TWM438765U - Audio quantization and de-quantization device - Google Patents

Description

M438765 Poor material (positive and negative values of voice signals). Therefore, after the first voice data is reduced by the down code converter 22, the new seven first voice data are obtained as 1i111〇11, 彳111〇〇11, 彳彳彳彳彳彳 、, 00000010, 00000101, 00001000, 〇〇〇〇1111 (5, ·13, small 2, 5, 8, 15). Next, the quantizer 230 quantizes the first speech data to generate a digital code, and the manner of quantization can be performed by a look-up table method. Although the voice data is positive or negative, in order to save the use of memory, a positive half-week quantization table (qUantizati〇n taWe) will be established. Before performing the quantification process, the symbolic bits representing the speech data belonging to the positive half cycle or the negative half cycle are recorded, and then the _ 4 tone data is taken to the absolute, and the side positive half cycle is used to quantize the speech data. The following lists a 5-bit table, and uses the quantization table 1 to quantize the first speech data (_5, _13, q, 2, 5, 8, 15), such as: record the speech data _5, · 13, μ symbolic bits The yuan is] and the 2, 5, 8, and 15 paying bits are 〇, and then all the data are taken as absolute values (5, a, 1, 2, 58, 15), 5 of which are based on the quantization table 1 The preferred index code is 3, and the corresponding quantized digit code is 00011, 13 according to the quantization table, the best index code is 7, and the corresponding quantized digit code is 00111, after the fifth bit is added with the sign bit. The digital data is 1〇〇11,1〇111. _ Yu's absolute value of the first-speech data (-5, -13, -1, 2, 5, 8, 15) according to Table 1 to get the index code (3, 7, 1, 2, 3, 4, 7) ), the corresponding binary digits are (0001), 00111, 00001, 00010, 0001Ί, 〇〇1〇〇, 〇〇111), and the digit of the digit after the fifth digit is replaced by the sign bit is (loon, mm, 10001, 00010, 〇〇〇11, 〇(1) 〇〇, 00111). Among the 13 tables, the closest one is 15, so select it as the corresponding table value. Table 1 5 Digital Code Index Code Quantization Table Code 00000 0 0 00001 1 Ί - " 00010 2 2 00011 3 "5 ' 00100 4 8 00101 5 ~9 00110 6 10 ' 00111 7 15 ' 01000 8 30 01001 9 " 40 " 01010 10 "45 " 01011 "11 50 01100 12 ~55 " ' 01101 13 Bu 60 01110 14 90 01111 15 1〇〇' ' 11111 31 Frame switching control code M438765 In practical applications, To reduce the quantization error, multiple quantization tables are often used to correspond to different dynamic range speech data. Please refer to Figure 4, the digit code 6〇〇 is composed of symbol data 612 plus digital data 614. The concatenated speech data is multiple The digital code consists of 6 而 and the 7 digits of the code 600 has a total of 35 bits. The serial voice data usually also contains the frame marker data 6 〇 6 (Frame Header) 'frame marker data 606 records the current frame The index corresponding to the optimal quantization table, and the frame switching control code (generally a special code that uses the same number of bits as the voice encoded data but does not repeat). The data length of the frame mark data 606 depends on the number of quantization tables. Example 'When 8 quantization tables are used, the frame mark data 606 needs 3 bits to correspond to the index of the best quantization table, and 32 quantization table time frame mark data 606 needs 5 bits to correspond to the index of the best quantization table. The length of the frame tag data is an example (the index of the 5-bit frame switching control code plus the 5-bit optimal quantization table), and the size of the serialized speech data is 35+10=45 bits. On M438765 M438765

V. New type description: [New technical field] This creation is a kind of quantization device, especially for an audio quantization device. [Prior Art] The speech signal is originally analog signal, which will be distorted after digitization and compression. In terms of compression ratio, signal distortion is large, but the required transmission code rate is low. Therefore, in the case of insufficient transmission bandwidth, 'below the cake under the rhyme call content', the limbs with high heights are usually selected. If there is no problem with the transmission bandwidth, it is generally better to use the G711 protocol with less signal distortion. Please refer to FIG. 1 , which is a prior art speech encoding and decoding system diagram, including: a voice input signal 100, a voice encoder 200, a memory 300, a voice decoder 400, and a voice output signal 500. Among them, the voice input signal is a true sound of the segment, which is an analog signal. For example, if the speech encoder 200 is a 16-bit mono, if the sampling rate is 8 KHz per second, the data amount is 12 sec per second, and when the speech input signal is input to the speech coder 2 〇〇, After the voice input, it will be sampled in the 128kbit channel surface of the Lin seconds. After the dust is encoded, it will be stored in the memory training +. The vocoder 2 (five) is actually a compressor. In practical applications, in order to reduce the usage of memory 3 (5), the tree generally compresses the 16 ❾ ❾ voice data into lower resolution data (such as 5bit or 4bit) and exists in the memory to effectively reduce the memory. 300 usage. Finally, the speech decoder 4 解读 interprets the lower resolution data after the memory 300 has been compressed and converts it into a mono speech material' having one and converts it into a speech output signal 500. Next, the reference to Fig. 2A is a detailed block diagram of the prior art speech coder. The voice knitting machine 200 includes: an analog digital converter 21 〇, a down code converter 22 量, a quantity 3 M438765 I 月 month f day double 2 230 and data editing · 24Ge, the digital converter plus receiving class The first voice data converted into a digit by inputting a signal. The subtractive conversion $22〇 connection analog digital converter 210' de-codes the first speech data. The quantizer 23 is coupled to the down-converter, receives the first-speech data and quantizes to generate a digital code containing the symbol data and the digital data. The data encoder 240 is coupled to the quantizer 23A to receive at least one digit to generate 'a series of speech data. In the other previous embodiment, please refer to FIG. 2B, in which an external first memory 110 has stored digital first-speech data, wherein the down-converter 22〇 enters the first voice. Reduce the code. The quantizer 230 is coupled to the down-converter 22A, receives the first speech data and quantizes to produce a-digit code containing symbol data and digital data. The data encoder 24 〇 connection quantizer 230' receives at least one digital code to generate a concatenated speech material. An example is given below: Next, please refer to Figure 3 for the speech data converted by the analog-to-digital conversion S 210, which contains the first speech data of 7 digits: 11111〇11〇〇〇1〇〇〇, 1111001100001000. 1111111100001000 . 0000000100001000 . # 0000010100001000,〇〇〇〇1〇〇〇〇〇〇〇1〇〇〇,〇〇〇〇1 i ] 1〇〇〇〇1 〇〇〇. The reduction mother converter 220 converts the seven I6bit first voice data into an 8-bit first voice data with positive and negative signs. The down code converter 22 directly removes the first bit to the eighth bit of the first voice data of the 16bjt, and retains only the data of the first to the first voice of the original voice. This is the new first voice material. Therefore, the first speech data only leaves 8 positive and negative symbols, and the data range is -128 to 127. Among them, the first bit to the seventh bit represent digital data (the amount of voice signals), and the first _ represents the symbol 4 M438765 丨 (four) month K correction is the original 7-note 第 first voice data a total of 112bit, after the reduction of the conversion of coffee The data of t is converted into the first voice data of only 7 8 bits and 56 bits. The quantization result of the re-quantizer 230 converts each 8-bit data (form code) into the index data of the lion. Finally, the data amount of the 7-bit 5-bit is 35 bib. From this, we can know that the amount of data of the original buckle (five) has passed. Reduced code to Lin 220 and quantified (four) G, and finally Wei only 35bit raw material. After that, add 1 〇 of the frame mark data 606, and now the total amount of data is 45. It can be known from the above prior art that when performing speech quantization coding, each quantized digital code _ contains a symbol code and a digital code, and each digital code contains a symbol code, which inevitably wastes the stored data amount. Therefore, in order to reduce the amount of data that is wasted, it is necessary to propose a new architecture to reduce the amount of data stored. [New Content] ▲This creation provides a kind of audio quantization and coding device, which uses memory to encode signals. The memory records a plurality of digital first speech data, including: signal divider, quantizer and information. - the signal number, reading the first voice data of the record and performing a plurality of zero-crossing enchantment judgments to sequentially generate a plurality of first symbol data, and cutting the digital first voice data into a plurality of frames; The quantizer is connected to the signal divider, and receives the digital-speech data and the first symbol data corresponding to each frame, and quantizes the digital-speech-quantization corresponding to each received frame After the generation of complex (four) - digital data, and quantized according to the frame: = miscellaneous should be generated - the first frame of the mark data; and - data compiled, connected to the quantizer and the signal L 'receiver quantizer generated for each frame The first digital data, the first symbol data, the eighth frame; the data is encoded and encoded into a first encoded data string. This creation also provides a kind of audio quantization decoding device, using a memory for signal solution M438765 T Zhuang t times 'memory record has - the second series, including: data decoder and inverse Wr-Bec decoder 'Connecting the memory, reading the second encoded data string and decoding to generate a plurality of first decoding tasks, each _ two decoding _ includes: - the second frame label _, a first symbol, gamma inverse峨 职 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , This creation proposes a more efficient round. In the existing encoding device, each quantized digit code contains a facet, and the invisible towel will have more (4) stored volumes. This creation only proposes to use only one symbol code, and a series of digital threats, which can reduce the amount of stored data under the premise of turning the resolution', and can achieve a significant increase by slightly increasing the amount of data under the premise of pursuing sound quality. The effect of sound quality. In order to make the above and other objects, features, and advantages of the present invention more obvious and easy to understand, the following are a few of the best _, and the age is closed, and the detailed cutting is as follows: [Embodiment] Please refer to the 5A 'Department, the embodiment of the audio quantization and editing group 2 of the creation, including: quantization ϋ 230, signal sub-office 25Q and (four) encoder 2. signal sub-turn 25 〇 includes a temporary storage H 251, signal segmentation The device 25Q reads the stored digital first-speech data from the first memory 11〇 and performs a zero-crossing point determination to sequentially generate a plurality of first symbol data 612′ and cut the digital first speech data into a plurality of Frame. The quantizer 23 is connected to the signal divider 250, and sequentially divides the plurality of digits of the first arpeggio and the corresponding first symbol data corresponding to the frame #cut by the signal according to the signal. The plurality of digits of the first voice data corresponding to the pivot is _--corresponding to generate the first digital data, and the first frame marker data 606 is generated according to the 8 M438765 result quantized by the frame. The data encoder 240 is coupled to the quantized number divider 250 ′ to receive the first digital data, the first symbol data and the frame marker data and encodes the first encoded data string into a first encoded data string, each first encoded data string comprising the first Frame mark data, first symbol data, and multiple first digital materials. Thereafter, the first encoded data string is stored in the second memory 310. In practice, the first memory 110 and the second memory 31 can be different blocks of the same memory. Next, please refer to FIG. 5B, which is an embodiment of the audio quantization coding module 2〇〇B of the present invention. The main difference between FIG. 5B and FIG. 5A is that the plurality of digit-the speech data corresponding to the frame read by the quantizer 23 in FIG. 5B is directly read from the first memory 11 . Take it back and quantify it. In FIG. 5A, the quantized n-bit data corresponding to each frame read by n 23Q is read by the first memory 11 () and then placed in the register of the signal divider 25 251 t, and then the digital voice data is read by the register 251 of the signal divider 25 to perform quantization. Next, please refer to Section 6A, which is an embodiment of the audio quantization coding module 2〇〇c of the present invention. In the embodiment of Fig. 5A, a down code converter 22 is added. The down code converter 220 is connected between the first memory 110 and the signal distribution room 25〇, and performs all the first voice data temporarily stored in the first memory training frame to be decoded, and then stored in the temporary memory. The memory (5) is in the middle. Next, Fig. 6B is an embodiment of the audio quantization coding module 2〇〇d of the present invention. In the embodiment of Fig. 5B, a down code converter 22 is added. The down-converter is connected between the first-memory no and the signal splitter 25A and the quantizer 23A, and performs the action of subtracting the first voice data stored in the memory of the first M438765 jms' memory. Then, the quantizer 230 includes a control unit and a vector unit. The control unit calculates the quantization error according to all the first voice data or the subtracted first voice data corresponding to each frame. For the selection of the quantization table, and generating the frame mark data, the vector unit receives the first voice data and performs the search of the quantization table to generate the digital data. The code reduction converter 220 of the present invention is not limited to only 16 bits and is reduced to 8 bits. It can also be changed from 16bit to i〇bit, or 24bl.t can be reduced to i2bit.

The system is designed to be selected. I The deduction can be a reduced data, for example, reduced to 丨丨 or _t. In some embodiments of the present invention, when the plurality of first voice data is a negative value, the symbol data representing the positive and negative symbols in the digit code 600 is omitted and integrated into the frame mark data, and a number of_ The data 614, which only contains the content representing the sound data, does not include the positive and negative of the Beixun 'as shown in Figure 7. Thus, the number of bits τ of the digit code 600 in the prior art can be reduced, for example, originally reduced to gamma to reduce the amount of data. Or, increase the number of bits of the digit code 6〇0, for example, from the _ of the original digit code, which only accounts for the ## data of the rhyme. In this creation, the digital data of _ is added to improve the encoding. The resolution of the data. - month reference to Figure 7 is a schematic diagram of the data structure of the serialized voice data of the creation. The individual tandem θ bead material 620 includes the frame mark data 6〇6, the symbol data M2 and the eve pen number sub-data state. In other words, each string of voice data is spliced to the frame tag data coffee and symbol data 612, and finally the next frame tag is 枓6〇6. The length of the concatenated speech data 622 depends on the number of points of the digital data among the lengths of the two zero crossing points, and also 10 M438765 food P 'frame mark data 606 + symbol data 612 + points of two zero crossing points长度 Length (eg '4bit or 5bit). In other words, the positive or negative sign of the symbol data is used to judge the generation of the zero crossing point by using the signal divider, and the judgment of the zero crossing point is based on the data of the two consecutive first voice data. Depending on the change. When the signal splitter 25 of the present creation receives the continuation-positive-speech data and the -negative-speech data (whenever it appears first), it can be judged that there is a zero crossing point, and the signal divider 250 will Generating symbolic data 612 representing the zero crossing point to provide a data compilation

Coder 240. The positive sign of symbol data 612 can be represented by 〇, and the negative sign of symbol data 612 can be represented by 1. For example, the first-the first voice data is A, the second first voice data is B, and when A<0' and B>=〇, the signal splitter 250 generates the symbol data 612 as "〇,, ( No.), that is, the case where the first voice data is negatively positive, that is, the subsequent first voice data will be positive; when A>=〇' and B<0, the signal splitter 25〇 generates symbols The data 612 is Ί" (negative sign), that is, the case where the first-speech data is positively negative, that is, the first language that appears subsequently

The audio data will all be I. The above is only the implementation of the zero crossing point judgment of this creation - the embodiment, this creation does not advocate this way. Example 1: In this embodiment, Wei Ming's digital data encoding length representing the voice signal is fixed to _. There may be two or more quantization tables corresponding to the frame mark data 6〇6. At this time, the frame mark data 606 shall use at least one bit to indicate which table to use. For example, when only two quantization tables are used, the frame mark data 6〇6 can set the table values corresponding to Table 2 (the first table in this embodiment) to be, 〇,,, and Table 3 The table value of (the second table in this embodiment) can be set to "1,". When five quantization tables are used, at this time, the frame mark data 6 (5) requires 3 bits, and the table 11 M438765 丨 0 On the tenth day of the month, f, the value is 〇〇Q, 0〇1, Q1Q, 011, 1(X), etc. The number of quantified tables that can be mapped by this creation, ^ a table, or a plurality of tables. In the case of a quantization table, the most suitable table will be determined according to the size of the quantization error of the data. Next, please refer back to Figure 3, which is a sequence of multiple first speech data (7, _12, • 1, 3, 5, 8, 15, 8, 5), which has two zero crossing points. It is assumed that there are 8 tables available for use. After comparison and comparison, Tables 2 and 3 are the most suitable tables. They are applied to the sequence of (_6, -12, -1) and the sequence of (3, 5, 8, 15, 8, and 5). The selection of suitable quantization tables is well known to those skilled in the art and will not be described again. First After the first zero crossing point, first encounter (·6, _13, _υ, first take the negative symbol data 612 to 1 in the front, and then (-6, _13, take the absolute value into (6, 13, ^, firstly, the frame mark data 606 of Table 2 is first set to _, and finally, after looking up the table 2, the index code is (4, 8, 1), and finally the sequence obtained by the digital face of the table 2 is ( (1) (8), 1〇〇〇, 0001). After that, the value of the frame mark data 6〇6 and the value j of the symbol data 612 are added. The finally obtained serial voice data 620 is (〇〇〇, 1). 〇1〇〇, 10〇〇, 〇〇〇1). Then 'after the second zero crossing point', the frame marking information of Table 3 is set to Μ6, and the last look of Lu (3, 5 , 8, 15, 8, 5) 'This symbol data 612 is 0 for a positive value, and then for the mapping table 3 'to find the closest form code, which is well known in the art, you can get the index code is 〇, 2, 3, 5, 3, 2) 'The code of the digital data of the final re-enactment table 3 is (〇〇〇1,〇〇1〇,〇〇1^, 0101,0011, 0010). After that, add The value of the message box 6〇6 of the upper frame is 〇〇1 and the character The concatenated speech data obtained by the value 0 of the data 612 is (001, 〇, 〇〇〇 1, 〇〇1 〇, 〇 () 11, (1) (7), 001, 0010) » Next 'Please refer to Figure 7' and finally Combine positive and negative frame-encoded data and add 12 M438765

—— · W IVW ixjx^yj - UUU1 '1111' 001 〇 0001, coffee, 〇〇11, 〇1〇1, 001, 〇_ complete speech data sequence string. In the embodiment described above, by adding a symbol data of one bit to the start code, the sign bit of the positive and negative digit code can be followed by the guest. When the number of points per two zero crossing points is increased, the amount of data that can be omitted is increased. From another point of view, taking 4-bit quantization coding as an example, the coded quantization result of the prior art includes one symbol data in every four bit digit code, and the numerical data has only 3 bits. In this creation, in the 4bit architecture of the cum, the original 4 bits can all be used as numerical data. In the application of such a look-up table method, the result of the look-up table, that is, the resolution of the audio signal coding can be nearly doubled on the basis of the same amount of data, and the quality of the voice signal coding is greatly improved.

Table 2 Digital data index code table code 0000 0 0 0001 1 Ί ~ 0010 2 2 0011 ~ 3 ~ 4 0100 4 ~ 6 0101 " 5 ~8 -- 01Ϊ0 " 6 9 01ΪΪ " 7 1 1Ϊ 1000 8 12 100Ϊ ----- 9 ~14~~~~' 1010 10 16~ 1011 11 17 '~~-- ΙΤόδ " 12 19~ 1101 13 ~20 Ιίΐό ~~ " 14 22 13 M438765

Frame switching control code

^ ------- Frame switching control code Hi [disclosed, is the editor. Score. Receiver, please refer to Figure 8, which illustrates the creation of the Xi Li 'month t gas spear, soldiers description. Vertical: Once the quantized decoding touches, the serialized speech encoded by this _ can be solved. The 曰1 里 解码 decoding module includes: data decanter 4 deduction and inverse quantizer 42 〇. The data decoder 410 reads the following 31Q _ two W syllabus Meisheng plural second correction data strings, each of the second decoded data strings includes: - the second frame mark data, a second symbol data, a plurality of second a number of data; and - the amount of silk coffee, connected to the data decoder 'connected to the second decoding identification, and according to the second pivot mark, the second symbol of the value of the second material to carry out the second number The inverse quantization of the data sequentially generates a plurality of digits of the second voice data and is then stored in the third memory 510. 14 M438765 丨· \ 1^11 In practice, the second memory 310 and the third memory 510 may be different blocks of a phase memory. - For example, after the decoding of the decoder 41Q, the speech data sequence string of the foregoing example 1 can be obtained (1111, 000, Ί, 0100, ] 〇〇〇, 〇〇〇 1, ii i, 〇〇 1, 〇 , 〇〇〇 1, 〇〇 1 〇, 〇〇 11, 〇 1 〇 1, 0101, 〇〇 1 〇). Next, first remove the frame switching control code 1111 of the speech data sequence string, and then inversely quantize 15 420 to take the table 2' of the matching data and take the symbol data as 1, indicating that the next digital data will be _ negative value, and then The index code obtained by Table 2 is (4, 8, 1) 'The final mapping table 2 can obtain the table code (6, 12, 1), and since the symbol data 612 is negative, the obtained speech data is obtained. For (.6, -12, .... the same 'in the second sequence, first remove the frame switching control code of the speech data sequence string, and 001 represents the second quantization table (Table 3), the symbol The data M2 is 〇, the index is exhausted as (2, 3, 4, 8, 4, 2), and finally, corresponding to Table 3, the form code (5, 8, 12, 25, 12, 5) can be obtained individually. The meter will restore the data to a plurality of _voice data (_5, 12 _1, 5, 8, 12, 25, 12, 5). Finally, 'save multiple _ voice data to the third memory 51 In the actual spear force, the first § memory 110, the second memory 31 〇 and the third memory (10) may be different blocks in the same memory. If you are a fan of the same ship, you will be able to make a certain change and refinement without any deviation from the spirit and scope of this creation. Therefore, the scope of patent protection of this creation must be seen in this book. The scope of the patent application attached to the specification shall prevail. [Simple description of the diagram] 15 I, q month, and the first picture are the previous speech coding and decoding system diagram (prior art). Figure 2A is First embodiment of the functional block diagram of the prior speech coder (Prior Art) Fig. 2B is a second embodiment of the functional block diagram of the previous speech coder (prior art); Fig. 3 is a previous analog digital sampling diagram (Prior Art); Figure 4 is a previous concatenated speech data map (previous technique); Figure 5A is a first embodiment of a functional block diagram of the speech coder of the present invention; The second embodiment of the functional block diagram of the speech encoder; the sixth embodiment is the third embodiment of the functional block diagram of the speech coder of the present invention; Lu 6B is the functional block diagram of the speech coder of the present invention. Fourth implementation Figure 7 is a diagram of an embodiment of the serial voice data of the present creation; and Figure 8 is a functional block diagram of the speech decoder of the present invention. [Description of main component symbols] 100 voice input signal 110 first memory 200 speech encoder 210 analog to digital converter 220 down code converter 230 quantizer 240 data encoder 250 signal divider 251 register 300 memory M438765

310 second memory 400 speech decoder 410 data decoder 420 inverse quantizer 500 speech output signal 510 third memory 600 digital code 606 frame mark tribute 612 symbol data 614 digital data 624 first encoded data string 626 first Coded data string 700 audio codec

17

Claims (1)

M438765 VI. Scope of application for patents: There are a plurality of digits of the first voice data, including: the memory records a signal sequence, reads the light material-voice cake and performs Wei zero crossing point judgment to sequentially generate a plurality of first symbol data, and The digital-speech data is cut into a plurality of frames; a quantizer is connected to the "branch" to receive the digital-speech data corresponding to each of the frames and the first symbol data, and each The first frame tag data is generated corresponding to the result of the plurality of digital data frames corresponding to the number of digits received by the second digit-voice dragon after the quantization; and - the data Edit · 'Lianzhu 4彳 old no signal division, the Wei quantizer generates the first digital data, the first symbol data and the first frame marker data for each frame and encodes it into a first Encoded data strings. 2. The audio quantization and coding apparatus according to claim 1, further comprising: a subtractive code converter ??? connected between the memory and the 峨 splitter, and performing a code reduction operation on the first voice data. 3. The audio quantization and coding apparatus according to claim 1, further comprising: a down code converter coupled between the memory and the signal divider and the quantizer, and performing down coding on the first voice data The action. 4. The audio quantization and coding apparatus according to claim 1, wherein the first coded data string corresponds to the frame and includes the first frame mark data, the first symbol data, and the frame corresponding to the frame The first digital material. [0085] The apparatus of claim 1, wherein the quantizer reads the digits of the first voice data corresponding to the frame of the video frame from the memory. The audio quantization device of the present invention, wherein the signal splitter includes a temporary save, the temporary storage is used to store the digital first voice data corresponding to the frame, and the quantizer is The digits of the first voice data corresponding to the frame read by the second reading are read from the register of the signal divider. The audio quantization and coding apparatus according to claim 1, wherein the quantizer uses one or more quantization tables to calculate the quantization error of each of the 帛-pins (four) 每 wires of each of the martial arts and to calculate the quantization error. The quantization table of the error, the sharp amount of the wire to get the first digital data. 8. The audio quantization coding apparatus according to claim 1, wherein each of the zero crossing points is multiplied by two consecutive first speech data, and when it is negative, the zero crossing point is established. 9. The audio quantization coding apparatus of claim 1, wherein the first symbol data system represents a positive value or a negative value. 1〇_-A kind of audio quantization decoding device, using a memory for signal decoding, the AVI memory record has a first encoded data string, comprising: - data decoding H' is connected to the memory, and the first encoding is read Serializing and decoding to generate a plurality of second decoded data strings, each of the second decoding strings comprising: _ second frame marker data, - second symbol data 'plurality of second digital data; and - inverse quantizer 'Connecting the data solution, receiving the second decoded f-string, and performing inverse quantization of the second digital data according to the second frame mark data and the value of the second match material to sequentially generate a plurality of digits Second voice data. </ RTI> </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; The two symbol data 'the second digital data corresponding to the quantization table corresponding to the second frame mark data corresponding to the second voice data. 20