function of the frame length selector 1 is to divide the input audio signal into frames according to a certain frame length. As for the choice of frame length, it is generally believed that the larger the frame length, the higher the encoding compression efficiency, but the greater the encoding delay, the greater the demand for memory units by the encoder and decoder. Therefore, when the code rate is high, a smaller frame length is generally selected, and when the code rate is low, a larger frame length is generally selected. Since the encoding delay is proportional to the product of the frame length and the sampling rate, the sampling rate must be considered when choosing the frame length.

When the audio signal is the accompanying sound of the video signal, in addition to the above factors, some special requirements for the accompanying sound of the video signal must be considered when selecting the frame length. The most prominent of these special requirements, which is also one of the problems to be solved by the present invention, is that the compressed audio code stream can be edited synchronously with the video signal. The present invention uses "synchronous editing" to mean that the reasonable editing points of the compressed audio code stream and video code stream are fully synchronized in time, so that after editing the audio and video code stream at any of their editing points, the audio There is no error in the video code stream, and the decoded audio and video signals can transition smoothly at the editing point.

The basic requirement for realizing synchronous editing is that the time occupied by one frame of audio stream must have a simple multiple relationship with the time occupied by one frame of video stream. Otherwise, when the video code stream is clipped on the boundary of the video frame, it will be clipped inside the audio frame, destroying the structure of the audio frame and causing a decoder error. We consider two cases. The first is the case where the video frame length is an integer multiple or equal to the audio frame length. At this point, clipping can be made on the boundary of each video frame, and the clip point will always fall on the boundary of the audio frame.

For a video signal of 25 frames per second (PAL), the time taken by each frame is 0.04 seconds. Relative to an audio signal sampled at 48000kHz, it is equivalent to 1920 samples. Since 1920=3×5×2 ⁷ , the audio frame length formed by any combination of these factors can be edited synchronously with the 25 frame/second video signal.

For a video signal of 24 frames/second (movie), the number of samples of an equivalent frame of audio signal (48000kHz sampling rate) is 2000=2 ⁴ ×5 ³ . Similarly, the audio frame length formed by the combination of these factors can be edited synchronously with the video signal.

Another restriction on the selection of the audio frame length is that it must be an integer multiple of the number of subbands of the subband decomposition filter bank to be experienced later. If some combinations are selected from the greatest common factor 2 ⁴ ×5 of 1920 and 2000 as the number of sub-bands, we can achieve the two video signals of 24 frames per second and 25 frames per second without changing the number of sub-bands Synchronized editing. This will be further described in the description of subband decomposition filter banks.

Now consider the second case, that is, the case where the audio frame length is an integer multiple (such as V times) of the video frame length. In order to ensure the integrity of the audio frame, there must be an audio signal clipping point every V video frames. The compressed video stream often has a clip point every several frames, and this interval may be dynamic. Assuming that the maximum value of this dynamic interval is W, then the maximum clip interval that can satisfy both video and audio clip constraints is WV video frames. This obviously requires a lot of storage. However, the methods described earlier still apply.

Subband Decomposition Filter Bank 2

The sub-band decomposition filter bank 2 is used to decompose the audio signal transmitted by the frame length selector 1 into M sub-band signals. A wide variety of filterbanks can be employed here [Ref 8], including but not limited to perfect reconstruction and imperfect reconstruction filterbanks, orthogonal and non-orthogonal filterbanks, tree-like filterbanks, cosine-modulated filter banks, wavelet packets, etc. In order to ensure the smoothness of the audio signal after clipping the coded stream encoded by the present invention, it is particularly recommended to use a filter bank with a long overlap.

Due to the low calculation amount and simple design, the cosine modulated filter bank is the preferred filter bank of the present invention. Its kth subband filter is [Ref 8]:

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 0.5</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

where n is the sample index, k is the subband index, M is the number of subbands, p ₀ ( ) is the prototype filter, N is the length of the prototype filter, ${\mathrm{\&lambda;}}_k={(-1)}^k\frac{\mathrm{\&pi;}}{4}.$

The present invention has certain restriction to the sub-band number M of sub-band decomposition filter bank 2, and its principle is that the sub-band number M must be a common factor of the audio frame length selected by the frame length selector 1 corresponding to different video frame frequencies . That is, all available audio frame lengths can be expressed by the following formula:

Fsize＝k*M (2)

Among them, Fsize is the audio frame length, and k is a positive integer. In this way, when the frame rate of the video changes, the frame length selector 1 only needs to select another k to select an audio frame length corresponding to the new video frame rate while keeping the number of subbands M constant. . Keeping the number of subbands constant means that the various delay lines of the encoder and decoder are unchanged. When the frame frequency of the video signal changes, both the encoder and the decoder do not need to be reset, and only need to adjust k in the above formula to adapt to the change smoothly.

For example, when describing the frame length selector 1, it has been mentioned that for an audio signal sampled at 48kHz, 2000 can be used as the audio frame length for a video signal of 24 frames/second, and 1920 can be used as a video frame length for a video signal of 25 frames/second The audio frame length of the signal. If the number of subbands M is selected as the greatest common factor 80 of 2000 and 1920, then only k=25 and 24 can be selected to realize the audio frame lengths of 2000 and 1920 respectively under the condition of keeping the number of subbands unchanged. Of course, other common factors, such as 40 and 20, etc., can all achieve the same effect. Table 1 has listed the preferred audio frame length of the present invention and the number of sub-bands of the filter bank:

Table 1

The short frame lengths in the above table, such as 200, 240, 400, 480, etc., obviously provide low-latency modes for this audio compression method. Therefore, the limitations of the present invention on the audio frame length and the number of subbands do not prevent the present invention from having a low-latency mode.

Transient Detector 3

The sub-band signal output from the sub-band decomposition filter bank 2 then enters the transient detector 3 . The transient state detector 3 analyzes the transient state of each frame of sub-band signals according to a certain detection scale, and then further divides a frame of sub-band signals into transient segments and steady-state segments, and outputs the position information of each segment. It is important to emphasize that the segment positions are adaptively changed with the transient conditions. Subsequent processing of sub-band signals in the present invention takes a segment (hereinafter referred to as a sub-band segment) as a basic unit.

Taking the sub-band signal shown in Figure 3 as an example, the detector 3 will judge that A is a steady state segment, B is a transient state segment, and C is a steady state segment, and output the position information of each segment such as segment length 30, 20, and 50.

The scales used for transient detection include but not limited to the energy of the subband signal, the logarithm of the energy, the energy entropy, and so on. The detection technique can be simple threshold detection, or some very complex techniques, including but not limited to the well-known K-Means algorithm [Reference 8].

Bit allocator 4:

The bit allocator 4 can allocate bits for each subband segment of each channel according to the signal-to-noise ratio (SNR) or the signal-to-coverage ratio (SMR) by a method commonly used in the art [references 1-9]. The result of the bit allocation is the number of bits per subband sample used to quantize each subband segment for each channel. The result is sent to subband quantizer 5 and multiplexer 6 .

Subband Quantizer 5

The sub-band quantizer 5 includes a set of quantizers with different quantization methods and quantization levels [Reference 9]. The selection and configuration of this group of quantizers has a great influence on the compression efficiency. Because the probability distribution of the quantization index is uneven (especially when there are few quantization levels), the general compression technology mostly adopts the method of scalar quantization plus entropy coding (such as Huffman code). However, the decoding method of entropy coding is complicated, and the amount of calculation is large and uneven, which brings many difficulties to the commercial realization of the decoder.

For this reason, the present invention preferably adopts vector quantization when there are few quantization levels, and scalar quantization when there are many quantization levels.

The sub-band quantizer 5 selects a specific quantizer from the quantizer group for each sub-band segment according to the bit allocation result, and quantizes each sub-band sample in the sub-band segment with this quantizer.

The quantization process of each subband sample in each subband segment is divided into the following four steps (Fig. 2):

1) Estimated scale factor: The scale factor estimator 51 may use the maximum value of the absolute value of all samples in the subband segment, the variance of all samples in the subband segment, or other variables as the scale factor.

2) Quantization of the scale factor: the scale factor itself also needs to be quantized in order to be transmitted to the decoder. Since the sensitivity of the human ear to the volume decreases as the volume increases, the quantization of the scale factor should adopt a nonlinear method, such as logarithmic quantization. Taking the variance as an example, if the variance of the dth segment of the kth subband of the cth sound channel is σ(c, k, d), then the quantization index selected by the scale factor quantizer 52 for the scale factor is:

$\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; \&sigma;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

where α is the quantization step size.

3) Normalize sub-band samples: the sub-band sample normalizer 53 normalizes all samples in the sub-band segment with the scale factor reconstructed after quantization.

4) Quantize the subband samples: the quantizer selector 54 selects a specific quantizer according to the number of bits sent by the bit allocator 4, and then the subband sample quantizer 55 uses it to quantize the normalized subband samples to obtain The quantization index for each sample.

Multiplexer (MUX) 6

The multiplexer 6 packs together the following information generated by the above encoder components to form a complete bit stream or code stream.

The audio frame length selected by frame length selector 1.

Segment position information output by transient detector 3.

Number of bits allocated by bit allocator 4 for each subband segment.

The subband quantizer 5 produces the quantization indices of the scale factors and the quantization indices of each subband sample.

Multiplexer 6 also packs other auxiliary information. Such auxiliary information includes but is not limited to the sampling frequency of the input audio signal, speaker settings, error correction code, time code, etc.

[Example 2]

The encoder of the second embodiment of the present invention is shown in FIG. 4 . Most of its components are the same as those in Embodiment 1, the difference lies in the unique bit allocation method of this embodiment. Specifically, the bit allocation in this embodiment is not as in Embodiment 1 to allocate bits for each subband segment according to the signal-to-noise ratio (SNR) or the signal-to-coverage ratio (SMR), but according to the scale factor quantizer 52 output The quantization index of the scale factor uses the following formula to allocate bits for each subband segment.

b(c,k,d)=f(α·s(c,k,d))-θ(k)-β(3)

in:

1) b(c, k, d) is the number of bits per sample allocated to the current subband segment.

2) f(α·s(c, k, d)) is a strictly monotonically increasing function. It can preferably be set as f(α·s(c, k, d))=[α·s(c, k, d)] ^q , where 0<q≦2. It can further preferably be set to f(α·s(c, k, d))=α·s(c, k, d).

3) θ(k) can preferably be set as the approximate curve of the hearing threshold (Threshold in Quite) of the human ear in a quiet environment as shown in Figure 5 [References 1, 2 and 10], more preferably set to zero to simplify calculations.

4) β is the bit allocation adjustment factor.

It can be seen from formula (3) that when the bit allocation adjustment factor β is determined, the number of bits allocated to each sample of each subband completely depends on the quantization index of its scale factor.

Obviously, the smaller β is, the more bits are allocated to each sub-band segment by formula (3); the larger β is, the less bits are allocated to each sub-band segment by formula (3). The task of bit allocation is to find the minimum value of β under the condition that the sum of the bits occupied by each sub-band segment does not exceed the total number of bits allocated to each frame of audio signal for a given target code rate (bit rate).

The bit allocation can be globally optimal, that is, all channels share a bit allocation adjustment factor. Assume that the total number of bits allowed to be allocated to each frame of audio signal at a given target bit rate is Total Bits (this is the total value after subtracting the number of bits required to transmit various auxiliary information, later always so assumed, no further explanation), the bit allocation adjustment factor searcher 41 must search for different β values to find a minimum β value that satisfies the following conditions:

$\underset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; d</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; Total</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; Bits</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

where n(c, k, d) is the number of samples of the subband segment.

The obtained β can be used to allocate bits for each subband segment of all channels according to formula (3). At the same time, the encoder only needs to pass this bit allocation adjustment factor to the decoder, and the decoder can reconstruct the bit allocation used by the encoder for each sub-band segment of all channels according to formula (3) according to it and the quantization index of the scale factor .

The bit allocation can also be locally optimal, eg each channel has an individual bit allocation adjustment factor. Assume that the total number of bits of each frame of audio signal allocated to the c-th channel in a predetermined way at a given target bit rate is Total Bits(c). The bit allocation adjustment factor searcher 41 has to search through different β values to find a minimum β value that satisfies the following conditions:

$\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; d</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; Total</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; Bits</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

In this case, the encoder has to transmit a bit allocation adjustment factor for each channel to the decoder. Apparently, this method can be extended directly to the situation of sharing the adjustment factor of bit allocation in other forms.

In summary, the bit allocation procedure of this embodiment is as follows:

1) All scale factor quantizers 52 of each channel send the scale factor quantization index of each subband segment to the bit allocation adjustment factor searcher 41 .

2) The bit allocation adjustment factor searcher 41 finds a globally optimal bit allocation adjustment factor at a given code rate according to formulas (3) and (4), and passes it to the multiplexer 6 and the bit allocator 42. Or, the bit allocation adjustment factor searcher 41 finds a local optimal bit allocation adjustment factor for each channel at a given code rate for each channel according to formulas (3) and (5), and uses them are sent to the multiplexer 6 and the respective bit allocator 42 respectively.

3) The bit allocator 42 allocates bits for each sub-band segment according to formula (3), and sends the bit to its corresponding quantizer selector 54 .

It can be seen from the above that the bit allocation of the present invention only uses the human auditory model very limitedly to achieve the purpose of simplifying the bit allocation. In this way, the computational complexity of bit allocation can be greatly reduced, and only one bit allocation adjustment factor can be used to express the result of bit allocation. Only this bit allocation adjustment factor must be included in the coded code stream. After receiving this parameter, the decoder can easily and accurately reconstruct the bit allocation used by the encoder using equation (3). This saves the bit resources required by other techniques to convey the number of bits allocated to each sub-band segment. These saved bit resources can be used to transmit the quantization index of the sub-band samples, thereby further improving the sound quality.

[Example 3]

The encoder of the third embodiment of the present invention is shown in FIG. 6 . Most of its components are the same as those of other embodiments, the difference is that this embodiment has one more joint strength encoder 7 than other embodiments. The theoretical basis of joint intensity coding is that the human ear's spatial localization of sound at high frequencies (such as above 7 kHz) is mainly based on the intensity of the sound. As shown in Figure 6, the joint strength encoder 7 can add up the high-frequency subbands output by the transient detector 3 of the left and right (or other joint) channels when encoding, and only pass this sum subband (called is the quantization index of each sample of the jointly coded sub-band of the source channel, plus the intensity index of the combined sub-band of the combined channel to achieve the purpose of saving bits.

When joint intensity coding is used, the allocation of bits to jointly coded subband segments of source channels must take into account the bit requirements of other jointly coded subband segments of the same subband. Assuming that the source channel is c, and the total set of other jointly coded channels is J, the scale factor that should be used when calculating the bit allocation of the jointly coded subband segments of channel c is:

$\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; J</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

If the starting frequency of the joint strength coding is too low, the spatial positioning may be narrowed. Therefore, in this embodiment, the joint strength coding is only used when the code rate is low.

Example 4

An encoder of a fourth embodiment of the present invention is shown in FIG. 7 . Most of its components are the same as those of other embodiments, the difference is that this embodiment has one more cross-channel long-term and short-term linear predictor 8 than other embodiments. For each sub-band signal, the present invention searches for its short-term and long-term autocorrelation with the sub-band, and its co-correlation with signals of the same sub-band of other channels to find a linear predictor that minimizes the prediction error 8 . Let x(c, k, n) be the n-th sample of the k-th subband of the c-th channel, then the linear prediction of the c-th channel based on the s-th channel is

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{\overset{\mathrm{<span\; class="notranslate">\; A</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; B</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

where a(m) and b(m) are the coefficients of the prediction filters of the short-term and long-term predictors, respectively, and τ is the delay of the long-term prediction filter. When s=c, the above prediction is completely based on the current channel, m ₀ =1; when s≠c, the above prediction is cross-channel, m ₀ =0.

After making predictions for each subband sample x(c, k, n) using formula (7), a corresponding prediction error can be obtained:

$\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

The task of the encoder is to find an optimal set of prediction coefficients a(m), b(m), and delay τ to minimize the total prediction error in this subband segment as the following mean square error

${\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; .</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

If the periodicity of the subband signal x(c, k, n) is strong, the variance of the prediction error will be much smaller than the variance of the subband signal x(c, k, n) itself. This means that the prediction gain of linear prediction is high and the prediction is successful. At this time, the prediction error signal e(c, k, n) can be used to replace x(c, k, n) and sent to the subband quantizer 5 . Otherwise, the subband signal x(c, k, n) is directly sent to the subband quantizer 5 . Therefore, the working procedure of linear predictor 8 can be summarized as follows:

1) Estimate the prediction coefficient and prediction gain.

2) If the prediction gain is high, send the decision of linear prediction, prediction coefficient and delay to the multiplexer 6 for this sub-band segment. At the same time, a prediction error signal is generated according to (8) and sent to the quantizer 5 .

3) If the prediction gain is not high, send the decision of not using linear prediction to the multiplexer 6 for this sub-band segment. At the same time, the samples x(c, k, n) of this sub-band are sent to the quantizer 5 .

相加来重建子带样本。而本声道预测器81也只能用这些重建的子带样本来预测未来的子带样本。也即，代入(7)式的子带样本实质上是重建的子带样本。When local channel prediction is used, the quantization of the prediction error signal must be performed in a closed loop [References 8, 9 and 11], as shown in Figure 8, in order to avoid quantization error spreading during decoding. Since the decoder can only obtain the quantization index of the prediction error output by the quantizer 5, the decoder can only use the inverse quantizer 9 to reconstruct the prediction error, and then compare it with the prediction value

are summed to reconstruct the subband samples. However, the channel predictor 81 can only use these reconstructed subband samples to predict future subband samples. That is, the subband samples substituted into Equation (7) are essentially reconstructed subband samples.

Similarly, when making cross-channel prediction, in order to avoid the spread of quantization errors during decoding, x(s, k, n) must also be reconstructed subband samples after decoding, as shown in FIG. 9 . Note that predictor 82 in Figure 9 is a cross-channel predictor because its input subband samples come from another channel.

The cross-channel predictor 82 and the local-channel predictor 81 can be used together, as shown in FIG. 10 . First, calculate the error for the cross-channel prediction:

${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{\overset{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

Then, the channel prediction is made for this error:

${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}{\overset{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

When cross-channel prediction is used, the encoder must ensure that the source channel is already decoded when decoding. That is to say, the encoder must ensure that the decoding of each channel must follow a certain achievable order: the first channel of decoding must only be predicted by this channel, and the second channel can only be predicted by this channel Or the previously decoded channel for cross-channel prediction. and so on.

The encoder can search all achievable decoding orders to find the decoding order with the largest prediction gain. The encoder can also only do a limited search for a suboptimal solution.

Forecasting Example 1: Long-Term or Short-Term Forecasting

Since it is difficult to estimate the prediction coefficient when long-term and short-term forecasting are used together, either long-term or short-term forecasting can be used to reduce complexity:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; B</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; )</span>$

At this time, when τ is small, it is short-term forecast, and when τ is large, it is long-term forecast.

When the above formula is used in the case of both cross-channel prediction and local-channel prediction, first calculate the residual of cross-channel prediction:

${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>$

Among them, the minimum value of the delay τ ₁ can be zero. Then, make this channel prediction for this residual:

${\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span>$

Among them, the minimum value of the delay τ ₂ is 1.

Prediction Example 2: Cross-Channel Prediction with Limited Search

In order to simplify the complexity of searching for the decoding order of cross-channel prediction, this example first finds the optimal order of the first two channels. Only local-channel predictions and cross-channel predictions using the first two channels as source channels are searched for subsequent channels.

In order to further reduce the complexity, no sequential search may be performed, and the first channel is always used as the source channel for cross-channel prediction of all other channels.

Preferably, the encoder of the present invention can also perform cross-channel and/or local-channel prediction on the audio frame before the audio frame is input to the subband decomposition filter bank 2, that is, between the frame length selector 1 and the subband decomposition A cross-channel and/or local-channel predictor is further included between the filter banks 2, after which, the prediction error is input to the sub-band decomposition filter bank 2, and the prediction error is coded according to a similar process as described above.

Coding process:

The aforementioned four embodiments can all be a complete encoder. However, if all their functions are combined to form an encoder, the best compression efficiency can be achieved. Fig. 11 shows the coding process when all the functions of the four embodiments of the present invention are combined. Among them, joint intensity coding, cross-channel long-term and short-term prediction, and global bit allocation are independently optional. When any one of them is not selected, it only plays a function of sending out the data sent in without any modification in Fig. 11 . Of course, if the bit allocator in Fig. 11 is not selected, another general bit allocator discussed in Embodiment 1 must be introduced to achieve the same function. The encoding steps of the present invention are described below in conjunction with FIGS. 2, 4, 6, 7, 8, 9 and 10 (see FIG. 11).

E1) frame length selection: frame length selector 1 receives samples of the audio signal input from each channel, according to the sampling rate of the audio signal, the target code rate, and the video frame frequency (when the multi-channel audio signal is used as the video signal When accompanied by audio), select the audio frame length, and then pass the frame length information to other components of the encoder. Since the encoder and its method of the present invention use frames as the basic unit, all components of the encoder and each step of the encoding process directly or indirectly use the frame length. However, for the sake of brevity and clarity of description, FIG. 11 does not mark all the transmission paths of the frame length information. After the frame length is determined, the frame length selector 1 also divides the samples of the input audio signal into frames according to the frame length, and sends them to the sub-band analysis filter bank 2 frame by frame. This step does not do any processing on the input audio signal itself.

E2) Subband decomposition: the subband decomposition filter bank 2 decomposes the audio signal of each channel into M subband signals.

E3) Transient state detection: the transient state detector 3 analyzes the transient state of each sub-band signal, and accordingly divides it into a transient state segment and a steady state segment. The location information for each segment is then passed to the other components of the encoder. For reasons similar to step E1, Fig. 11 does not show all transmission paths of the position information of the temporary/stable segment. This step does not perform any processing on the sub-band signal itself.

E4) Joint strength coding: after joint strength coding, the joint strength encoder 7 discards the samples of the joint subbands, and only transmits the strength index of each subband segment to the bit allocator (E7) device 4 and the multiplexer Multiplexer (E9) 6. This step is preferred in the present invention when the code rate is low. Not using this step will not affect the integrity of the coder or method, but the coding efficiency will be reduced.

E5) Cross-channel long-term and short-term prediction: After the cross-channel long-term and short-term predictor 8 completes the linear prediction, it sends the decision whether to use prediction to the multiplexer (E9) 6. If it is decided to use prediction, the delay of the prediction filter and the prediction coefficient are also passed to the multiplexer (E9) 6, and the prediction error signal is used instead of the subband signal to the subband quantizer 5. This step is preferred in the present invention, and not using this step will not affect the integrity of the encoder or the method, but the encoding efficiency will decrease.

For ease of description, Figure 11 divides the function of the subband quantizer 5 into two steps: scale factor (E6) and vector/scalar quantization (E8).

E6) Scale factor: the subband quantizer 5 estimates and quantizes the scale factor in units of temporary/steady-state segments from the subband signal (if linear prediction is adopted, it is the prediction error signal, hereinafter collectively referred to as the subband signal). Then, the quantization indices of the scale factors are sent to the bit allocator (E7) 4 and the multiplexer (E9) 6. This step does not perform any processing on the sub-band signal itself.

E7) Bit allocation: the bit allocation adjustment factor searcher 41 finds the optimal bit allocation adjustment factor according to the quantization index of the scale factor input by step E6, and the intensity index (if joint intensity coding is selected) input by E4, And pass it to the multiplexer (E9) 6. Then, the bit allocator 42 distributes the bits to each subband segment according to the formula (3), and the assigned bit number of each subband segment is passed to the quantizer selector 54 for vector/scalar quantization (step E8) use. The above bit allocation method is preferred in the present invention. If this method is not preferred, an alternative method of bit allocation must be used to maintain the integrity of the encoder. This step does not perform any processing on the sub-band signal itself.

E8) vector/scalar quantization: quantizer selector 54 selects a quantizer for each subband section according to the number of bits allocated to each subband section sent by bit allocator (E7) device 4, and then sends it to Subband sample quantizer 55 . The subband sample quantizer 55 then quantizes each subband sample in units of subband segments, and passes its quantization index to the multiplexer ( E9 ) 6 .

E9) multiplexing (MUX): the multiplexer 6 packs (multiplexes) the quantization index of each subband sample and the following auxiliary information into a complete audio frame and outputs it: audio frame Length, location of transient/stationary segments, intensity index (if joint intensity coding is selected), decision whether to use prediction, delay and coefficient of prediction filter, quantization index of scale factor, and bit allocation adjustment factor. The multiplexer 6 can also package (multiplex) and output other auxiliary data, such as sampling frequency, speaker settings, error correction code, time code and so on.

decoding

The decoder and its decoding method of the present invention are essentially the inverse process of the encoder and its method. Here, the decoding process and components of the decoder are described based on FIGS. 12 and 13 .

Decoding process:

The code stream that produces through encoding method of the present invention and encoder must go through the following main steps (Fig. 12) to decode and reconstruct the multi-channel audio signal:

D1) Unpacking (DEMUX) auxiliary information: the multiplexing depacker 110 unpacks the following auxiliary information:

• Frame length.

• Location of all subband segments (transient/stationary segments).

• Bit Allocation Adjustment Factor.

• Scalefactor quantization index for each subband segment.

• A decision whether to use cross-channel long-term and short-term prediction for each subband segment; if so, further unpacking out the delays and coefficients of the prediction filter.

The strength index of the subband segments that are jointly strength coded

D2) Bit allocation: The bit allocator 42 allocates bits for each sub-band segment according to the input bit allocation adjustment factor and the scale factor quantization index of each sub-band segment. This bit allocator is exactly the same as the bit allocator 42 used by the encoder, so the reference number 42 is still used. This bit allocation method is preferred in the present invention; if the encoder adopts other bit allocation methods, this step must be skipped. But step D 1 usually has to add an unpacking item: unpacking the bit allocation from the input code stream.

D3) Unpacking the quantization index of the sub-band samples: the multiplexing unpacker 110 unpacks the quantization index of each sub-band sample from the input code stream according to the number of bits allocated in step D2.

D4) Inverse quantization and reconstruction of sub-band samples: the sub-band sample inverse quantizer 120 reconstructs each sub-band sample according to the scale factor quantization index unpacked in step D1 and the sub-band sample quantization index unpacked in step D3.

D5) Cross-channel long-term and short-term prediction: For each sub-band segment, if the prediction decision unwrapped in step D1 is positive, then the cross-channel long-term and short-term predictor 130 makes predictions for all samples in this sub-band segment. Otherwise, the cross-channel long-term and short-term predictor 130 does nothing for all samples of this subband segment. This step is preferred in the present invention; if the encoder does not use cross-channel long-term and short-term prediction, then this step must be skipped.

D6) Joint intensity decoding: For each subband segment that is jointly intensity encoded, the joint intensity decoder 140 first copies the subband samples from the source subband segment to the current subband segment. Then, rebuild the intensity scale factor according to the intensity index unpacked in step D1, and use it to correct the sample value copied to the sub-band. This step is preferred by the present invention when the code rate is low; if the encoder does not use joint strength coding, this step must be skipped.

D7) Synthesis filter bank: The synthesis filter bank 150 synthesizes and reconstructs the subband samples into an audio signal.

If the encoder of the present invention has carried out cross-channel and/or local-channel prediction to the audio frame before the audio frame is input to the sub-band analysis filter bank 2, the signal reconstructed by the synthesis filter bank 150 must also be made at this moment. corresponding predictions to reconstruct the audio signal.

decoder:

The decoder of the present invention is shown in FIG. 13 . The individual components are described below:

Demultiplexer (DEMUX) 110:

The demultiplexer 110 is responsible for depacketizing the data listed in the decoding steps D1 and D3 from the compression-encoded code stream. It is also responsible for unpacking other auxiliary data from the compressed code stream, such as sampling frequency, speaker settings, error correction code, time code, etc.

Bit allocator 42:

This bit allocator is exactly the same as the bit allocator 42 used by the encoder, so the reference number 42 is still used. The function of this bit allocator is to substitute the bit allocation adjustment factor and the scale factor quantization index of each subband segment into equation (3) to obtain the number of bits per sample allocated to each subband segment.

Subband Sample Inverse Quantizer 120:

The subband sample inverse quantizer 120 selects a quantizer according to the bit allocation of each subband segment. Then, the subband samples are reconstructed from the quantization indices of the subband samples with this quantizer and the scale factor of the subband segment.

Note that when a certain sub-band segment adopts cross-channel long-term and short-term prediction, the sub-band sample dequantizer 120 reconstructs the prediction error signal of the sub-band segment instead of the sub-band sample itself.

Cross-channel long-term and short-term predictors 130:

For each sub-band segment, if the prediction decision unwrapped in step D1 is positive, the cross-channel long-term and short-term predictor 130 makes predictions for all samples in this sub-band segment. Otherwise, the cross-channel long-term and short-term predictor 130 does nothing for all samples of this subband segment.

The predictor for this channel prediction is shown in Figure 14. The predictor 81 is exactly the same as the predictor 81 used by the encoder, so the reference number 81 is still used.

The predictor for cross-channel prediction is shown in Figure 15. The predictor 82 is exactly the same as the predictor 82 used by the encoder, so the reference number 82 is still used.

When cross-channel and local-channel prediction are used together, the local-channel prediction is performed first, and then cross-channel prediction is performed, as shown in FIG. 16 .

Joint intensity decoder 140:

For each subband segment to be jointly intensity coded, the joint intensity decoder 140 first copies the subband samples from the source subband segment to the subband segment. Then, the intensity scale factor is reconstructed from the intensity index and used to correct the sample values copied into this subband.

Subband synthesis filter bank 150:

The subband synthesis filterbank 150 synthesizes and reconstructs the subband samples into an audio signal. The subband synthesis filterbank 150 is the inverse of the subband analysis filterbank 2 in the encoder and is designed simultaneously. That is, after the subband analysis filter bank 2 in the encoder is determined, the subband synthesis filter bank 150 is also completely determined [Reference Document 8].

application solution

The present invention relates to multi-channel digital audio compression encoding/decoding technology, which has the advantages of high compression efficiency, simple decoder, high fidelity of the decoded audio signal, and can be applied to high, medium, and low bit rates various applications. Therefore, it is completely suitable for pure audio applications, such as digital audio broadcasting, etc.; its decoder can be installed independently in pure audio equipment, such as power amplifiers, walkmans, etc.

Because the audio signal mostly appears in the form of accompanying sound of the video signal, and because the compression efficiency of the encoding and decoding method of the present invention is high, the audio signal encoded by the present invention can be edited synchronously with the video signal and can withstand more than ten tandem encodings , so it has the following advantages in practical application:

1) It can meet the requirements of distribution and transmission of programs at the same time.

2) The encoding technology of the present invention greatly simplifies the link and equipment of program distribution. Taking digital TV as an example, Fig. 17 shows the process of program distribution and transmission using this technology. Obviously, it is much simpler than Dolby's solution (Fig. 1).

3) This technology greatly improves the fidelity of the program distribution process because multiple transfer coding links are omitted.

4) Since multiple encoders and decoders in FIG. 1 are omitted, this technology also greatly reduces the cost of program distribution.

references

[1] ISO/IEC 13818-7, 1997.

[2] ISO/IEC 14496-3, 1998.

[3] S.Smyth, M.Smyth, and W.P.Smith, "Multi-channel Predictive Subband Audio conder using Psychoacoustic AdaptiveBit Allocation In Frequency, Time, And Over The MultipleChannels," US Patent 5956674.

[4] M. Smyth, "An overview of the Coherent Acoustics coding system," http://www.dtsonline.com/whitepaper.pdf , 1999.

[5] S.Smyth, W.P.Smith, M.Smyth, M.Yan, and T.Jung, "DTS Coherent Acoustics Delivering High Quality MultichannelSound to the Consume," AES 100th Convention, 1996.

[6]L.Fielder and C.Todd, "The design of a video friendly audio coding system for distribution applications," AES ^17th Internaltional Conference, pp.86-92, 1999.

[7] C.Todd, G.Davidson, M.Davis, L.Fielder, B.Link, and S.Vernon, "AC-3, Flexible perceptual coding for audiotransmission and storage," 96 ^th AES Convention, Amsterdam, 1994 .

[8] P.P. Vaidyanathan, "Multirate systems and filterbanks," Prentice Hall, 1993.

[9] A.Gersho and R.M.Gray, "Vector quantization and signal compression," Kluwer, 1992.

[10] B.C.J.Moore, "An introduction to the psychology of hearing," Academic Press, 1997.

[11] A.M. Kondoz, "Digital Speech," John Wiley & Sons, 1994.

Claims (14)

1. A coding method for compressing and coding a multi-channel digital audio signal, comprising the following steps: A) frame length selection step, select the frame length according to the sampling rate of the audio signal and the target code rate, and divide the audio signal into frames according to the frame length; B) the sub-band decomposition filtering step, decomposing the frame-by-frame input audio signal into a plurality of sub-band signals with the sub-band decomposition filter bank; c) a transient state detection step, which divides the input sub-band signal into a transient state segment and a steady state segment, and the segment length of each transient state segment or steady state segment is adaptively changed with the transient state and the steady state condition; d) the first prediction step, following the transient detection step, using samples from other channels or past samples of this channel to predict the current sample of this channel; e) a scaling factor estimation step, estimating and quantizing the scaling factor of each subband signal, and outputting the quantization index of the scaling factor; f) bit allocation step, the number of bits determined by the target code rate is adaptively allocated to each sub-band segment according to the scale factor quantization index of each sub-band segment, wherein it is allocated to the k-th sub-band of the c-th channel in the d-th segment The number of bits b(c,k,d) of the samples of is determined by the following formula: b(c,k,d)=f(α·s(c,k,d))-θ(k)-β Where f(α·s(c, k, d)) is a strictly monotonically increasing function, s(c, k, d) is the quantization index of the scale factor of the subband segment, α is its quantization step size, θ( k) is an approximate curve of the hearing threshold of the human ear in a quiet environment, and β is a bit allocation adjustment factor; g) sub-band quantization step, quantizing the sub-band signal in units of segments according to the number of bits allocated to each sub-band segment; h) multiplexing step, the quantization index of the sub-band samples produced by the sub-band quantizer and the auxiliary information including the quantization index of the frame length, segment position information, scale factor and bit allocation information are multiplexed and packaged into one The complete code stream in frame unit. 2. encoding method as claimed in claim 1, when described audio signal is used as the accompanying sound of video signal, the frame length selection step wherein also selects audio frame length according to video frame frequency, makes audio frame length be video frame length Rational fractions of or integer multiples of . 3. coding method as claimed in claim 2, wherein said sub-band decomposition filter bank can decompose the sub-band number that audio signal is divided into each audio frame length that is selected by frame length selection step according to different video frame frequencies common factor of . 4. The encoding method according to claim 1, wherein the sub-band quantization step selects a scalar quantizer or a vector quantizer according to the quantization level. 5. The encoding method as claimed in claim 1, wherein f(α·s(c, k, d))=[α·s(c, k, d)] ^q , where 0<q≦2. 6. The encoding method according to claim 5, wherein θ(k)=0. 7. The encoding method according to any one of claims 1-6, wherein said bit allocation adjustment factor is either shared by all sound channels, or used separately for each sound channel, or for all sound channels Divided into several groups and each group uses one separately, but these bit allocation adjustment factors make the sum of the number of bits allocated to each subband sample plus the number of bits required to transmit auxiliary information not exceed the target bit rate allowed to be allocated to each frame of audio The total number of bits of the signal. 8. The encoding method as claimed in claim 1, further comprising a joint intensity encoding step, arranged between said steps c) and d), for each jointly encoded subband segment, used in the bit allocation step The scalefactor quantization index of must be the maximum value of the scalefactor quantization indices of all channels encoded together in the same subband segment. 9. An encoding method as in any one of claims 1-6, further comprising a second prediction step, which is arranged before said sub-band analysis and filtering step, and utilizes samples of other channels or past samples of this channel to estimate the current channel The current sample is used for prediction. 10. The coding method as claimed in claim 9, wherein any one of the prediction steps comprises using samples of other sound channels as a source to make long-term and/or short-term predictions for the current samples, and then perform long-term and/or short-term predictions on the current samples. Errors in the short-term forecast and then make a long-term and/or short-term forecast. 11. The encoding method as claimed in claim 9, wherein said other sound channels are limited to the first and second sound channels only. 12. The encoding method as claimed in claim 9, wherein said other channels are limited to the first channel only. 13. The encoding method according to claim 9, wherein the predictor used in any one of said predicting steps consists of a linear predictor including a delay factor. 14. The encoding method as claimed in claim 9, wherein the predictor used in any one of said predicting steps consists of a short-term linear predictor including a delay factor and a long-term linear predictor including a delay factor.

Images