JP4978539B2 - Encoding apparatus, encoding method, and program. - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve coding efficiency of data, while securing the quality of the data. <P>SOLUTION: A modified discrete cosine transform (MDCT) section 23 performs frequency conversion upon digital signals of blocks, respectively, to produce a MDCT coefficients. A time-order permutation section 30 time-sequentially permutes the MDCT coefficients, regarding the same frequency. A coefficient VQ section 31 uses a codebook 51, to vector-quantizes the group of MDCT coefficients for each frequency and arranges obtained indexes in the order of frequencies to produce a data stream of transform coefficient indexes. An entropy-coding section 32 codes these data streams and information about a flag. A data deletion section 34 calculates the degree of importance for the group MDCT coefficients for each frequency. The data deletion section 34 compresses the data streams of the transformation coefficient indexes, based on the degree of importance and changes the flag, to thereby compress the data to be coded. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

The present invention is an encoding apparatus for encoding a digital signal, the encoding method of encoding a digital signal, a program to be executed by a computer to perform 及 beauty digital signal processing.

  A speech processing apparatus that performs speech coding based on human auditory characteristics and decodes the coded data is disclosed (for example, see Patent Documents 1 to 4).

JP-A-2005-128404 JP 2006-119363 A JP 2006-259517 A JP 2006-262295 A

  This type of speech processing device is also used for speech of words that are pronounced in a language dictionary such as an electronic dictionary device. A speech processing device for a language dictionary must secure a data rate of about 16 kbps while ensuring sufficient sound quality.

The present invention has been made in view of such circumstances, provide while maintaining the quality of the data, the encoding device which can improve the coding efficiency of the data, marks Goka method, the 及 beauty program The purpose is to do.

In order to achieve the above object, an encoding apparatus according to the present invention includes a dividing unit that divides a digital signal having a predetermined time length into a plurality of blocks, a frequency conversion of the digital signal of each block, and a first conversion. A band that divides a frequency conversion unit that generates a coefficient group for each block and a first conversion coefficient group generated by the frequency conversion unit into a plurality of small frequency bands that have a wider bandwidth as the frequency increases. By searching for the maximum value of the absolute value of the first transform coefficient belonging to the small frequency band for each of the small frequency bands, and by dividing the searched maximum values in order of frequency, A maximum value search unit to be generated for each block, and a maximum value sequence of each block is vector-quantized using a maximum value sequence codebook, and the obtained indexes are arranged in time series A maximum value sequence vector quantization unit that generates a data sequence of a maximum value sequence index, and the index of the maximum value sequence of each block obtained by the maximum value sequence vector quantization unit is inverted using the maximum value sequence codebook. A division unit that quantizes and divides a first transform coefficient group belonging to each small frequency band of each block using an inverse quantization value corresponding to the block and using the inverse quantization value of the small frequency band And generating a second transform coefficient group for each frequency by rearranging the transform coefficients of the same frequency included in the first transform coefficient group of each block divided by the division unit in time series A time series rearrangement unit that performs vector quantization on the second transform coefficient group of each frequency using a transform coefficient codebook, and arranges the obtained indexes in order of frequency. Based on the importance of the transform coefficient vector quantization unit that generates the transform coefficient index data string and the second transform coefficient group of each frequency, the transform coefficient index data string is compressed, A data compression unit that generates information on a flag indicating whether or not the transform coefficient group of 2 is an encoding target, a data sequence of the maximum value sequence index generated by the maximum value sequence vector quantization unit, and the data compression An encoding unit that encodes the information about the flag generated by the unit and the compressed data string.

The data compression unit further includes a code amount determination unit that repeats the determination as to whether or not the code amount of the data encoded by the encoding unit is smaller than a target code amount until the determination is positive. When the determination of the code amount determination unit is negative, the data sequence of the transform coefficient index is compressed by deleting the second transform coefficient group from the encoding target in ascending order of importance. The information on the flag is generated, and the encoding unit, the data sequence compressed by the data compression unit until the determination of the code amount determination unit is affirmed , the information on the generated flag, May be encoded.

  In addition, the data compression unit is configured such that the code of the encoding unit is in descending order of importance until the code amount of the encoded data is less than the target code amount and is close to the target code amount. Selecting a frequency frequency to be encoded, compressing a data sequence of the conversion coefficient index with a second conversion coefficient group corresponding to the selected frequency frequency as an encoding target, and generating information on the flag Also good.

  In addition, the data compression unit forms a flag sequence by arranging the generated flags in order of frequency, and based on the formed flag sequence, a continuous number sequence in which the same value continues in the flag sequence. It is good also as producing | generating as the information regarding the said flag.

  In this case, when the number of consecutive consecutive identical values is equal to the upper limit value in the flag string, the data compression unit, between the number of consecutive numbers in the number sequence and the next consecutive number, It is also possible to insert 0.

  The data compression unit may insert 0 at the head of the number sequence when the flag sequence starts from 1.

  In addition, it is determined whether or not the sum of the code amounts of the encoded data encoded by the encoding unit with respect to the series of the plurality of digital signals having the predetermined time length is smaller than the entire target code amount, An overall code amount determination unit that repeats until the determination is affirmative, and a transform coefficient index corresponding to the second transform coefficient group that minimizes the degree of importance overall when the determination of the overall code amount determination unit is denied Are excluded from the transform coefficient index data string, and an adjustment unit that changes the flag corresponding to the excluded second transform coefficient group to a value indicating that it is not an encoding target; and the transform coefficient index is excluded. And a re-encoding unit that re-encodes data related to the digital signal having the predetermined time length in which the flag is changed.

  According to the present invention, encoding efficiency can be improved while ensuring the quality of data.

<< First Embodiment >>
Next, a first embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 1 shows a schematic configuration of a sound processing apparatus 1 according to the present embodiment. As such a voice processing device 1, for example, a mobile phone or a terminal device such as an electronic dictionary is assumed.

  The voice processing device 1 includes a voice input / output device 11, a storage device 12, a ROM 13, a RAM 14, and a CPU 15. These are connected via an internal bus.

  The voice input / output device 11 converts the input voice into a digital signal. For example, the voice input / output device 11 generates the digital signal Sound0 by sampling the input voice at a sampling frequency of 16 kHz and quantizing the voice by 16 bits. Further, when a digital signal is supplied, the voice input / output device 11 outputs a voice corresponding to the digital signal.

  The storage device 12 stores data in which the digital signal Sound0 generated by the audio input / output device 11 is encoded. The storage device 12 also stores data necessary for decoding the encoded data. These data will be described later.

  The ROM 13 stores various data such as program codes necessary for processing executed by the CPU 15. The RAM 14 stores data necessary for processing executed by the CPU 15.

  The CPU 15 executes processing according to the program code stored in the ROM 13. The encoding unit 16 and the decoding unit 17 are realized by execution of processing by the CPU 15.

  The encoding unit 16 encodes the digital signal converted by the voice input / output device 11. As shown in FIG. 2, the encoding unit 16 includes a DC removing unit 21, a framing unit 22, an MDCT (Modified Discrete Cosine Transform) unit 23, a normalizing unit 24, and a band division. Unit 25, maximum value search unit 26, maximum value sequence vector quantization (VQ) unit 27, maximum value division unit 28, quantization unit 29, time order rearrangement unit 30, coefficient vector quantization ( VQ) unit 31, entropy encoding unit 32, code amount comparison unit 33, data deletion unit 34, and codebooks 50 and 51.

  As shown in FIG. 3, the DC removal unit 21 deletes a direct current (DC) component Xdc from the digital signal Sound0 of the sampling period Ts generated by the audio input device 11. The reason why the direct current component Xdc is deleted is that the direct current component Xdc is irrelevant to the sound quality. The DC removal unit 21 can be realized by a high-pass filter, for example. In the following equation (1), an example of the transfer function H (z) of the high-pass filter is shown.

ＤＣ除去部２１は、直流成分Ｘdcが除去されたデジタル信号Sound1を記憶装置１２に格納する。そして、ＤＣ除去部２１は、フレーム化部２２に処理開始を通知する。

The DC removing unit 21 stores the digital signal Sound1 from which the direct current component Xdc has been removed in the storage device 12. Then, the DC removal unit 21 notifies the framing unit 22 of the start of processing.

  Upon receiving this notification, the framing unit 22 reads the digital signal Sound1 stored in the storage device 12 and divides it into frames. FIG. 4 schematically shows a relationship with a frame signal (a digital signal having a predetermined time length) generated by frame division of the digital signal Sound1. As shown in FIG. 4, each frame signal partially overlaps with the immediately preceding frame signal. Let this overlap time be T. FIG. 4 also shows an MDCT block (1MDCT) which is an MDCT processing unit described later. The time length of each MDCT block is 2T. That is, the MDCT block overlap time is half of its length. The overlap time between frames is also a time T that is half the time length of the MDCT block. In this way, the time interval of each MDCT block is constant over a plurality of frames (all sections of the digital signal).

  In FIG. 4, although there are four MDCT blocks per frame, the following description will be made assuming that N blocks (N is an integer of 2 or more) are generated per frame. The framing unit 22 stores a plurality of frame signals generated by frame division in the storage device 12 in units of frames. Then, the framing unit 22 notifies the MDCT unit 23 of the start of processing.

Upon receiving this notification, the MDCT unit 23 serving as the dividing unit and the frequency converting unit performs frequency conversion on each frame signal read from the storage device 12. More specifically, the MDCT unit 23 divides the frame signal read from the storage device 12 into a plurality of MDCT blocks, performs frequency conversion for each block, and MDCT coefficient X _k (k: code indicating frequency). Is calculated for each block. The MDCT unit 23 calculates the MDCT coefficient X _k using the following equations (2) and (3). MDCT coefficient X _k of one MDCT block in a single MDCT is calculated. The tap length M of the MDCT coefficient X _k is ideally 512 taps.

ＭＤＣＴ部２３で生成された、各ＭＤＣＴブロックのＭＤＣＴ係数Ｘ_k（すなわち第１の変換係数群）は、記憶装置１２に格納される。ＭＤＣＴ部２３は、正規化部２４に処理開始を通知する。

The MDCT coefficient X _k (that is, the first transform coefficient group) of each MDCT block generated by the MDCT unit 23 is stored in the storage device 12. The MDCT unit 23 notifies the normalization unit 24 of the start of processing.

Upon receiving this notification, the normalization unit 24 reads the MDCT coefficient X _k of each MDCT block from the storage device 12. The normalizing unit 24 normalizes the MDCT coefficient X _k of each MDCT block in units of frames. Normalization unit 24 obtains the maximum value gain of MDCT coefficient X _k is separated by dividing each MDCT coefficient X _k at this maximum gain, normalization is performed.

More specifically, the normalization unit 24 obtains the maximum value gain of the MDCT coefficient X _k in the frame using the following equation (4).

図５（Ａ）には、各ブロックにおける補正係数Ｘ_kの最大値Ｘｍａｘ_iが示されている。この最大値Ｘｍａｘ_iが、ブロック間で比較され、最も大きいＸｍａｘ_iが、最終的な最大値gainとして求められる。

FIG. 5A shows the maximum value Xmax _i of the correction coefficient X _k in each block. This maximum value Xmax _i is compared between the blocks, and the largest Xmax _i is obtained as the final maximum value gain.

  Next, the normalization unit 24 normalizes the MDCT coefficient using the following equation (5).

In FIG. 5 (B), an example of the normalized MDCT coefficients Xn _k is shown. By this normalization, the MDCT coefficient is quantized to, for example, about 16 bits to 8 bits. Normalization unit 24, the MDCT coefficients Xn _k, collectively for each block, frame by frame, and stores in the storage device 12. Further, the normalizing unit 24 stores the maximum value gain in the storage device 12. Then, the normalizing unit 24 notifies the band dividing unit 25 of the start of processing.

In response to this notification, the band division section 25, as shown in FIG. 5 (B), the entire frequency band of the MDCT coefficients Xn _k read from the storage device 12, for each block, P number (P: 2 or more Are divided logarithmically (divided) into subbands b_p (p: subband number). The band dividing unit 25 divides the frequency band logarithmically so as to be narrower as the low frequency band (low frequency band) and wider as the high frequency band (high frequency band) according to the characteristics of human hearing. As a result, the divided band b_p conforms to the auditory characteristics. The division number P of the divided band is preferably about 16, for example.

  When the frequency band is divided, the band dividing unit 25 notifies the maximum value searching unit 26 of the start of processing.

In response to this notification, the maximum value search unit 26, as shown in FIG. 6 (A), belonging to each frequency band b_p, it retrieves the maximum value of the absolute values of the MDCT coefficients Xn _k, for each frequency band b_p The maximum value env_p is acquired for each block. Then, as shown in FIG. 6B, the maximum value search unit 26 generates a maximum value sequence env [p] (p = 1 to P) including a set of these maximum values env_p for each block. . Maximum value detection section 26, MDCT coefficients Xn _k and maximum column env [p] a (p = 1 to P), are summarized for each block, frame by frame, and stores in the storage device 12. The maximum value search unit 26 notifies the maximum value sequence vector quantization unit 27 of the start of processing.

Upon receiving this notification, the maximum value sequence vector quantization unit 27 vector quantizes the maximum value sequence env [p] read from the storage device 12 for each block. For this vector quantization, a code book 50 stored in the ROM 13 is used. FIG. 7A shows a code book 50. As shown in FIG. 7 (A), q vectors V _j (j = 1 to q) of the same dimension P as the dimension P of the maximum value sequence env [p] are registered in the codebook 50. Yes. The maximum value sequence vector quantization unit 27 refers to the code book 50 and obtains the value of the index j that minimizes the value e _{j of} the following equation (6) for each block.

図７（Ｂ）では、各ブロックの最大値列ｅｎｖ［ｐ］が、ｅｎｖ［ｐ］₁〜ｅｎｖ[ｐ]_Nとして示されている。最大値列ベクトル量子化部２７は、求められたｅｎｖ［ｐ］₁〜ｅｎｖ[ｐ]_Nにそれぞれ対応するインデックスｊを、時系列順に並べることにより、図７（Ｂ）に示されるような最大値列インデックスのデータ列ｉｎｄｅｘ１［ｉ］（ｉ＝１〜Ｎ）を生成する。データ列ｉｎｄｅｘ１［ｉ］は、記憶装置１２に格納される。そして、最大値列ベクトル量子化部２７は、最大値除算部２８に処理開始を通知する。

In FIG. 7B, the maximum value sequence env [p] of each block is shown as env [p] _{1 to} env [p] _N. The maximum value sequence vector quantization unit 27 arranges the indexes j respectively corresponding to the obtained env [p] ₁ to env [p] _N in time series order, thereby obtaining the maximum value as shown in FIG. A data string index1 [i] (i = 1 to N) of the value string index is generated. The data string index1 [i] is stored in the storage device 12. Then, the maximum value sequence vector quantization unit 27 notifies the maximum value division unit 28 of the start of processing.

In response to this notification, the maximum value the division unit 28, the MDCT coefficients Xn _k, a data string index1 [i], from the storage device 12. Then, the maximum value the division unit 28, the MDCT coefficients Xn _k belonging to each divided band of each block, the value (i.e., the inverse quantized value) of each element of the codebook 50 corresponding to the index index1 [i] of the block Then, the division is performed using the inverse quantization value corresponding to the divided band. Thereby, MDCT coefficient Xe _k (k = 1 to M / 2-1) is obtained for each block. 8, this was produced by the division, and an example of the MDCT coefficients Xe _k of a block is illustrated. Maximum division unit 28, the MDCT coefficients Xe _k, collectively for each block, frame by frame, and stores in the storage device 12. Then, the maximum value division unit 28 notifies the quantization unit 29 of the start of processing.

In response to this notice, the quantization unit 29, the MDCT coefficients Xe _k read from the storage device 12 is quantized with accuracy (number of bits) which is previously set for each divided band b_p. 9A and 9B show the state of quantization. The quantization unit 29 obtains the MDCT coefficient Xq _k shown in FIG. 9B as a result of quantizing the MDCT coefficient Xe _k shown in FIG. The quantization unit 29 stores the acquired MDCT coefficient Xq _k in the storage device 12. Then, the quantization unit 29 notifies the time order rearrangement unit 30 of the start of processing.

In response to this notification, the time-order rearrangement unit 30 as the time-series rearrangement unit includes the MDCT coefficients Xq _k included in the MDCT coefficient group (that is, the first transform coefficient group) of each of the plurality of MDCT blocks in one frame. Are read from the storage device 12. Then, the time order rearrangement unit 30 regroups the MDCT coefficients Xq _k into groups of the same frequency, and converts the MDCT coefficients Xq _k to time for the regrouped MDCT coefficient group (that is, the second transform coefficient group). Sort in order. 10A and 10B show the rearrangement. As shown in FIG. 10 (A), i and (i = 1 to N) th MDCT coefficients Xq _k of th block, the correction coefficient Xq _{k, i.} As shown in FIG. 10B, the correction coefficients Xq _{k, i} are grouped into MDCT coefficient groups having the same frequency and rearranged in time series. Here, a vector having each MDCT coefficient Xq _{k, i} of the transform coefficient group having the same frequency as each element is a coefficient vector F _k (k = 1 to M / 2-1). The time order rearrangement unit 30 stores the coefficient vector F _k in the storage device 12. Then, the time order rearrangement unit 30 notifies the coefficient vector quantization unit 31 of the start of processing.

Upon receipt of this notification, the coefficient vector quantization unit 31 performs vector quantization on the coefficient vector F _k read from the storage device 12. For this vector quantization, a code book 51 stored in the ROM 13 is used. FIG. 11A shows a code book 51. As shown in FIG. 11A, in the code book 51, s representative vectors W _j having the same dimension N as the dimension N of the coefficient vector F _k are registered. The coefficient vector quantization unit 31 refers to the code book 51 to obtain the value of the index j that minimizes the value of the following equation (7) for each frequency (in short, k).

これにより、図１１（Ｂ）に示されるように、周波数毎に、係数ベクトルＦ_kに最も近いベクトルのインデックス値ｉｎｄｅｘ２［ｋ］（ｋ＝１〜Ｍ／２−１）が求められる。係数ベクトル量子化部３１は、ｉｎｄｅｘ２［ｋ］（ｋ＝１〜Ｍ／２−１）を、記憶装置１２に格納する。そして、係数ベクトル量子化部３１は、エントロピ符号化部３２に処理開始を通知する。

Thus, as shown in FIG. 11 (B), for each frequency, the index value index2 closest vector to the coefficient vector _{F k [k] (k =} 1~M / 2-1) is obtained. The coefficient vector quantization unit 31 stores index2 [k] (k = 1 to M / 2-1) in the storage device 12. Then, the coefficient vector quantization unit 31 notifies the entropy encoding unit 32 of the start of processing.

Upon receiving this notification, the entropy encoding unit 32 receives the maximum value gain, the index data string index1 [1] to index1 [N] of the maximum value string, and the index data string index2 [k] to MDCT coefficient group index. index2 [K] (K = M / 2-1 in the initial stage) and a code sequence C _t described later are read from the storage device 12. Then, the entropy encoding unit 32 uses the entropy encoding method such as a range coder or Huffman code to perform entropy encoding on the read data using a code table (not shown) to generate encoded data.

FIG. 12 shows data encoded by the entropy encoding unit 32. As shown in FIG. 12, as the data to be encoded by the entropy encoding unit 32, there is a code sequence C _t. The code number sequence C _t is _obtained by encoding the presence flag FLG _k . The presence flag FLG _k is a flag indicating whether or not the MDCT coefficient group of each frequency is an encoding target, as shown in FIG. If the presence flag FLG _k is 1, the MDCT coefficient group of the corresponding frequency is an encoding target, and if the presence flag FLG _k is 0, the MDCT coefficient group of the frequency is not an encoding target. In the initial stage, since MDCT coefficient groups of all frequencies are to be encoded, as shown in FIG. 13, the presence flag FLG _k is set to 1 at all frequencies. That is, the initial values of the presence flags FLG _k are all 1.

Thus, the flag sequence of the encoding flag FLG _k (hereinafter, abbreviated as the flag sequence FLG _{k as} appropriate) is a sequence of 0s and 1s. A code number sequence C _t is encoded by expressing this flag sequence FLG _k as a continuous number of 0s and 1s. FIG. 14 (A), the in FIG. 14 (B) is an example of a code sequence C _t is shown. As shown in FIG. 14 (A), in the presence flag sequence FLG _k , when the number of consecutive 0s and 1s is 2, 2, 1, 3, 3, 1, 1 in order, the code sequence C _t is {2, 2, 1, 3, 3, 1, 1}.

Further, as shown in FIG. 14 (B), in the flag column FLG _k, when the continuous number of the same value continues is equal to the upper limit value, the code sequence C _t, and the number of consecutive, the following 0 is inserted between the consecutive numbers.

As shown in FIG. 14B, when the flag sequence FLG _k starts from 1, 0 is inserted at the head of the code sequence C _t .

As described above, since the presence flags FLG _k are all 1 at the initial stage, the initial value of the code sequence C _t is also uniquely determined. The storage device 12 stores in advance the initial value of the code number sequence C _t when the presence flags FLG _k are all set to 1. In the first encoding, the entropy encoding unit 32 reads an initial value of the code sequence C _t and entropy encodes it.

  The entropy encoding unit 32 stores the encoded data in the storage device 12. Then, the entropy encoding unit 32 notifies the code amount comparison unit 33 of the start of processing.

  Upon receiving this notification, the code amount comparison unit 33 reads the code data stored in the storage device 12 and compares the code amount of the data with the target code amount. This target code amount is set in advance to ensure a data rate of about 16 kbps while ensuring sufficient sound quality.

  The code amount comparison unit 33 compares the total code amount with the target code amount, and determines whether the total code amount is equal to or less than the target code amount. If the determination is negative, the code amount comparison unit 33 notifies the data deletion unit 34 to that effect.

Upon receiving this notification, the data deletion unit 34 deletes a part of the encoded data. First, the data deletion unit 34 calculates the importance (degree of influence on sound quality) of the MDCT coefficient group, that is, the MDCT coefficient group corresponding to each coefficient vector F _k for each frequency.

The simplest method for calculating the importance is a method for calculating the total energy g _k for each frequency. The total energy g _{k for} each frequency is expressed by the following equation (8).

Note that the data deleting unit 34 may multiply the energy g _k by a weighting coefficient depending on the frequency. For example, the data deletion unit 34 multiplies the MDCT coefficient belonging to the frequency band below 500 Hz by 1.3, and multiplies the MDCT coefficient belonging to the frequency band of 500 or more and less than 3500 Hz by 1.1 to 3500 Hz or more. The frequency band correction coefficient can be multiplied by 1.0. That is, the weight of the low frequency can be increased.

Next, the data deleting unit 34, as shown in Figure 15, the elements that energy g _k corresponding to the frequency is a minimum, from index2 [k], by removing, compressing index2 [k]. Further, the data deletion unit 34 sets the presence flag FLG _k corresponding to the frequency _where the value of the energy g _k is 0 to 0.

Further, the data deleting unit 34 generates the code sequence C _t using the method described above based on the changed presence flag FLG _k . The data deletion by the data deleting unit 34, an increasing number of existence flag FLG _k value becomes 0, increases the number of continuous 0, the code length of the code sequence C _t becomes shorter. The data deleting unit 34 stores the compressed index 2 [k] and the code number sequence C _t in the storage device 12. Then, the data deletion unit 34 notifies the entropy encoding unit 32 of the start of processing.

  Then, the entropy encoding unit 32 reads out the data shown in FIG. 12 from the storage device 12, performs entropy encoding again, and stores the encoded data in the storage device 12. Then, the entropy encoding unit 32 notifies the code amount comparison unit 33 of the start of processing.

The code amount comparison unit 33 compares the code amount of the encoded data with the target code amount again. In this way, data deletion in the data deletion unit 34 and entropy encoding in the entropy encoding unit 32 are repeated until the determination in the code amount comparison unit 33 is affirmed. In this iterative process, as shown in FIG. 15, the MDCT coefficient group is excluded from the encoding target in ascending order of energy g _k , and the number K of elements of index2 [k] (K <M / 2-1) is reduced, it becomes code sequence C _t is gradually decreased, as a result, the code amount of data to be encoded becomes smaller.

  When the code amount becomes equal to or less than the target code amount and the determination in the code amount comparison unit 33 is affirmed, the encoded data at that time is stored in the storage device 12 as a code string.

  Next, the decoding unit 17 will be described. The decoding unit 17 reads the encoded data from the storage device 12, decodes the encoded data, and generates a digital audio signal. As shown in FIG. 16, the decoding unit 17 includes an entropy decoding unit 41, a coefficient inverse vector quantization (VQ) unit 42, a frequency rearrangement unit 43, an inverse quantization unit 44, and a maximum value sequence inverse vector. A quantization (VQ) unit 45, a maximum value multiplication unit 46, a gain synthesis unit 47, an IMDCT unit 48, and codebooks 50 and 51 are provided.

  The entropy decoding unit 41 performs entropy decoding of the encoded data read from the storage device 12, and acquires various data shown in FIG. These data are stored in the storage device 12. The entropy decoding unit 41 notifies the coefficient inverse vector quantization (VQ) unit 42 of the start of processing.

In response to this notification, the coefficient inverse vector quantization (VQ) unit 42 refers to the code book 51 based on the code sequence C _t and index2 [k] (k = 1 to K), and performs the inverse vector. Quantization is performed to generate MDCT coefficients Xq _k . More specifically, the coefficient inverse VQ unit 42 first decodes the presence flag FLG _k based on the code sequence C _t as shown in FIG. In decoding, each element of index2 [k] (k = 1 to K) corresponds to the presence flag FLG _k being 1. Therefore, the coefficient inverse vector quantization (VQ) unit 42 refers to the decoded presence flag FLG _k , sets the coefficient vector F _k to 0 vector at the frequency at which the presence flag FLG _k is 0, and sets the presence flag FLG _k. Is set to a coefficient vector F _{k in} the order of _k .

In this way, the coefficient inverse vector quantization (VQ) unit 42 generates the MDCT coefficient group F _k at all frequencies. FIG. 17 schematically shows an example of the MDCT coefficient group F _k generated in this way. The MDCT coefficient group F _k is either a coefficient group in which all elements are 0 or a coefficient group corresponding to the coefficient vectors W _{1 to} W _s in the code book 51. Each element of the MDCT coefficient group F _k is an MDCT coefficient Xq _{k, i} . The coefficient inverse vector quantization (VQ) unit 42 stores the MDCT coefficient Xq _{k, i} in the storage device 12. Then, the coefficient inverse vector quantization (VQ) unit 42 notifies the frequency rearrangement unit 43 of the start of processing.

In response to this notification, the frequency rearranging unit 43 rearranges the MDCT coefficients Xq _{k, i} stored in the storage device 12 in the MDCT coefficient group of each block, that is, in the frequency direction, and stores the rearranged MDCT coefficients Xq _{k, i} in the storage device 12. Then, the frequency rearrangement unit 43 notifies the inverse quantization unit 44 of the start of processing.

In response to this notification, the inverse quantization unit 44 performs inverse quantization on the MDCT coefficient Xq _{k of} each block read from the storage device 12 with an accuracy set in advance for each divided band b_p, and the MDCT coefficient to generate a Xe _k for each block. Inverse quantization unit 44, the acquired MDCT coefficients Xe _k, stored in the storage device 12, and sends the processing start to the maximum column inverse vector quantization (VQ) unit 45.

In response to this notification, the maximum value sequence inverse vector quantization (VQ) unit 45 refers to the code book 50 and is based on the data sequence index1 [i] of the maximum value sequence decoded and stored in the storage device 12. Then, vector inverse quantization is performed to obtain a maximum value sequence env [p] _N. The maximum value sequence env [p] _N is stored in the storage device 12. The maximum value sequence inverse vector quantization (VQ) unit 45 notifies the maximum value multiplication unit 46 of the start of processing.

Upon receiving this notification, the maximum value multiplier 46 extracts the MDCT coefficient Xe _k belonging to the divided band b_p, and multiplies the extracted MDCT coefficient Xe _k by the maximum value env_p stored in env [p]. To do. This multiplication, MDCT coefficients Xn _k of each block is obtained. Maximum value multiplying unit 46 stores the MDCT coefficients Xn _k of each block acquired in the storage unit 12, and sends the processing start the gain combining unit 47.

In response to this notification, the gain combining section 47, the MDCT coefficients Xn _k of each block read out from the storage unit 12, by multiplying the maximum value gain decoded, to obtain the MDCT coefficient X _k of the blocks. The gain synthesis unit 47 stores the acquired MDCT coefficient X _k in the storage device 12 and notifies the IMDCT unit 48 of the start of processing.

The IMDCT unit 48 performs inverse MDCT for each block on the MDCT coefficient X _k read from the storage device 12. The IMDCT unit 48 further synthesizes the digital signals of the blocks obtained by the inverse MDCT to restore the digital audio signal Sound1. The restored digital audio signal Sound1 is sent to the audio input / output device 11 and reproduced.

Next, the operation of the speech processing apparatus 1 according to this embodiment will be described. In the encoding operation in the encoding unit 16, first, as shown in FIG. 18, the digital signal from which the DC component X _dc has been deleted in the DC deletion unit 21 is divided into N block signals in the MDCT unit 23. . The number of samples per block is M.

  Thereafter, MDCT is performed for each block in the MDCT unit 23. FIG. 19 schematically shows the code amount at this time. At this point, if the bit length of one transform coefficient is 16 bits, the code amount in units of frames is 16 × N (number of blocks) × M / 2-1 (number of transform coefficients per block). It has become.

  Thereafter, the normalization unit 24 normalizes the MDCT coefficients. When normalization is performed, the data length of all MDCT coefficients is reduced from 16 bits to 8 bits, for example, and the code amount per frame is reduced to ½ as shown in FIG. .

  Next, as shown in FIG. 21, the frequency division unit 25 divides the frequency band into P pieces, and searches for maximum values env [1] to env [P] in the respective divided bands. Then, as shown in FIG. 22, the maximum value sequence VQ unit 27 performs vector quantization on the determined maximum value sequence for env [1] to env [P]. The data string index1 [i] (i = 1 to N) of the maximum value string is generated.

  Next, the maximum value division unit 28 divides the MDCT coefficient using the maximum values env [1] to env [P]. FIG. 23 schematically shows how the MDCT coefficients belonging to each divided band are divided using each element (inverse quantization value) of the vector of the codebook 50 corresponding to index1 [i]. This division further reduces the number of bits of the MDCT coefficient.

Next, after the quantization of the MDCT coefficients in the quantization unit 29, the MDCT coefficients are rearranged in the time order rearrangement unit 30 as shown in FIG. In the quantization, the bit number of the MDCT transform coefficient increases as the frequency becomes lower. In FIG. 24, the bit number of the MDCT transform coefficient over the entire frequency range in order to prevent complication of the drawing. Are assumed to be the same. Then, as shown in FIG. 25, the coefficient VQ unit 31 refers to the code book 51, performs vector quantization of the MDCT coefficient, and obtains index2 [k]. Then, as shown in FIG. 26, the energy g _k is calculated in the data deletion unit 34, the MDCT coefficient group is deleted in order from the smallest energy g _k , index2 [k] is compressed, and the flag FLG _k is changed. The code number sequence C _t is compressed.

Then, as shown in FIG. 27, the entropy encoding unit 32, gain, index1 [i] , index2 [k], C t is to be coded. As the MDCT coefficient group corresponding to the frequency is deleted in ascending order of energy, the data length of the index 2 [k] and the code sequence C _t is shortened, and the data compression rate is improved.

Then, upon decoding in the decoding unit 17, as shown in FIG. 28, the flag FLG _k is decoded from the code sequence C _t. Then, index2 [k] is restored based on the flag FLG _k and the compressed index2 [k]. Then, based on the restored index2 [k], the MDCT coefficient group of each frequency is restored with reference to the code book 51. On the other hand, the maximum value sequence env [p] (env_p) is restored based on index1 [i]. Then, for each divided band, the decoded MDCT coefficient group and the maximum value env_p are multiplied to restore the MDCT coefficient Xq _k . Then, MDCT coefficients Xn _k is multiplied by the maximum gain, MDCT coefficient X _k is restored. By performing inverse MDCT on the restored MDCT coefficient X _k , the audio signal of each block is restored, and these audio signals are synthesized into audio signals in units of frames.

  As described above, the data encoded as described above is decoded by the decoding unit 17, quantized with 16 bits, and restored to the original audio signal sampled with a frequency of 16 kHz. As a result, the sound quality of the sound reproduced by the sound processing apparatus 1 is suitable for learning applications.

  As described above, according to the present embodiment, the data of the MDCT coefficient group is deleted in ascending order of energy. In this way, data that does not affect the sound quality can be preferentially deleted, so that the coding efficiency can be increased without degrading the sound quality.

<< Second Embodiment >>
Next, a second embodiment of the present invention will be described. In the present embodiment, the configuration of the encoding unit 16 is different from the configuration of the encoding unit 16 according to the first embodiment.

  As illustrated in FIG. 29, the encoding unit 16 according to the present embodiment includes an encoding frequency selection unit 35 instead of the data deletion unit 34. Further, in the encoding unit 16 according to the present embodiment, the code amount comparison unit 33 is not provided, and a loop based on the comparison result is not provided.

  The encoding frequency selection unit 35 selects a frequency to be encoded in the entropy encoding unit 32. The encoding frequency selection unit 35 calculates importance for each frequency, and selects a frequency to be encoded based on the importance.

More specifically, the encoding frequency selection unit 35 calculates the energy g _k according to the above equation (8), and determines that the importance of this frequency is higher as the energy g _k is higher. Then, the encoding frequency selection unit 35 selects encoding target frequencies in descending order of importance.

The encoding frequency selection unit 35 may multiply the energy g _k by a frequency-dependent weighting factor. For example, the encoding frequency selection unit 35 multiplies the MDCT coefficient in the frequency band of less than 500 Hz by 1.3, and multiplies the MDCT coefficient in the frequency band of 500 to less than 3500 Hz by 1.1. Then, the MDCT coefficient of the frequency band of 3500 Hz or higher may be multiplied by 1.0.

The encoding frequency selection unit 35 determines whether or not the code amount of the encoded data has reached the target code amount, and selects a frequency to be encoded until the code amount of the encoded data reaches the target code amount. The encoding frequency selection unit 35 generates index2 [k] by rearranging the indexes obtained by vector quantization for the MDCT coefficient group of the selected frequency in order of frequency. Also, the encoding frequency selection unit 35 sets the flag FLG _k corresponding to the MDCT coefficient group of the selected frequency to 1, and sets the remaining flag FLG _k to 0. Then, the encoding frequency selection unit 35 encodes the generated flag FLG _k into a code number sequence C _t . The encoding frequency selection unit 35 stores the index 2 [k] and the code number sequence C _t in the storage device 12. Then, the encoding frequency selection unit 35 notifies the entropy encoding unit 32 of the start of processing.

  The entropy encoding unit 32 encodes the data shown in FIG. 12 to generate encoded data, as in the first embodiment.

  As described above, instead of deleting data from the lower importance as in the present embodiment, the data selected before reaching the target code amount is selected from the higher importance. Since the amount can be reduced, the time required for encoding can be shortened.

<< Third Embodiment >>
Next, a third embodiment of the present invention will be described. Since the configuration of the speech processing apparatus according to this embodiment is the same as the configuration of the speech processing apparatus according to each of the above embodiments, detailed description thereof is omitted.

  In the present embodiment, the audio data is compressed not only in units of frames, but also is compressed over a plurality of frames. FIG. 30 shows a flowchart of the encoding operation of the speech processing apparatus according to this embodiment. First, as shown in step 201, the CPU 15 encodes the digital audio signal for each frame in the encoding unit 16 as described in the above embodiments. Here, the target code amount can be changed for each frame. In the next step 203, the CPU 15 determines whether or not all frames have been encoded. If this determination is negative, the CPU 15 returns to step 201. In this way, all frames are encoded.

  If the determination in step 203 is affirmed, the CPU 15 proceeds to step 205. In step 205, the CPU 15 calculates the sum of the code amounts of all frames. In the next step 207, the CPU 15 determines whether or not the sum of the code amounts is equal to or less than the overall target code amount. If this determination is positive, the CPU 15 ends the encoding process. On the other hand, if this determination is negative, the CPU 15 proceeds to step 209.

In step 209, the CPU 15 searches the MDCT coefficient group having the minimum importance from all frames. For example, as shown in FIG. 31, in frame 1, five MDCT coefficient groups are to be encoded, in frame 2, four MDCT coefficient groups are to be encoded, and in frame 3, Assume that four MDCT coefficient groups are to be encoded. Also, let g _{i, k} be the energy of the frequency k of the i-th frame. In this case, CPU 15, all of the MDCT coefficient set, energy g _i, sorted in ascending order of _k, looking for energy g _i, the minimum conversion coefficient set of _k. In the example of FIG. 31, the energy g _1,9 of the MDCT coefficient group F ₉ of frame 1 is minimum.

In the next step 211, the CPU 15 excludes the MDCT coefficient group having the smallest energy g _{i, k} from the encoding target. In the example of FIG. 31, the transform coefficient group F ₉ of frame 1 is excluded from the encoding target. In the next step 213, the CPU 15 re-encodes the entire frame in which the MDCT coefficient group is excluded from the encoding target. In the example of FIG. 31, data relating to frame 1 is re-encoded.

  After performing Step 213, the CPU 15 returns to Step 205, calculates the sum of the code amounts of all frames (Step 205), and compares it with the entire target code amount (Step 207). In this way, steps 205 → 207 → 209 → 211 → 213 are repeated until the determination in step 207 is affirmed. In this way, the encoding amount of a plurality of frames is suppressed to the overall target encoding amount.

  In step 211, one MDCT coefficient group is excluded one by one, but a plurality may be excluded at a time.

  In the present embodiment, Step 207 corresponds to the overall code amount determination unit, Steps 209 and 211 correspond to the adjustment unit, and Step 213 corresponds to the re-encoding unit. According to the present embodiment, it is possible to compress data in a plurality of entire frames. For this reason, in order to maintain the sound quality, the data compression rate of a frame that has a low data compression rate does not affect the sound quality even if the data compression rate is low or the data compression rate is increased. And the overall data compression rate can be improved.

  For example, the target code amount of each frame is set to 16 to 20 kbps, and the entire target code amount is set to 12 kbps. In this way, depending on the frame, the code amount is set to about 20 kbps in order not to deteriorate the sound quality, and the data compression in other frames that do not affect the sound quality even if the data compression rate is increased is improved, and the code amount is set to 12 kbps as a whole. It will be possible to: As a result, the coding efficiency of the data can be improved while ensuring the sound quality.

  As described above in detail, according to each of the above embodiments, the MDCT coefficients are grouped for each frequency by the time order rearrangement unit 30 to remove groups with small energy. Thereby, encoding efficiency can be improved without impairing sound quality. Furthermore, since the MDCT coefficient group grouped for each frequency is vector-quantized, the data compression rate can be increased as compared with scalar quantization. As a result, encoding efficiency can be improved without deteriorating sound quality.

  Further, according to each of the above embodiments, the maximum value sequence VQ unit 27 vector-quantizes the data sequence of the maximum value sequence index, so that the data compression rate can be increased as compared with scalar quantization. . As a result, encoding efficiency can be improved without deteriorating sound quality.

  By performing the encoding operation according to each of the above embodiments, a 16 KHz sampling audio signal can be compressed to about 12 kbps while maintaining the sound quality for learning use.

Further, according to the above embodiments, the data deletion portion 34 or encoding the frequency selection unit 35, the flag column FLG _k is converted into the number of consecutive sequence C _t in which the same value continues in the flag column FLG _k The As a result, the data compression rate can be further increased while maintaining a reversible state. As a result, encoding efficiency can be improved without deteriorating sound quality.

Further, according to each of the above embodiments, in the data deletion unit 34 or the encoding frequency selection unit 35, when the consecutive number of consecutive identical values is equal to the upper limit value in the flag sequence FLG _k , the code sequence C _{In t} , 0 is inserted between the continuous number and the next continuous number. In this way, if the upper limit value of the continuous number is provided, the data length of the code number sequence can be shortened equally regardless of the appearance pattern of the continuous number of the flag sequence FLG _k. It becomes like this.

Further, according to the above embodiments, the data deletion portion 34 or encoding the frequency selection unit 35, the flag column FLG _k is, when starting from 1, 0 to the beginning of the code sequence C _t is inserted . In this way, reversible data compression can be performed on the flag string FLG _k .

In order to improve the sound quality of the audio signal, it is desirable to increase the coding amount related to the low sound range as much as possible and relatively reduce the coding amount related to the high sound region in terms of human auditory characteristics. Therefore, in each of the above embodiments, the number of bits in the quantization unit, the weight multiplied by the energy g _k , and the like are increased in the low sound range. From this point of view, two codebooks for low-frequency and high-frequency codebooks are prepared for use in vector quantization of MDCT coefficients. The low-frequency codebook increases the number of vectors q, In the codebook of the range, the number q of vectors may be relatively reduced.

  In each of the above embodiments, MDCT is applied as frequency conversion. However, the frequency conversion method is not limited to MDCT, and DCT can also be adopted.

  In each of the above-described embodiments, the program is described as being stored in advance in a memory or the like. However, a program for executing the above-described processing is stored on a computer-readable recording medium such as a flexible disk, a CD-ROM (Compact Disk Read-Only Memory), a DVD (Digital Versatile Disk), or an MO (Magneto Optical disk). It may be stored and distributed, installed in another computer, operated as the above-mentioned means, or the above-mentioned steps may be executed.

  Furthermore, the program may be stored in a disk device or the like included in a server device on the Internet, and may be downloaded onto a computer by being superimposed on a carrier wave, for example.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation and application are possible. The above-described hardware configuration, block configuration, and flowchart are examples, and are not limited. For example, in each of the embodiments described above, a mobile phone or an electronic dictionary has been described as the voice processing device. However, the present invention can be easily applied to PHS (Personal Handyphone System), PDA (Personal Digital Assistants), or general personal computers. That is, the said embodiment is for description and does not restrict | limit the scope of the present invention.

1 is a block diagram showing a schematic configuration of a speech processing device 1 according to a first embodiment of the present invention. It is a block diagram which shows the structure of the encoding part of FIG. It is a figure which shows an example of DC component removal of a digital signal. It is a figure which shows an example of a frame division | segmentation. FIG. 5A is a diagram illustrating an example of the maximum value of the transform coefficient of each block, and FIG. 5B is a diagram illustrating a divided band. FIG. 6A is a diagram illustrating an example of the maximum value, and FIG. 6B is a diagram illustrating an example of the maximum value sequence. FIG. 7A is a diagram illustrating an example of the code book of the maximum value sequence, and FIG. 7B is a diagram illustrating an example of the data sequence of the maximum value sequence index. It is a figure which shows an example of the division by the maximum value. FIG. 9A is a diagram illustrating an example of an MDCT coefficient before quantization, and FIG. 9B is a diagram illustrating an example of an MDCT coefficient after quantization. FIG. 10A is a diagram illustrating an example of the MDCT coefficient group before rearrangement, and FIG. 10B is a diagram illustrating an example of the MDCT coefficient group after rearrangement. FIG. 11A is a diagram illustrating an example of an MDCT coefficient codebook, and FIG. 11B is a diagram illustrating an example of a data string of an MDCT coefficient index obtained by vector quantization. It is a figure which shows an example of the data of encoding object. It is a figure which shows an example of the data row | line | column and flag row | line | column of the MDCT coefficient index produced | generated by importance. FIG. 14A and FIG. 14B are diagrams illustrating an example of a code number sequence generation method based on a flag sequence. It is a figure which shows an example of data deletion. It is a block diagram which shows the structure of the decoding part of FIG. It is a figure which shows typically an example of the inverse vector quantization of a MDCT coefficient. It is a figure which shows typically the change of the data at the time of block division. It is a figure which shows the change of the data at the time of MDCT typically. It is a figure which shows typically the change of the data at the time of normalization. It is a figure which shows typically the change of the data at the time of a band division | segmentation and a maximum value search. It is a figure which shows typically the change of the data at the time of maximum value column vector quantization. It is a figure which shows typically the change of the data at the time of the maximum value division. It is a figure which shows typically the change of the data at the time order rearrangement. It is a figure which shows typically the change of the data at the time of the vector quantization of a MDCT coefficient. It is a figure which shows typically the change of the data at the time of data compression. It is a figure which shows typically the change of the data at the time of an encoding. It is a figure which shows typically the change of the data in decoding operation | movement. It is a block diagram which shows the structure of the encoding part which concerns on the 2nd Embodiment of this invention. It is a flowchart of the encoding operation | movement which concerns on the 3rd Embodiment of this invention. It is a figure which shows an example of the data compression between frames.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Voice processing device 11 Voice input / output device 12 Storage device 13 ROM
14 RAM
15 CPU
16 encoding unit 17 decoding unit 21 DC removal unit 22 framing unit 23 MDCT unit 24 normalization unit 25 band division unit 26 maximum value search unit 27 maximum value sequence vector quantization unit 28 maximum value division unit 29 quantization unit 30 time Order rearrangement unit 31 Coefficient vector quantization unit 32 Entropy encoding unit 33 Code amount comparison unit 34 Data deletion unit 35 Encoding frequency selection unit 41 Entropy decoding unit 42 Coefficient inverse vector quantization unit 43 Frequency rearrangement unit 44 Inverse quantization Unit 45 maximum value sequence inverse vector quantization unit 46 maximum value multiplication unit 47 gain synthesis unit 48 IMDCT unit 50, 51 codebook

Claims (9)

A dividing unit that divides a digital signal having a predetermined time length into a plurality of blocks;
A frequency conversion unit that frequency-converts the digital signal of each block and generates a first conversion coefficient group for each block;
A band dividing unit that divides the first transform coefficient group generated by the frequency converting unit into a plurality of small frequency bands such that the bandwidth increases as the frequency increases;
By searching the maximum value of the absolute value of the first transform coefficient belonging to the small frequency band for each small frequency band, and arranging the searched maximum values in order of frequency, a maximum value string is obtained for each block. A maximum value search unit to be generated;
Maximum value sequence vector quantization that generates a data sequence of the maximum value sequence index by vector quantizing the maximum value sequence of each block using a maximum value sequence codebook and arranging the obtained indexes in time series order And
A first transform coefficient belonging to each small frequency band of each block by inversely quantizing the maximum value sequence index of each block obtained by the maximum value sequence vector quantization unit using the maximum value sequence codebook A division unit that divides the group by using the inverse quantization value corresponding to the block and the inverse quantization value of the small frequency band;
When generating a second transform coefficient group for each frequency by rearranging the transform coefficients of the same frequency included in the first transform coefficient group of each block divided by the division unit in time series A series sorting section;
A transform coefficient vector quantization unit that vector-quantizes the second transform coefficient group of each frequency using a transform coefficient codebook and generates a data string of transform coefficient indexes by arranging the obtained indexes in order of frequency. When,
A flag indicating whether or not the second transform coefficient group of each frequency is a coding target by compressing the data string of the transform coefficient index based on the importance of the second transform coefficient group of each frequency. A data compression unit that generates information about,
An encoding unit that encodes the data sequence of the maximum value sequence index generated by the maximum value sequence vector quantization unit, the information about the flag generated by the data compression unit, and the compressed data sequence;
An encoding device comprising: A code amount determination unit that repeats the determination of whether or not the code amount of the data encoded by the encoding unit is smaller than the target code amount until the determination is affirmative,
The data compression unit
When the determination of the code amount determination unit is denied, by compressing the data sequence of the transform coefficient index by deleting the second transform coefficient group from the encoding target in order of increasing importance, Generating information about the flag,
The encoding unit includes:
2. The code according to claim 1, wherein the data sequence compressed by the data compression unit and the information on the generated flag are encoded until the determination of the code amount determination unit is affirmed. Device. The data compression unit
The frequency to be encoded by the encoding unit is selected in descending order of importance until the code amount of the encoded data is less than the target code amount and close to the target code amount. ,
2. The encoding according to claim 1, wherein the second conversion coefficient group corresponding to the selected frequency is used as an encoding target, and the data sequence of the conversion coefficient index is compressed to generate information on the flag. apparatus. The data compression unit
Forming a flag string by arranging the generated flags in order of frequency,
2. The encoding apparatus according to claim 1, wherein, based on the formed flag sequence, a continuous number sequence in which the same value continues in the flag sequence is generated as information on the flag. The data compression unit
In the flag sequence, when the consecutive number of consecutive identical values is equal to the upper limit value, 0 is inserted between the consecutive number and the next consecutive number in the number sequence. The encoding device according to claim 4 . The data compression unit
The flag sequence will, when starting from 1, the encoding apparatus according to claim 4 or 5, characterized in that inserting a 0 to the beginning of the sequence. It is determined whether or not the sum of the code amounts of the encoded data encoded by the encoding unit with respect to the series of the plurality of digital signals having the predetermined time length is smaller than the entire target code amount. An overall code amount determination unit that repeats until affirmed;
When the determination of the overall code amount determination unit is negative, the conversion coefficient index corresponding to the second conversion coefficient group having the minimum importance as a whole is excluded from the data string of the conversion coefficient index and excluded An adjustment unit that changes the flag corresponding to the second transformed coefficient group to a value indicating that it is not an encoding target;
The transform coefficient index is excluded, any of claims 1 to 6, further comprising a, a re-encoding unit for re-encoding the data relating to the flag modified digital signals of said predetermined time length The encoding device according to one item. A dividing step of dividing a digital signal having a predetermined time length into a plurality of blocks;
A frequency conversion step of frequency-converting the digital signal of each block to generate a first conversion coefficient group for each block;
A band dividing step of dividing the first transform coefficient group generated in the frequency converting step into a plurality of small frequency bands such that the bandwidth becomes wider as the frequency becomes higher;
By searching the maximum value of the absolute value of the first transform coefficient belonging to the small frequency band for each small frequency band, and arranging the searched maximum values in order of frequency, a maximum value string is obtained for each block. A maximum value search step to be generated;
Maximum value sequence vector quantization that generates a data sequence of the maximum value sequence index by vector quantizing the maximum value sequence of each block using a maximum value sequence codebook and arranging the obtained indexes in time series order Process,
The first transform coefficient belonging to each small frequency band of each block by inversely quantizing the maximum value sequence index of each block obtained in the maximum value sequence vector quantization step using the maximum value sequence codebook A division step of dividing the group by using the inverse quantization value corresponding to the block and the inverse quantization value of the small frequency band;
When generating the second transform coefficient group for each frequency by rearranging the transform coefficients of the same frequency included in the first transform coefficient group of each block divided in the division step in time series A series rearrangement process;
Transform coefficient vector quantization step of generating a data sequence of transform coefficient indices by vector quantizing the second transform coefficient group of each frequency using a transform coefficient codebook and arranging the obtained indexes in order of frequency When,
A flag indicating whether or not the second transform coefficient group of each frequency is a coding target by compressing the data string of the transform coefficient index based on the importance of the second transform coefficient group of each frequency. A data compression process for generating information about,
An encoding step for encoding the data sequence of the maximum value sequence index generated in the maximum value sequence vector quantization step, the information about the flag generated in the data compression step, and the compressed data sequence;
An encoding method including: A division procedure for dividing a digital signal of a predetermined time length into a plurality of blocks;
A frequency conversion procedure for frequency-converting the digital signal of each block and generating a first conversion coefficient group for each block;
A band division procedure for dividing the first transform coefficient group generated by the frequency conversion procedure into a plurality of small frequency bands in which the bandwidth becomes wider as the frequency becomes higher;
By searching the maximum value of the absolute value of the first transform coefficient belonging to the small frequency band for each small frequency band, and arranging the searched maximum values in order of frequency, a maximum value string is obtained for each block. The maximum value search procedure to generate,
Maximum value sequence vector quantization that generates a data sequence of the maximum value sequence index by vector quantizing the maximum value sequence of each block using a maximum value sequence codebook and arranging the obtained indexes in time series order Procedure and
A first transform coefficient belonging to each small frequency band of each block by inversely quantizing the maximum value sequence index of each block obtained by the maximum value sequence vector quantization procedure using the maximum value sequence codebook A division procedure for dividing a group by using an inverse quantization value corresponding to the block and using the inverse quantization value of the small frequency band;
When generating the second transform coefficient group for each frequency by rearranging the transform coefficients of the same frequency included in the first transform coefficient group of each block divided by the division procedure in time series Series reordering procedure,
A transform coefficient vector quantization procedure for vector-quantizing the second transform coefficient group of each frequency using a transform coefficient codebook and generating a data sequence of transform coefficient indexes by arranging the obtained indexes in order of frequency. When,
A flag indicating whether or not the second transform coefficient group of each frequency is a coding target by compressing the data string of the transform coefficient index based on the importance of the second transform coefficient group of each frequency. A data compression procedure to generate information about,
An encoding procedure for encoding the data sequence of the maximum value sequence index generated by the maximum value sequence vector quantization procedure, the information about the flag generated by the data compression procedure, and the compressed data sequence;
A program that causes a computer to execute.

Images