TW584835B - Method and architecture of digital coding for transmitting and packing audio signals - Google Patents

Abstract

A digital coding method converts inputted audio signal into frequency sampling sequence to represent a spectrum integration of the audio signal; further based on a bit allocation procedure, quantitize the frequency sampling sequence into quantitized value. By means of a mask threshold value, the bit allocation procedure uses a parameter predictor to predict quantitized parameter. The quantitized value is coded as the coded data containing multiple bits. If the bit number of the coded data exceed a preset available bit number of the coded data, a recursive rate control loop will adjust quantitized parameter and quantitized gap size. Prior to quantitizing frequency sampling sequence, the method also follows a cutoff frequency to cut off the high frequency portion of inputted audio signal where the cutoff frequency is decided by the recursive rate control loop.

Description

584835 V. Description of the invention (1) Technical field to which the invention belongs The present invention relates to a digital coding method and structure for transmitting and packing signals, particularly in audio signai coding Bit allocation (bH aiiocation). Perceptual speech coding in the prior art, such as MPEG layer 1-3, advanced expressions, 'symbols' or time / frequency (T / F) coding, etc. have been widely used in consumer electronics, telecommunications and telecommunications Broadcasting technically. In these perceptual speech encoders, bit allocation is a major task in key modules that guide high complexity and determine encoding quality. FIG. 1 is a block diagram of a coding process (pr occess) in a perceptual audio coding. The time / frequency converter (T / j? Mapper) 1 0 1 is connected by a window, window-by-window Based on this, the audio S (n) is converted into a frequency range (frequenCy segments) S (m, f) from time range to frequency range. A variety of encoders (co d e r s) 103 are also used in the encoding process to achieve high compression ratios. The output value X (m, f) is the sequence encoded by the window segment index m and the frequency segment f, and the frequency range is used as the definition domain. Quantizer (quantizer) 丨 05 will be x (m, f) amount

584835

Formed into a limited number of levels (丨 e v e 1 s), χ, (一 立 广 + · νηι, ί) represent 0 and quantify the amount of treasure generated by Uuantizatlon nolse (,,) to minimize A-day contamination. For this purpose, the level of generalization is controlled by quantization parameters. === 缩 (C_FessiC) r0 is to classify the frequency lines into groups and handle them as uuantlzatlon bands. The number of frequency lines formed by a quantization is determined according to the important frequency bands (crltical / bands) and the bits that can be provided to transmit the quantization parameters. A variable length encoder (VLc) i07 represents a quantized sequence x, (m, f) after variable length encoding, which is considered in terms of the statistical probability of the converted signal . The aggregation unit (packing u n t) 1 0 9 aggregates the final transcoding results into a sequence defined by a specific audio protocol. The auditory analysis mode (^ ych〇acoustic model) analyzes the signal and provides the signal-pair-masking rati0 (SMR) from the analysis result to the quantized band from the signal analysis result. The bit allocator 113 refers to the masking thresholds provided by the auditory analysis mode in and the predetermined available bits (budge t) 11 5 to determine the quantization parameter. A non-uniform quantizer quantizes the spectral lines under the control of a bit allocation system. This bit allocation device considers the final audio quality and the required bits to determine the quantization method. because

Page 6 584835 V. Description of the invention (3) Therefore, the control of audio quality and the number of bits is a basic requirement of a bit allocator. US Patent Document 5,579,430 discloses an optimal, A digital transcoding program for optimal coding in the frequency domain '〇CF. It provides a music transcoding that is comparable to CD (compact-disc) quality, its feasible data rate (date rate) is about 2 bits / ATW, and 1 ^ 5 bits / under good FM_radio broadcast quality ATW data rate. Another US patent document 5, 924, 0 60 discloses a digital encoding program for transmitting and / or storing acoustic signals, which uses a factor of 4 and 6 to reduce the data rate without substantially reducing the music. The quality of the signal. Regarding the first and second layers of MPEG, a uniform quantizer (unifom == 11 zer) is used to control quality and bit requirements. Therefore, the bit allocator early allocates the total number of available bits to the sub-frequency band signal (beijing band sjg ^ ls) for quantization, so as to make the quantization noise audibility two; ^ TtlZaUon as minimized. In encoders such as MPE (^ 3 layer, MPEG-2 AAC, and MPEG4 T / F encoding, the second reason is that they all use non-uniform quantizers. Noise: change: change ... according to the perceptual allowable layer 3 and Changhua product = different values will be given, I, I / Λ code specifies the variable bit length, instead of quantizing the bit should be obtained from the quantization result 584835 5. Invention description (4) & lt The above disadvantages cause problems in estimating the quantization parameters. A two-layer nested loop iterative method (referred to as OCF) is proposed to solve this problem. As shown in Figure 2 It is shown that it is estimated by two kinds of iterative loops, that is, a rate control loop (rate-control loop) and a mouth-loop control loop (quaHty-c0ntr〇iiing ι〇〇ρ). The loop repeatedly adjusts the parameter values to conform to the limited bits obtained by the jjuff man encoding of the spectral lines of the refining process. The quality control loop repeatedly adjusts the parameter values to conform to the inverse quantization process

CinveTse quantization) The perceptual criterion of the quantization noise. For a frame method with F spectral lines, its complexity can be described as 〇 (F · R · π + F · q · r), where q and r are repeated quality control loops and ratios, respectively. The number of control loops, and "r" are the computational complexity of processing the spectral lines under the ratio control loop and the quality control loop, respectively. The complexity of the ratio control loop 7 is from Xiahua and spectrum Line VLC encoding, and the quality control loop complexity γ comes from deequ anti zat ion and noise measure. Both 々 and 7 are high complexity. And, the number of repetitions q And r depend on the initial value of the reification parameter and the adjustment method, and this complexity is even greater than the total complexity of hybridtransfom and auditory analysis mode in FIG. 1.

V. Description of the invention (5) Specify the quality of the bit code in the mouth quality control loop. The finger position 70,, and the localized band determine the two ways of processing the voice editor. Congshi, + θ yuan is assigned to every 10,000 # garden by address processing / 2: One method is to only have the worst noise-to-masking ratl in the loop repeatedly. ). The number of repetitions is large, which means a high degree of complexity: the method assigns bits in the mother's repeated loops. ::: All available bits are used up. The first- The complexity of this method is much lower. However, it makes people wonder whether the quality of this method is satisfactory. The first method has the same threshold and the sample encoding of this standard iso. The problem of the method is that it controls the quality and position. The number of loops is one. Generally used to deal with a limit, and the quality and number of loops. Of course, the noise can be deformed, so that the mask threshold and noise threshold have been widely accepted. The second method has been The use of the Tong A crane makes the quality substantially better. Two kinds of nested loops can not be used to cause convergence. Because there are two different rules for the yuan / Xiao consumption in the two loops, it may cause the bodyless to call it "Stagnation π problem (dead 1 〇ckpr 〇b 1 em) + The method of stagnation problem is to set the maximum value of the number of iterations to use some kind of heuristic parameter adjustment method to care about the product. In these methods, the quality still cannot be obtained. Summary of the invention

Page 9 584835, description of the invention (6) The present Mao Ming overcomes the shortcomings of the traditional digital coding process described above. Its main purpose is to provide a digital encoding method with high-quality and less complex operations for communicating and t-set speech signals. According to the present invention, the input audio is first converted into a sequence of frequency samples (feculence of frequency samples) to represent the spectral composition of the audio. The frequency sample sequence is quantized according to a bit allocation procedure and a parameter estimator (parameter). This parameter estimator estimates the quantization parameters directly based on the mask threshold. These quantized values are encoded in a variable-length encoding method, or they are directly assembled into a specific, fixed voice specification. If the full length of the encoded data exceeds the number of available bits, adjust the parameters and increase the quantization step size. This process is repeated until the number of available bits is greater than the number of bits required for this encoding. Finally, the final encoded sequences are aggregated into a sequence defined by specific speech specifications.

Three layers; can also be applied to most of the MPEG methods. The method of the present invention uses the MPEG third layer non-uniform sentence quantizer to explain the derivation steps in detail, and checks the complexity and speech quality of this perceptual coding method. According to this, the present invention uses segmental murmur_pair_mask rate to explain this derivation step 'and provides a closed-form equation to represent the bit / quantization grid size (bu / step size) and quantized speech. This method is not limited to liver "AAC (Advanced Speech Coding

Page 10 584835

The bit allocation standards are also as the first layer of mpeg and the perceptual editor. Since the novel provided by the present invention can be applied to an encoder having a uniform quantizer, the second layer. Another object of the invention is to mention. This structure contains a converter, a parameter estimator, a packaging unit, and a processor that can be implemented by a signal processor. 〇 = A structure quantizer that implements this digital encoding program, a variable-length encoder, and an adjustment. According to the present invention, the quantization parameters of the low bit rate encoding program are directly evaluated from the quality conditions. Because by using the ratio control loop, the quantizaton bandwidth and the required number of bits in the non-uniform frequency lines are considered and can be adjusted. For variable bit rate programming, the ratio control loop can be completely removed. The above-mentioned and other objects and advantages of the present invention are described in detail below in conjunction with the following drawings, detailed description of the embodiments, and the scope of patent applications. Implementation method Figure 3a illustrates the procedure of the audio coding method of the present invention. Referring to FIG. 3a, the turn-in audio is first converted into a sequence of frequency samples to represent the audio frequency.

刈 4835 V. Description of the invention (8) ^ Please synthesize. Then, this frequency sample sequence is allocated according to the _ species bits, and the quantization is performed to obtain a symbol with a lower accuracy. One moxibustion number estimation state directly estimates the quantization parameter based on the mask threshold value. The second threshold value is the noise limit that can be heard by the human hearing system. ^ Gate A compression system parameter that determines the signal level resolution. … Estimated one Selenium = ΐ 6 The encoding of the quantization symbol is coded with a τ variable-length encoder. One step is to check whether a predetermined number of available bits is sufficient. If the full length of the encoded data exceeds the available digits, adjust the parameters and increase the size of the quantization interval. Repeat until the number of available bits is greater than the number of bits required for this encoding. Definition: Ϊ The final encoded sequence is packed into a high-level 7L rate audio coded audio coded by a specific audio specification before the low bitization parameter The required ^ frequency (CUt-off amount will be adjusted in the estimated quantization parameter size. According to the required 2 loop steps, the rate of the audio encoding program is the encoding of the audio frequency. The frequency can be cut off in the evaluation parameter estimation. (CUt-Off). Tupu explains the steps of this low program. As shown in Figure 3b, when the number of low-bit rate speech bits exceeds the number of available bits, adjust the frequency) and then transmit to make the high-frequency The score was cut off before. If necessary, quantize the interval. For encoding of variable bit rate audio, the bit quality can be adjusted. In this case, the iterative ratio control can be completely removed. Figure 3c illustrates this variable bit rate step, where the step of iterative rate control loops is shown in Figure 584835.

Removed in 3a. The coding procedure of FIG. 3a to FIG. 3c of the present invention may be implemented by a signal processor. 工 工,. ^ _Private The detailed architecture is only disclosed later. According to the implementation architecture shown in FIG. 3a and FIG. 4a, the input audio is received and converted into a 产 1 婵婵 product, and a state sequence of 401 ^ frequency samples is obtained to represent a kind of spectrum synthesis of the continuous signal. A quantized cry 4n9 # 衣衣 Θ said — + is changed to 4 0 2, according to a bit allocation process ^, this frequency sample sequence is quantized into a finite number of parameter estimators 4 0 5, directly ώ) Α ^ ^ ^ 7 η The error is estimated by a cover threshold to quantify the parameters, and then an optimal stone horse 5 | 4 〇Cj Chiang when the number of reserved available bits is less than 22! Level encoding. Adjuster 4. 7 adjust this ... Dagger 4 is enough, when the data is used’- a

One dagger majority. A comparator 4 0 8 compares the number of available bits with the predetermined number of 〆 〆 151; ^ I The number of bits used is sufficient to allow the length of η: to check this may be 409, which will eventually be ㈣ΛΛΛ A sequence defined by a packing unit. Figure 4b and Figure 4c are respectively implemented by a specific audio specification-an adjuster 413 is used to adjust the architecture of the two pictures: and 3C. Referring to FIG. 4b, in the case of encoding, this is cut off ^, 胄 rate ', and the audio rate 411 is at a low rate. The adjuster 413 may also transmit the frequency to the high-frequency cut-off unit size. The high-frequency cut-off unit (: the quantization interval 411 in the quantizer 4 ° 2 is the cut-off frequency added to the converters 401 and FreQuency cut-off unit) and passes it :: Between the receivers 40 and 2 for receiving adjustments to the e-number estimator 4 05. Variable position

Page 13 584835 V. Description of the invention (ίο) In the case of meta-rate encoding, the components related to the love, _ h ϋ ^ _ override ratio control loop step have been removed, as shown in Figure 4c. In the present invention, a basic decision formula (closed method of parameter estimator and noise estimator in the determi allocation procedure is used to detail an MPEG ACC quantizer with MPEG third layer, and a constant mask_pair_ The noise rate p is a basic nistic formula) to calculate the quantization parameter at bit 2. This decision equation provides a formula for the non-uniform quantizer. The method and derivation steps of the present invention are described as examples. A similar procedure can be used for. The bit allocation of the present invention meets the requirements for the bit rate and noise shaping of each sub-band (SUb-band) through a single-step prediction. By directly using a mask threshold, the optimal global factor and the scaling factor of each sub-band are estimated. This global factor controls the total number of exhausted bits used, while this ratio adjustment factor controls the quantization noise of the relevant band relative to other bands. The following chapters first explain the requirements for this bit allocation, and then explain in detail the derivation steps of the noise predictor (n〇ise predict〇r) and the zero noise and negative noise masking rate (negative noise- to-masking ratio (NMR) The upper and lower bounds of a ratio adjustment factor. Bit points i and 4 conditions

584835 V. Description of the invention (11) First, consider the minimum value of the segmented N M R and express it as = 魟 gMinT / m

M {i) Jj where σA ^ /; > and the noise energy $ (noise energy) and masking energy related to the key band / respectively. Is the bit rate to minimize NMR. In a PCM encoder with Qiu bits per sample, the quantization error variance (Quantization error variance) is expressed as follows: (2) (3) Ν (η = Ρ2, σ2χ (〇 Therefore, this minimum iin2 ~ 2R (0a2.) Ar§MmZ a ^ —— mi [V σΜ {〇J must be limited by this total bit rate, which is (4)-according to the Lagrange multipliers method, its λ solution The following equations must be satisfied:

Page 15 584835

B (j) B (j) then

A (2 log 2) P2 v (./) 2 log 2 (-'(.n for all (5) (j) (j) So we must find i? ⑺ so that the noise mask ratio is proportional to this. Also That is,） ·), for all (6) noise, the standard level (noise level) must be proportional to the mask gate root value multiplied by a frequency rabbit 'in order to have the best segmented NMR. Secondly, consider the mask at this quantized band Mask thresholds and key bandwidths to select a standard level of noise in the quantization band. In other words, 4 'substitutions will be found to minimize segmented NMR. wood

among them,? It is an indicator of the quantization band. The energy that minimizes the segmented N M R. This problem is to find s⑼ for the best approximation, that is,

584835 V. Description of the invention (13) The masking energy of the key band in the quantized band is even after the nose is selected. (9) No. ΛΑ FREE, in order to avoid the bits being allocated to the umbrella with noise higher than the noise = The band level of the mask level standard must be modified to match the segmentation criteria so that the band with negative NMR is rounded to 1. This means that the 々 2 杂 murmurs that have been converted must have a lower a. On the other hand, it is higher than the cover: the wave / miscellaneous phenomenon will cause a phenomenon. This phenomenon means that the relevant band will be thresholded by J2 ::: the zero band. These zero bands are quite obvious-for some levels of virtualization, they must also be limited to no more than the signal energy. Ting Cong, in short, in the allocation of conditional elements of the zero band and negative NMR, the noise must be multiplied by the mask threshold multiplied by _ ~, specifying the value of this hate. In the present invention, a quantizer of the third layer of MPEG is used to describe the% derivation step. From the standard of the third layer of MPEG, here, the simplified formula of the splitter and splitter is the non-uniform sentence volume of the second layer (3 Λ int (1〇)

Page 17 584835 V. Description of the invention (14) Among them, the quantization interval is

(ID _ -scales ΑΦ = 1. From the standard of MPEG, the formula of non-uniform quantizer can also be expressed as = mt xr. 2 scale-gain -0.0946 (12)

V

For each quantization band q, the ratio adjustment factor is scaleq = 1/2 (1 + scalefac_scale) {scalefacq + preflag · pretabq); scalefac _ scale is 0 or 1, and scalefacq is between 0 and 15. And the pre-amplified flag is Prefias, pretabq; for each small detail of the third layer structure of MPEG, the global gain is gciingr = \ / 2 (gl〇bal′aingr2 \ Q). If 0 · 0 9 4 6 is ignored, then (1 2) can be derived as / is = int xr.2 scale-gain \ int ^ scaleq-gaing

int _3, xr} Δ ,, (13) where the gap size is ^ {gaing, -scale q

Page 18 584835 V. Description of the invention (15) Next, the input signal (inputsigna 1) and the reply signal (rec ο nstructedsigna 1) xr are expressed by the following two formulas: xrt = ((iSi ^ s,) / isJb}, and χη = (is} AsJh} The quantization error of the non-uniform quantizer e will be equal to the difference between the input signal xr, and the response signal xr, {(is. + sl) Asjhf = (1 + isf εί) 3 isi 3 Δφ '— {isf Δφ (14) Let / (^,) = 0 + Yi,'. hu. Take the first derivative of Taylor expansion (Ty 1 er expans i on) + f (^) s Approximating this difference, and deriving the assumption that the quantization signal h and the quantization error of the non-uniform quantizer & are independent of each other, the expected value of the quantization error of the non-uniform quantizer A is as follows: E [ef]-^ a} e [is } £ 2] ^ ^ A} E [IS}] E [£!] (1 5} If the spectrum of the quantization band is uniform, the noise of the line can be the average energy of this quantization band; that is,

P.19 584835 V. Explanation of the invention (16) E (ef) = E (e2g) (16) Since E [ε2], the formula (15) becomes Li 12 E [el] a ¥} q1 E [\ XRlY] (1 7)

Bring Eq. (7) into Eq. (1 6) and get E [e2q] = Ka2M {g) B (q) (18) Finally, by defining 7; two <, ⑷, this global benefit and ratio adjustment factor The difference between them approaches 0 0.5 gaingr-scaleq «fl〇g2 ^ K'Tq IE [XRq] 5 or

gain-scale log2 — + l〇g2 K + 1 ° §2 ^ ^ ° §2 ^ f \ ^ q | ^ 4 (19) because the value of the ratio adjustment factor is between 0 and 16 and these quantized bands The minimum ratio adjustment must be zero, so the global benefit is

Page 20 584835 V. Description of the invention (17) gain 2 Max {gaingr-scaleq \, s q and get the ratio adjustment factors of all sub-bands. It can be seen that the global benefit changes with the bit rate related constant / C, and the ratio adjustment factor of each sub-band is different according to this mask threshold and the input signal. Upper and lower limits of the ratio adjustment factor

As mentioned earlier, bit allocation should be limited to non-negative NMR and zero band conditions. For the case of non-negative NMR, the noise level is set to the mask threshold; that is, 7; and / c = 1. This results in an upper limit relative to the global ratio adjustment. gaingr-Uscaleq =-(l〇g2 — + 1〇§2 σ1ί {ς) " " | ^ (20) That is, scalea < Όscale = gaingr-\ (l〇g2 — + l ° g2 σΜ ( ς) " | J) q 4 (21)

This will be adjusted based on the available bits. The lower limit can be derived from the zero-band condition. When the noise is greater than the signal energy, a zero-band situation will occur; that is,

Page 21 (22) 584835 V. Description of the invention (18) 0.5 Therefore, the lower limit of the ratio adjustment will be scaleq > Dscaleq = gaingr-^ \ 0gi E [\ XRq 0.5 (23) Figure 5 illustrates the * MpEG bit of the present invention, respectively. Yuan distribution process, and R is the average number of iterations of ratio control. The repeated allocation method shown in Figure 5 removes the inverse selection needed for quality control iterations. The present invention uses a factor greater than 3 to reduce the number of iterations of rate control. And, FIG. 6 illustrates the bit allocation record of the present invention compared to 丨 s0. Here, the present invention uses a perceptual pre-evaluation of audio quality (perceptual evaluation of audio qualHy ^ tJ!). This PEAQ system is the 々MAQ recommended by the ITU-R Working Group 10/4. I SOI was changed using Lame's termination conditions. ' ", T ° 1 S〇 is the original style (stereo mode) and psychoanalysis mode 2 is based on the second test and three-dimensional. (Blt -ser ;; i〇V ^^^ off, so in the experiment these two mechanisms Turn off. The objective difference equal grading method (objecUve deference g): ade, 0DG) is the output variable of the objective measurement% method. The value of 0DG should ideally be between 0 and ~ 4, of which 0 are not perceptible. The damage is 'and -4' corresponds to one which is considered to be very annoying: 584835 V. Description of the invention (19) Damage. As shown in Figure 6, the quality of the method of the present invention is better than the quality of the method proposed in this figure and text Well, the architecture used by PEAQ in the present invention is a basic version. This basic version uses an FFT-based ear model (FFT-based ear model). It uses the following model output variables: BandwidthRefB,

BandwidthTestB, Total NMRB, WinModDifflB, ADBb, EHSB, AvgModDifflB, AvgModDiff2B, RmsNoiseLoudB, ^^? 08 and 1 ^ 10 丨 37? Plant 3111658. Using an artificial neural network, the 11 model output variables are converted into a single quality index. The hidden layer of the artificial neural network contains three nodes (node). Figure 7 provides a list of test signals used in objective and subjective tests. By setting the same iterative termination conditions, such as the number of iterations, non-increasing noise ratio adjustment factor bands, suitable ratio adjustment factor tables, etc. [URL http://www.inp3dev.org/inp3], the I SO algorithm can Use the method mentioned by Lame to improve (that is, mp 3 encoding with the best. Λ). The two nested loops used for comparison are based on the iterative algorithm used by Lame. However, the above are merely preferred embodiments of the present invention, and the scope of implementation of the present invention cannot be limited by this. That is to say, all equivalent changes and modifications made in accordance with the scope of the patent application of the present invention should still belong to the patent scope of the present invention.

Page 23 584835 5. Within the scope of (20).

Ι1 · Ι1ΙΙ page 24 Brief description of the figure Figure 1 is a block diagram of the current audio coding _ encoding process. Figure 2 shows the a-bit allocation procedure in the 10 C program. Fig. 3a illustrates the steps of the sequence of the sound of the present invention _ "oblate code. Fig. 3b illustrates the process of the present invention. You-die ^-flow rate of the encoding process of the message rate. Fig. 3c illustrates the process of the present invention. Figure 4a illustrates the implementation architecture of Figure 3 ^ ^ of the present invention. Figure 4b and Figure 4c illustrate the implementation architecture of Figure 3h known Q ^ n BMb and Figure 3C. Figure 5 Explain separately when the present invention is not μ pp wide, ^ # MPEG bit allocation program uses different test materials

Material% Oral control and bit rate i allow birth, 1 ΛΑ T half control the average number of iterations. Figure 6 illustrates the objective scores of the present invention compared to the I-spider, 1 bU allocation method. Figure 7 provides a set of tests / tables for testing the subjective and subjective tests. 'These tests Signals are used by customers

Page 25 584835

Brief description of the drawing Figure description 101 Time / frequency converter 103 Other encoders 105 Quantizer 107 Variable-length encoder 109 Packaging unit 111 Auditory analysis mode 113 Bit allocator 115 Available bits 401 Converter 402 Quantizer 403 Variable Length Encoder 405 Parameter Estimator 407 Adjuster 408 Comparator 409 Packaging Unit 411 Cut 1¾ Frequency 413 Adjuster

Page 26

Claims (1)

584835 6. Scope of patent application 1. A method for digitally encoding and transmitting audio, including the following steps: (a) converting the input audio into a sequence of frequency samples to represent a spectral synthesis of the audio; (b) according to A bit allocation program that quantizes the frequency sample sequence into a quantized value. With a mask threshold value, the bit allocation program uses a parameter estimator to estimate the quantization parameter; (c) uses a symbol encoder to The quantized value is encoded to form encoded data, the encoded data comprising a plurality of bits; and (d) packaging the encoded data into a sequence of data according to a specific audio specification. 2. The method for digitally encoding and transmitting audio as described in item 1 of the scope of patent application, wherein step (b) is performed by a uniform quantizer or a non-uniform quantizer. 3. The method for digitally encoding and transmitting audio as described in item 1 of the patent application scope, wherein the symbol encoder comprises a variable length-encoder. 4. The method of digital encoding for transmitting and packaging audio according to item 1 of the scope of patent application, wherein for a quantized band, the parameter estimator in the bit allocation program uses a formula to calculate and adjust at least Page 27 584835 VI. Patent Application Range A corresponding global factor and / or a frequency band ratio adjustment factor. The formula is based on a constant mask-to-noise ratio. 5. The method for digitally encoding and transmitting audio as described in item 4 of the scope of the patent application, wherein the bit allocation procedure of step (b) further includes the following steps: According to a predetermined availability of the encoded data Adjust the global factor by the number of bits; and generate an upper limit and a lower limit of the band ratio adjustment factor corresponding to the global factor of the quantized band. 6. The method for digitally encoding and transmitting audio as described in item 5 of the patent application scope, wherein the upper limit is limited by a non-negative noise-to-mask ratio. 7. The method for digitally encoding and transmitting audio as described in item 5 of the patent application scope, wherein the lower limit is limited by the zero band. 8. The method for digitally encoding and transmitting audio as described in item 4 of the scope of patent application, wherein the band ratio adjustment factor is different for each sub-band according to the mask threshold and the input audio. 9. The method for digitally encoding and transmitting audio as described in item 4 of the scope of the patent application, wherein the global factor is dependent on a bit rate Page 28 584835 6. The scope of patent application is constant and varies. 10. According to the method for digitally encoding and transmitting audio as described in item 1 of the scope of the patent application, before step (d), there is an iterative rate control loop, which includes the following steps : (C 1) if the number of bits contained in the encoded data does not exceed a predetermined number of available bits of the encoded data, continue with step (d), otherwise continue with step (c 2); and ( c2) Adjust the quantization parameter and the size of a quantization interval to be used in step (b), and return to step (b). 11. The method for digitally encoding and transmitting audio as described in item 10 of the patent application scope, wherein step (b) is performed by a uniform quantizer or a non-uniform quantizer. 12. The method for digitally encoding and transmitting audio packaging audio as described in item 10 of the scope of patent application, wherein if the number of bits included in the encoded data exceeds a predetermined available number of the encoded data Number of bits, at least one corresponding global factor and one frequency band ratio adjustment factor will be adjusted, and the size of the quantization interval in this step (c2) will also increase. 13. A method of digital encoding for transmitting and packaging audio as described in item 1 of the patent application scope, wherein the symbol encoder comprises a variable length Page 29 6. Application for patent range encoder 14 15. The method for transmitting and packetizing 1 = code described in item 10 of t Γ t patent range, wherein step (b) further includes a : Step by step: J § Before Xuan frequency sample sequence 'for low bit rate audio coding, cut off high frequencies. The method of transmitting and encapsulating the cubic digital code described in the scope of patent application No. 14 of Qian Yue Λ includes the step of removing the second rate more than that shown in the table. Step-by-step steps. The frequency is reduced by one person-a cut-off frequency of 16.17. If there is at least one corresponding parameter in the formula of the patent application, the formula is as follows: Global factor coded; to generate the method for transmitting and packaging audio as described in item 10 of the quantization circle, where for a quantized band, the bit-score estimator uses a formula to calculate and adjust the global factor And / or one-frequency band ratio adjustment is based on a constant mask ~ parallel noise ratio. The method for transmitting and packaging audio described in item 16, wherein the bit allocation procedure of step (b) uses a predetermined number of available bits of data to adjust the band ratio corresponding to the global factor of the band Adjustment factor 584835 6. An upper limit and a lower limit of the scope of patent application. 18 The digital coding method for transmission and packaging as described in item 17 of the scope of patent application, the upper limit is limited by a non-negative, -pair-to-mask ratio. Complex noise 19. The method for digitally encoding a transport and packing person as described in item 17 of the patent application scope, wherein the lower limit is limited by the zero band ^. Hole 20. The f-bit encoding method for transmitting and packaging audio as described in item 16 of the scope of patent application, wherein according to the mask threshold value and the input, the band ratio adjustment factor is Different bands. 21β ^ The number 2 encoding method for transmitting and packaging audio as described in item 16 of the scope of patent application, wherein the global factor varies with a bit rate-dependent hanging number. In ·: The structure of a digitally coded stone horse that transmits and packs audio, the structure 22. Contains: a converter that converts the input audio ^ ^ ^ sequence to represent the Z-type spectrum of the audio combined with-frequency Sample a parameter estimator that estimates the number of inflammations to estimate the quantization parameters; H mask thresholds-a quantizer, which quantifies the frequency based on the quantization parameters 1 Page 31 584835 6. Patent application scope This sequence is quantized into a quantized value; a variable-length encoder that encodes the quantized value into coded data, where the coded data includes a plurality of bits; and a packaging unit , The packaging unit packages the encoded data into a data sequence according to a specific audio specification. 23. The digitally encoded structure for transmitting and packaging audio as described in item 22 of the scope of the patent application, the structure further comprising: a comparator that compares the number of bits contained in the encoded data with the A predetermined number of available bits of encoded data; and an adjuster for adjusting when the number of bits included in the encoded data exceeds the predetermined number of available bits of the encoded data The quantization parameter. 24. The structure of digital encoding for transmitting and packaging audio as described in item 23 of the scope of patent application, the structure further includes a high-frequency cut-off unit connected to the converter and the quantizer There is also an input port for receiving a cut-off frequency from the regulator. Page 32