JP2008289085A - Decoding method, decoder, decoding apparatus, encoding method, encoder, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a decoding technology and an encoding technology for satisfying the maintenance of encoding quality and reduction in computational complexity in encoding while reducing a memory capacity to be used. <P>SOLUTION: This decoding method includes an spreading code decomposing step, a weighted form decoding step, a weighting removal calculating step, a gain decoding step and a multiplication step. In the spreading code decomposing step, an input code is decomposed into a form code and a gain code. In the weighted form decoding step, a weighted form code book is used to convert the form code into a weighted form vector. In the weighting removal calculating step, a weight of the weighted form vector is removed and a form vector is outputted. In the gain decoding step, a gain code book is used to convert the gain code into a gain. In the multiplication step, the form vector and the gain are multiplied to output a decoded signal. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a decoding method for decoding a code, an encoding method for encoding the code, a decoder, a decoding device, an encoding device, a program, and a recording medium using these methods.

  Non-linear waveform compression coding (μ-law / A-law PCM) used in G.711 (Non-Patent Document 1), G-711 (Non-Patent Document 1), G-711 There is a waveform coding method such as differential prediction waveform compression coding (ADPCM) used in 726 (Non-Patent Document 2) and the like. In voice communication (VoIP) using a public telephone network and the Internet, this encoding method is almost used.

  On the other hand, when the transmission band of a line such as a cellular phone is limited, the coding method based on linear prediction analysis is the mainstream, and noise is transformed and encoded based on the envelope information obtained by this linear prediction analysis. Is used. However, in the linear prediction analysis method, it is necessary to obtain an autocorrelation function having a large amount of calculation for each encoding processing time unit. Further, when selecting a code, it is necessary to select a code by reflecting this envelope information for each encoding processing time unit, and the amount of calculation required for encoding is several tens of times that of the above-described waveform encoding method. In addition to the coding methods based on waveform coding and linear prediction analysis, there are many coding methods with high sound quality and good compression efficiency.

  Thus, when there are a large number of encoding methods, the interoperability is not guaranteed due to the difference in the methods. Therefore, when communicating with a plurality of terminal devices equipped with different encoding methods, the corresponding encoders and decoders are operated on their own communication terminals, and the encoding implemented in the terminal device of the communication partner It is necessary to use different encoding methods depending on the method. However, in a terminal device in which the amount of computation that can be used is limited, it is impossible to simultaneously operate a plurality of encoders having a large amount of computation. On the other hand, the waveform coding method is generally implemented in any VoIP conference terminal. From the above, eventually, a waveform encoding method such as G.711 or G.726 must be used.

  Since the coding method based on μ-law / A-law PCM and ADPCM uses nonlinear compression of amplitude, the coding noise superimposed on the reproduced speech has a strong correlation with the power of the entire speech and does not depend on the input speech level. There is an advantage that the S / N ratio can be made constant. However, this coding noise becomes white noise. Conventional G.G. 711 and G.G. Since the input voice to 726 is assumed to be changed to the frequency characteristics of a signal output from a conventional telephone in which high frequency components typified by IRS characteristics are emphasized, the high frequency components are According to a signal in which is emphasized, white noise is not perceived remarkably. Here, the IRS characteristic (Non-patent Document 3) indicates a gentle high-pass filter type frequency characteristic as shown in FIG.

  However, it is rare that a microphone having a frequency characteristic in which a high frequency component is emphasized is used in communication such as VoIP. For this reason, the power signal is encoded without correcting the power concentration in the low frequency range (encoder mismatch), the SN ratio on the high frequency side deteriorates due to the increase in power on the low frequency side, and noise occurs on the decoding side. This causes a problem that the image is perceived prominently. For example, if a microphone with flat frequency characteristics is used to pick up speech, the signal to be encoded is also concentrated in the low frequency range (about 1 kHz), and the encoding noise level relative to the input speech level in the high frequency range. Becomes relatively large, and noise is easily perceived on the decoding side.

  In order to solve the above problem, the applicant has realized an encoding method and a decoding method that reduce quantization noise caused by encoder mismatch in waveform encoding to a low computational complexity and high efficiency. In order to realize this, the basic code provides means for maintaining the bit stream compatibility with the conventional terminal and enhancing the interconnectivity by making the bit stream scalable (Patent Document 1). Specifically, in order to avoid quality degradation due to encoder mismatch, G. 711 or G.I. A multi-stage encoder using 726 as the basic stage is used, and the second stage for reducing the noise of the basic stage is an encoding that operates with a much lower amount of computation than the encoding method based on linear prediction analysis. The method was used.

In the encoder of Patent Document 1, it is possible to reproduce with high quality using a plurality of codebooks such as a high-frequency weighted shape codebook pre-weighted and a high-frequency weighted power reciprocal table, and code with low computational complexity. The function to be realized.
ITU-T (Telecommunication Standardization Sector, International Telecommunication Union), Geneva, Switzerland.ITU-T G.711-Pulse code modulation (PCM) of voice frequencies, Nov.1988. ITU-T (Telecommunication Standardization Sector, International Telecommunication Union), Geneva, Switzerland.ITU-T G.726-40,32,24,16 kbit / s adaptive differential pulse code modulation (ADPCM), Dec. 1990. ITU-T (Telecommunication Standardization Sector, International Telecommunication Union), Geneva, Switzerland.ITU-T P.830 Annex D-modified IRS send and receive characteristics, Feb. 1996. JP 2006-119301 A

  As described above, conventionally, in order to change the weight based on the linear prediction analysis result or the like in both the encoding process and the decoding process, the same codebook without the weight is used. In Patent Document 1, the amount of calculation is reduced while maintaining the quality of encoding by preparing a codebook that is pre-weighted for encoding processing. However, in the method of Patent Document 1, since a codebook with no weight is required for the decoding process, it is necessary to secure a memory for recording the weighted codebook.

  The present invention provides a decoding technique that satisfies the requirement that the encoding quality is kept equal to the conventional one while the amount of memory used is lower than that of Patent Literature 1, and that the amount of computation of the encoding is equivalent to that of Patent Literature 1. An object is to provide an encoding technique.

  The decoding method of the present invention includes an extended code decomposition step, a weighted shape decoding step, a weighting removal calculation step, a gain decoding step, and a multiplication step. In the extended code decomposition step, the input code is decomposed into a shape code and a gain code. The weighted shape decoding step converts the shape code into a weighted shape vector using a weighted shape codebook. The weighting removal calculating step removes the weight of the weighted shape vector and outputs the shape vector. The gain decoding step converts the gain code into a gain using a gain codebook. The multiplication step multiplies the shape vector and the gain to output a decoded signal. In the weight removal operation step, the weight of the weighted shape vector may be removed using the inverse matrix U of the weight matrix W used for encoding.

The encoding method of the present invention includes a weighting step, a shape calculating step, a gain calculating step, and a multiplexing step. The shape calculation step includes an initialization substep, an inner product substep, an ideal gain calculation substep, a distance calculation substep, and a confirmation substep. Then, the inner product sub-step, ideal gain calculation sub-step, distance calculation sub-step, and confirmation sub-step are repeated until n is the number N of weighted shape vectors in the weighted shape codebook, and i _opt at the end of the iteration is shaped It is assumed that the code is _Is , and the shape code _Is and the optimum ideal gain g ^to _opt at the end of the repetition are output. In the gain calculation step, when g _x is the gain having the x-th smallest value in the gain codebook, the order of g ^to _opt <(g _x + g _{x + 1} ) / The first x satisfying 2 is searched, and x is output as the gain code _Ig .

  According to the decoding method of the present invention, since a weighted shape codebook can be used for decoding, the weighted shape codebook used for encoding can also be used for decoding. Therefore, when recording a weighted shape codebook in the memory, there is no need to record a non-weighted shape codebook. Since it is necessary to provide both an encoder and a decoder at the same time in bidirectional communication, the codebook of the encoder and the decoder becomes common when combined with the encoding method of Patent Document 1, so it is actually necessary. It is possible to provide a decoding technique and an encoding technique that satisfy the requirements of maintaining the quality of encoding and reducing the amount of calculation of encoding while greatly reducing the amount of memory used.

[First Embodiment]
[Encoding device]
FIG. 2 shows a functional configuration example of the encoding apparatus according to the first embodiment. The encoding apparatus according to the first embodiment includes a basic encoder 910, a basic decoder 920, an adder 930, and an extended encoder 100. The basic encoder 910 may be a conventional waveform encoding encoder such as G.711 or G.726. The basic decoder 920 is a decoder corresponding to the basic encoder 910. Basic encoder 910 encodes the input signal s, and outputs the base code I _b. Elementary decoders 920 decodes the base code _{I b.} The adding unit 930 obtains a difference (basic coding quantization noise) between the decoded signal and the input signal. The extension encoder 100 encodes quantization noise (basic noise signal e) of basic encoding. If real-time processing is performed, the extension encoder performs processing for each short processing frame of 8 samples (1 ms) to 160 samples (20 ms). If the number of samples of the processing frame at this time is K, the basic noise signal e can be expressed by a K-dimensional vector. The input signal to the encoding device is an audio digital signal in the 3.4 kHz band (telephone band) sampled at 8 kHz, for example. If real-time processing is not performed, batch processing may be performed within the range allowed by the memory.

Since the present invention particularly relates to the extension encoder 100 in the encoding apparatus of FIG. 2, the extension encoder 100 will be described below. FIG. 3 is an example of a processing flow of the extension encoder 100. The extended encoder 100 includes a weighting unit 110, a shape calculation unit 120, a gain calculation unit 130, a multiplexing unit 140, a weighted shape codebook 150, a weighted shape power reciprocal book 160, and a gain codebook 170. In the weighted shape codebook 150, N weighted shape vectors Wc ₁ ,..., Wc _N are recorded. Each weighted shape vector is a K + r-dimensional vector composed of weighted shape signals of K + r samples. Here, r is the number of taps of the weighting filter, and is 1 in the case of weighting according to equation (2) described later. In the weighted shape power reciprocal book 160, reciprocals of norms of weighted shape vectors (1 / ‖Wc ₁ ‖ ² ), ..., (1 / ‖Wc _N ‖ ² ) are recorded. Gains g ₁ ,..., G _M are recorded in gain codebook 170. The weighting unit 110 assigns a weight to the basic noise signal (input signal to the extension encoder 100) e and outputs a weighted noise signal We (S110). Shape calculation unit 120, as a distance between the weighted noise signal We is minimized, weighted shape vector Wc n _(n is an integer 1~N in weighted shape codebook 0.99, N is the weighted coded shape determine the optimum ideal gain g ^~ _opt with selecting the number) of the weighted shape vectors stored in the book, and outputs the coded shape I _s and optimal ideal gain g ^~ _opt indicating the weighted shape vector (S120). The gain calculator 130 stores the gain g _m in the gain codebook 170 (m is an integer from 1 to M, and M is stored in the gain codebook so that the distance ^{from the} optimal ideal gain g ^to _opt is minimized. select a gain number of) that are, for outputting a gain code _{I g} indicating the gain (S130). Multiplexer 140, and a coded shape _{I s} and the gain code _{I g} are multiplexed, and outputs the code _{I e} (S140). Hereinafter, the weighting unit 110, the shape calculation unit 120, and the gain calculation unit 130 will be described in detail.

と表現されるＦＩＲフィルタを用いる方法がある。このフィルタは、数学的には（Ｋ＋ｒ）行Ｋ列の行列として次のように表現できる。

式（１）の場合は１タップのフィルタであるため、ｒ＝１である。重みを付与する演算はこの行列Ｗを基本雑音信号ｅに乗じ、重み付き雑音信号Ｗｅを出力する演算である。すなわち畳み込み演算である。なお、例えば、ｂの値として0.550107181を用いた場合のフィルタの周波数特性を図４に示す。図４（Ａ）は振幅の周波数特性、図４（Ｂ）は位相の周波数特性を示している。このようなフィルタ（重み付け）を用いることによって、基本雑音信号ｅの低域成分は大幅に減衰され、高域を重視するため、拡張符号化器１００では高域の雑音を低減できる。なお、入力信号が音声でない場合には、入力信号の特性に合わせたフィルタを用意すればよい。また、このフィルタはｒが１よりも大きい高次のフィルタとしてもよい。 Weighting unit 110
The weighting unit 110 assigns a weight to the basic noise signal (input signal to the extension encoder 100) e and outputs a weighted noise signal We (S110). Specifically, for example,

There is a method using an FIR filter expressed as: This filter can be expressed mathematically as a matrix of (K + r) rows and K columns as follows.

In the case of Expression (1), since it is a one-tap filter, r = 1. The operation for assigning a weight is an operation for multiplying the basic noise signal e by this matrix W and outputting a weighted noise signal We. That is, a convolution operation. For example, FIG. 4 shows the frequency characteristics of the filter when 0.550107181 is used as the value of b. 4A shows the frequency characteristics of the amplitude, and FIG. 4B shows the frequency characteristics of the phase. By using such a filter (weighting), the low frequency component of the basic noise signal e is greatly attenuated and the high frequency is emphasized. Therefore, the extended encoder 100 can reduce high frequency noise. If the input signal is not voice, a filter that matches the characteristics of the input signal may be prepared. Further, this filter may be a higher-order filter in which r is larger than 1.

Shape calculator 120
Shape calculation unit 120, the distance between the vector (n is an integer of 1 to N) weighted shape vector _Wc n weighted noise signal We and the weighted shape codebook 150 multiplied by the optimal ideal gain ^g _{~ opt} to as D ^~ is minimized, selecting the weighted shape vector Wc _n and optimal ideal gain ^g _{~ opt} (S120). However, the ideal gain refers to the gain before quantization (gain obtained by calculation). Further, the optimal ideal gain, is ideal gain can be minimized distance D ^~ of the respective weighted shape vector Wc _n and weighted noise signal We.

が成り立つ。なお、式中のｔは、行列ないしはベクトルの転置を表す。この式をｇ^〜 _ｏｐｔについて解くと、

となる。ｇ^〜 _ｏｐｔを式（３）に代入すると、次式となる。

右辺の第１項‖Ｗｅ‖^２は、Ｗｃ_ｎに対して不変であるため、定数項である。したがって、右辺第２項を最大にすることが、距離Ｄ^〜を最小とすることになる。そこで、拡張符号化器１００では、距離ｄ_ｓを次式のように定義し、距離ｄ_ｓが最大となる重み付き形状ベクトルＷｃ_ｎと最適な理想利得ｇ^〜 _ｏｐｔを選定する。

Here, the distance D ^~ can be expressed as follows.
_{^{D ~ = ‖We-g ~ opt}} Wc n ‖ ² (3)
Distance ^{D ~} ideal gain ^g _~ to minimize _opt, since the distance obtained by partially differentiating ^{D ~} the ideal gain ^g _{~ opt} value of 0

Holds. Note that t in the equation represents transposition of a matrix or a vector. Solving this equation for g ^~ _opt ,

It becomes. When the g ^~ _opt into equation (3), the following equation.

The first term ‖We‖ ² on the right side are the invariant to Wc _n, constant terms. Therefore, maximizing the second term on the right side becomes the distance D ^~ to be minimized. Therefore, the extension encoder 100, the distance _{d s} is defined as: the distance _{d s} to select the most ideal gain ^g _{~ opt} and weighted shape vector Wc _n which maximizes.

  Next, a specific example of the configuration of the shape calculation unit 120 is shown. As shown in FIG. 2, the shape calculation unit 120 includes an initialization unit 121, an inner product unit 122, an ideal gain calculation unit 123, a distance calculation unit 124, and a confirmation unit 125. The processing of each means is shown in S120 of FIG. 3 and is as follows.

The initialization unit 121 sets n = 1 and d _smax = 0 (S121). The inner product means 122 records the inner product result (Wc _n ) ^t (We) of the _nth weighted shape vector Wcn in the weighted shape codebook 150 and the weighted noise signal We as a scalar variable q _n ( S122). Ideal gain calculating means, the n-th element of the weighted shape power reciprocal Book 160 (1 / ‖Wc _n ‖ ^2), the product of the variables _{q n,} is recorded as an ideal gain ^g _{~ n} (S123). The distance calculation unit 124 records the product of the ideal gains g ^to _n and the variable q _n as the distance d _s (S124). Note that q _n = (Wc _n ) ^t (We) is the inner product of the weighted shape vector and the weighted noise signal, and is the same as the numerator of Expression (5), so d _s is the ideal gain g ^to _n it can be obtained as the product of the variable q _n. If d _smax <d _s , the confirmation unit 125 sets d _s to d _smax , n to the optimal code i _opt , and g ^to _n to the optimal ideal gain g ^to _opt (S 1251, S 1252). If n is smaller than N (the number of elements in the weighted shape codebook), the shape calculation unit 120 sets n = n + 1 and returns to step S122 (S1261, S1262). If n = N, the _{i opt} at repeat end a shape code _{I s,} and outputs the coded shape _{I s} and optimal ideal gain ^g _{~ opt} for repeated at the end (S1263).

Shape calculation unit 120, the use of the fact that it is possible in this way the distance _{d s} is determined as the product of the ideal gain ^g _{~ n} and variables _{q n,} weighted shape vector Wc _n and optimal ideal gain ^g _{~ opt} The amount of calculation for selecting can be reduced as compared with Patent Document 1.

Gain calculation unit 130
The gain calculation unit 130 selects the gain g _m (m is an integer of 1 to M) in the gain codebook 170 so that the distance D ^{from the} optimum ideal gain g ^to _opt is minimized, and indicates the gain. The gain code _Ig is output. (S130). Here, the distance D is
D = _‖We-g m Wc _n ‖ ² (8)
It is. Then, the distance d _{g is changed} to d _g = ‖ g ^to _opt ⁻ g _m ‖ ² (9)
And a gain g _m that minimizes the distance d _g is selected.

The gain calculation unit 130 includes search means 135. Search means 135 selects gain g _m (m is an integer from ₁ to M) using gain g _x having the smallest value in x in gain codebook 170 and gain g _{x + 1} having the smallest value in _{x + 1.} . For example, in order from the smallest gain in the gain codebook,
g ^to _opt <(g _x + g _{x + 1} ) / 2 (10)
Check if you are satisfied. First, x that satisfies Expression (10) is searched.

A specific example of the search process is as follows. The search means 135 sets x = 1 (S131). It is confirmed whether the expression (10) is satisfied (S132). If step S132 is No, x = x + 1 is set (S133). x is M (M is the number of gain _{g m} in the gain codebook 170) checks whether less than (S134). If step S134 is Yes, the process returns to step S132. Step S132 is the case with step S134: Yes If No, and outputs a x as a gain code _{I g} (S135).
Since the gain calculation unit 130 calculates using the formula (10) in this way, the amount of calculation can be reduced as compared with Patent Document 1.

[Decoding device]
FIG. 5 shows a configuration example of the decoding apparatus according to the first embodiment. The decoding apparatus includes a basic decoder 920, an adder 940, and an extended decoder 200. The basic decoder 920 is a decoder corresponding to the basic encoder 910. The basic decoder 920 decodes the basic code _Ib into the reproduction basic signal S ^ _b . The extended decoder 200 decodes the extended code I _e into a reproduced noise signal e ^. The adder 940 adds the reproduction basic signal S ^ _b and the reproduction noise signal e ^ and outputs a reproduction signal s ^.

  Since the present invention particularly relates to the extended decoder 200 in the decoding apparatus of FIG. 5, the extended decoder 200 will be described below. FIG. 6 is an example of a processing flow of the extended decoder 200. The extended decoder 200 includes a decomposition unit 210, a weighted shape decoding unit 220, a weighting removal calculation unit 230, a gain decoding unit 240, a multiplication unit 250, a weighted shape codebook 150, and a gain codebook 170. The extended decoder 200 according to the first embodiment is significantly different from Patent Document 1 in that a weighted codebook can be used for the shape codebook for decoding. Since a weighted codebook can be used, the weighted shape codebook 150 used for encoding can also be used for decoding.

Decomposition unit 210 decomposes the input code _{I e,} the coded shape _{I s} and the gain code _{I g} (S210). Weighted shape decoding unit 220 uses the weighted shape codebook 150, integers coded shape I _s weighted shape vector Wc n _(n a is 1 to N, N is stored in the weighted shape codebook Number of weighted shape vectors). Weighting removal operation unit 230 removes the weight W of a weighted shape vector Wc _n, and outputs the shape vector _{c n} (S230). Gain decoding section 240, by using a gain codebook 170, and converts the gain code I _g gain g _{m (m} is an integer of 1 to M, M is the number of gain stored in the gain codebook) to. Multiplication section 250 multiplies the shape vector _{c n} and the gain _{g m,} the decoded signal (reproduced noise signal) e ^ (S250).

と表現されるＫ行（Ｋ＋ｒ）列の行列である。式（１１）の場合は、1タップのフィルタであるため、ｒ＝１である。重み付け除去は、重み付き形状ベクトルにこの逆行列Ｕを乗算し、形状ベクトルを出力することである。 The reason why the extended decoder 200 can use a weighted codebook resides in the weight removal operation unit 230. For example, the weight removal arithmetic unit 230 may use an inverse matrix (inverse filter) of a matrix (filter) to which weights used for encoding are applied. For example, the inverse matrix U of the matrix W in equation (2) is

Is a matrix of K rows (K + r) columns. In the case of Expression (11), since it is a 1-tap filter, r = 1. The weight removal is to multiply the weighted shape vector by this inverse matrix U and output the shape vector.

Removing the weight W of a weighted shape vector Wc _n by using the inverse matrix of Equation (11), in order to output the shape vector _{c n,} the weighting removal operation unit 230, the initialization unit 231, addition unit 232, division Means 233 and update means 234 are provided. The initialization unit 231 sets k = 1 and f _a = 0 (S231). Addition means, the k th element _{f k} of the weighted shape vector Wc _n is added to _{f a,} the new _{f a} (S232). The dividing unit 233 records the n-th element p _k of the shape vector as p _k = f _a / b (S233). When k is smaller than K, the updating unit adds 1 to k to obtain a new k (S2341, S2342). If k is equal to K (K is the vector length of the shape vector), the process ends. In this way, steps S232 to S2342 are repeated until all of p ₁ ,..., P _K (where K is the vector length of the shape vector) is obtained.

  Since the decoding apparatus according to the first embodiment includes the weight removal operation unit 230 as described above, the same weighted shape codebook 150 as that of the encoder can be used. If the weighted shape codebook used for encoding is also used for decoding, the amount of memory used for the codebook of the bidirectional communication device including both the encoding device and the decoding device is significantly reduced (specifically Can be almost halved).

  In general, the amount of calculation required for decoding is significantly smaller than the amount of calculation required for encoding for code search. In the case of the first embodiment, the amount of calculation required for encoding and the amount of calculation required for decoding are substantially in a relationship of N: 1. That is, even if the calculation amount for decoding increases slightly, there is almost no influence on the total calculation amount including encoding. As described above, the encoding device of the first embodiment reduces the amount of calculation while maintaining the quality equivalent to that of Patent Document 1. Therefore, overall, the amount of calculation and the quality can be kept equal while greatly reducing the amount of memory used.

  FIG. 7 shows G. in the basic code part in order to show the effect of the present invention. 7A shows an example of spectrum analysis of a reproduction signal (decoded signal) when the encoding apparatus and decoding apparatus of the first embodiment are used. FIG. 7A shows the original voice (dashed line) and its voice. Reproduced sound (solid line) encoded using 711 and decoded (solid line), FIG. 7B shows the original sound (broken line) and the reproduced sound obtained by encoding and decoding the sound using the encoding device of the first embodiment (broken line). ) Spectral analysis results. G. When the 711 unit is used, the high-frequency harmonic structure existing in the current sound is buried in the quantization noise. However, if the first embodiment is used, the high-frequency (2,500 KHz or higher) harmonic structure is reproduced. I understand that. This result is equivalent to Patent Document 1. Therefore, the amount of calculation can be reduced while maintaining the quality of encoding and decoding.

[Second Embodiment]
[Encoding device]
FIG. 8 shows a functional configuration example of the encoding apparatus according to the second embodiment. Gain calculating section 130 of the first embodiment, it explored the gain code _{I g} from the gain codebook 170. Gain calculator 330 of the second embodiment determines the gain code I _g by calculation. Therefore, the extension encoder 300 according to the second embodiment includes a gain calculation unit 330 instead of the gain calculation unit 130 and the gain codebook 170. The gain calculation unit 330 includes a quantization unit 335. Other configurations are the same as those in FIG. FIG. 9 shows an example of the processing flow of the extension encoder 300 of the second embodiment. The processing flow of the extension encoder 100 of the first embodiment shown in FIG. 3 is the same as steps S110, S120, and S140. Below, the gain calculation part 330 which is the difference with 1st Embodiment is demonstrated.

Quantization means of the gain calculator 330 335, the optimum ideal gain ^g _{~ opt} using a projection function f (x), determining the gain code _{I g.} The projection function f (x) is a straight line, a curve, or a continuous function that combines a straight line and a curve. The quantization means 335 first performs a projection function calculation f (g ^to _opt ) (S331). Next, the rounding operation as follows seek gain code _{I g} (S332).
_{^{I g = round (f (g}} ~ opt)) (12)
Gain calculating unit 330 outputs a gain code _{I g} obtained (S333).

そして、例えば、利得符号の数Ｍが６４の場合は、
ａ＝１３００
ｂ＝−１
ｃ＝１．０７
ｄ＝１９．２７
とすればよい。なお、射影関数は、連続関数であればよく、ｘの値に応じて複数の関数を切り替えてもよい。
利得計算部３３０は、このように射影関数を利用して計算するので、演算量を特許文献１よりも低減できる。また、利得符号帳が必要ないので、メモリ使用量を低減できる。 As a specific example of the projection function, there is a hyperbola as follows.

For example, when the number M of gain codes is 64,
a = 1300
b = -1
c = 1.07
d = 19.27
And it is sufficient. The projection function may be a continuous function, and a plurality of functions may be switched according to the value of x.
Since the gain calculation unit 330 calculates using the projection function in this way, the amount of calculation can be reduced as compared with Patent Document 1. Further, since no gain codebook is required, the amount of memory used can be reduced.

[Decoding device]
FIG. 10 shows a configuration example of the decoding apparatus according to the second embodiment. Gain decoding unit 240 of the first embodiment, to determine the gain _{g m} using a gain codebook 170. The gain decoding unit 440 of the second embodiment obtains the gain g _m by calculation. Therefore, the extended decoder 400 according to the second embodiment includes a gain decoding unit 440 instead of the gain decoding unit 240 and the gain codebook 170. Other configurations are the same as those in FIG. FIG. 11 shows an example of the processing flow of the extended decoder 400 of the second embodiment. The processing flow of the extended decoder 200 of the first embodiment shown in FIG. 6 is the same as steps S210, S220, S230, and S250. Below, the gain decoding part 440 which is a difference with 1st Embodiment is demonstrated.

Gain decoding section 440, inverse projection functions used in coding ^f -1 with (y), determining the gain _{g m} from the gain code _{I g.} Specifically, the gain decoding unit 440
g _m = f ⁻¹ (I _g ) (14)
The gain g _m is obtained by the calculation of (S440).

このときのａ、ｂ、ｃ、ｄは、利得計算部３３０と同じ値を用いる。 If the hyperbola of equation (13) is used as the projection function, the inverse function is

At this time, the same values as those of the gain calculation unit 330 are used for a, b, c, and d.

  With such a configuration and processing method, the memory usage can be reduced with the same amount of computation as in the first embodiment. Further, since the processing related to the quality of encoding and decoding is the same as that in the first embodiment, the same result as the spectrum analysis result shown in FIG. 7 can be expected. Therefore, the amount of computation can be reduced while maintaining the quality of encoding and decoding.

[Modification]
One important point of the decoding method of the present invention is steps S220 and S230. Further, components necessary for executing these steps are a weighted shape decoding unit 220, a weighted shape codebook 150, and a weighting removal calculating unit 230. The other components need not be limited to the decoding device of the first embodiment (FIG. 5) or the decoding device of the second embodiment (FIG. 10). FIG. 12 shows a functional configuration example of a decoding device according to a modification of the second embodiment. In this configuration, there is a division unit 650 instead of the multiplication unit 250 in FIG. 10, and an addition unit 640 instead of the addition unit 940. The division unit 650 divides the reproduction basic signal ＾ _b by the gain g _m . Adding section 640 adds the outputs _s ^ b / _{g m} and shape vector _{c n} of the divider 650 to obtain a reproduction signal e ^ '.

^ Reproduced signal e of the decoding device in FIG. 10 is the sum of the product of the shape vector c _n and the gain g _m and the reproduction fundamental signal s ^ _b. Therefore, the reproduction signal e ^ and the reproduction signal e ^ '
_{e ^ '= e ^ / g} m (16)
It becomes the relationship. That is, the reproduction signal e ^ 'is the same signal as the reproduction signal e ^ although the volume is different. In the encoding and decoding processes, the volume of the reproduced signal is adjusted by other processes, and there is no problem in quality if the waveform is reproduced. Therefore, even with the configuration shown in FIG. 12, the same effect as that of the decoding device of the second embodiment can be obtained.

  As described above, the decoding method including steps S220 and S230 illustrated in FIGS. 6 and 11, the decoding including the weighted shape decoding unit 220, the weighted shape codebook 150, and the weighting removal calculating unit 230 illustrated in FIGS. 5 and 10. If it is an apparatus, the effect of this invention can be acquired.

  FIG. 13 shows a functional configuration example of a computer. In the encoding method and decoding method of the present invention, a program for causing the recording unit 2020 of the computer to execute each step of the above method is read and operated by the control unit 2010, the input unit 2030, the output unit 2040, and the like. Can be implemented. In addition, as a method of causing the computer to read, the program is recorded on a computer-readable recording medium, and the program recorded on the server or the like is read into the computer through a telecommunication line or the like. There is a method to make it.

The characteristic curve figure for demonstrating the IRS frequency characteristic used for the telephone network. The figure which shows the function structural example of the encoding apparatus of 1st Embodiment. The figure which shows the example of the processing flow of the extended encoder of the encoding apparatus of 1st Embodiment. The figure which shows the frequency characteristic of a weighting part. The figure which shows the function structural example of the decoding apparatus of 1st Embodiment. The figure which shows the example of the processing flow of the extended decoder of the decoding apparatus of 1st Embodiment. G. 7 is a diagram illustrating a spectrum analysis example of a decoding result when the encoding device and the decoding device according to the first embodiment are used. The figure which shows the function structural example of the encoding apparatus of 2nd Embodiment. The figure which shows the example of the processing flow of the extended encoder of the encoding apparatus of 2nd Embodiment. The figure which shows the structural example of the decoding apparatus of 2nd Embodiment. The figure which shows the example of the processing flow of the extended decoder of the decoding apparatus of 2nd Embodiment. The figure which shows the function structural example of the decoding apparatus of the modification of 2nd Embodiment. The figure which shows the function structural example of a computer.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Extended encoder 110 Weighting part 120 Shape calculation part 121 Initialization means 122 Inner product means 123 Ideal gain calculation means 124 Distance calculation means 125 Confirmation means 130 Gain calculation part 135 Search means 140 Multiplexing part 150 Weighted shape codebook 160 Shape Power reciprocal book 170 Gain code book 200 Extended decoder 210 Decomposition unit 220 Shape decoding unit 230 Removal operation unit 231 Initialization unit 232 Addition unit 233 Division unit 234 Update unit 240 Gain decoding unit 250 Multiplication unit

Claims (20)

A decoding method for decoding a shape code indicating a shape vector,
A weighted shape decoding step of converting the shape code into a weighted shape vector using a weighted shape codebook;
A weighting removal calculating step of removing weights of the weighted shape vector and outputting the shape vector;
A decryption method. A decoding method for decoding a code composed of a shape code indicating a shape vector and a gain code indicating a gain,
A decomposing step of decomposing the input code into a shape code and a gain code;
A weighted shape decoding step of converting the shape code into a weighted shape vector using a weighted shape codebook;
A weighting removal calculating step of removing weights of the weighted shape vector and outputting the shape vector;
A gain decoding step of converting the gain code into a gain;
A multiplication step of multiplying the shape vector and the gain and outputting a decoded signal;
A decryption method. The decoding method according to claim 1 or 2, comprising:
The decoding method according to claim 1, wherein the weighted shape codebook is the same as the weighted shape codebook used for encoding. A decoding method according to any one of claims 1 to 3,
The weighting removal calculating step is a step of outputting, as the shape vector, a result obtained by multiplying the weight of the weighted shape vector by the inverse matrix U of the weight matrix W used for encoding. . The decoding method according to claim 4, wherein
The inverse matrix U is

A decoding method characterized by being expressed as: The decoding method according to claim 5, wherein
The weighting removal calculating step includes:
an initialization sub-step with k = 1 and f _a = 0;
An adding substep of component k f _k of the weighted shape vector is added to f _a, the new f _a,
A division sub-step in which the k-th element p _k of the shape vector is set to p _k = f _a / b;
an update step of adding 1 to k to give a new k;
Have
The decoding method, wherein the addition sub-step, the division sub-step, and the update step are repeated until all of p ₁ ,..., P _K (where K is a vector length of a shape vector) is obtained. A code encoded by the first encoding method (hereinafter referred to as a “basic code”) and a code obtained by encoding an error generated by the encoding by the first encoding method by the second encoding method ( Hereinafter, a decoding method for decoding “quality extension code”),
The quality extension decoding step of decoding a quality extension code and outputting a reproduction quality extension signal includes the steps of the decoding method according to any one of claims 1 to 6,
A basic decoding step of decoding a basic code and outputting a reproduced basic signal;
A decoding method comprising: an addition step of adding the reproduction basic signal and the reproduction quality extension signal to output a reproduction signal. A decoder for decoding a shape code indicating a shape vector,
A weighted shape codebook that associates shape codes with weighted shape vectors;
A weighted shape decoding unit that converts the shape code into a weighted shape vector using the weighted shape codebook;
Removing a weight of the weighted shape vector and outputting a shape vector;
A decoder. A decoder for decoding a code composed of a shape code indicating a shape vector and a gain code indicating a gain,
A decomposition unit that decomposes the input code into a shape code and a gain code;
A weighted shape codebook that associates shape codes with weighted shape vectors;
A weighted shape decoding unit that converts the shape code into a weighted shape vector using the weighted shape codebook;
Removing a weight of the weighted shape vector and outputting a shape vector;
A gain codebook that associates the gain code with the gain;
A gain decoding unit that converts the gain code into a gain using the gain codebook;
A multiplier that multiplies the shape vector and the gain to output a reproduction quality extension signal;
A decoder. The decoder according to claim 8 or 9, comprising:
The decoder according to claim 1, wherein the weighted shape codebook is the same as the weighted shape codebook used for encoding. A decoder according to any one of claims 8 to 10,
The weight removal operation unit outputs a result obtained by multiplying the weight of the weighted shape vector by the inverse matrix U of the weight matrix W used for encoding, as the shape vector. . The decoder according to claim 11, comprising:
The inverse matrix U is

Decoder that can be expressed as The decoder according to claim 12, comprising:
The weight removal operation unit
initialization means for k = 1 and f _a = 0;
Adding means for the k-th element f _k of the weighted shape vector is added to f _a, the new f _a,
A dividing means for setting the k-th element p _k of the shape vector to p _k = f _a / b;
an updating means for adding 1 to k and setting it as a new k;
It said updating means and said adding means and said dividing means has _{a, p 1, ..., p K} ( although, K is the shape vector length of the vector), characterized in that to Kaee treated repeatedly until finding all Decoder. A code encoded by the first encoding method (hereinafter referred to as a “basic code”) and a code obtained by encoding an error generated by the encoding by the first encoding method by the second encoding method ( Hereinafter, it is a decoding apparatus that decodes “quality extension code”),
A decoder according to any one of claims 8 to 13, comprising a decoder according to any one of claims 8 to 13, as a quality extension decoder that decodes a quality extension code and outputs a reproduction quality extension signal.
A basic decoder for decoding a basic code and outputting a reproduced basic signal;
A decoding apparatus comprising: an adder that adds the reproduction basic signal and the reproduction quality extension signal and outputs a reproduction signal. An encoding method for converting an input signal into a code composed of a shape code indicating a shape vector and a gain code indicating a gain,
A weighting step of giving a weight to the input signal e and outputting a weighted noise signal We;
As the distance between the weighted noise signal We is minimized to obtain the optimal ideal gain g ^~ _opt with selecting the weighted shape vector Wc _n weighted in shape codebook, the shape representing the weighted shape vector A shape calculation step for outputting a code _Is and an optimum ideal gain g ^to _opt ;
As the distance between the optimal ideal gain g ^~ _opt is minimized, and the gain calculation step of selecting a gain g _m in the gain codebook, and outputs the gain code I _g indicating the gain,
A multiplexing step of multiplexing the shape code I _s and the gain code _Ig and outputting the code I _e ,
The shape calculation step includes:
an initialization sub-step with n = 1 and d _smax = 0;
An inner product sub-step for recording an inner product result (Wc _n ) ^t (We) of an _nth weighted shape vector Wcn in the weighted shape codebook and a weighted noise signal We as a variable q _n ;
An ideal gain calculation sub-step for recording the product of the _nth element (1 / ‖Wc _n ‖ ² ) in the weighted shape power reciprocal book and the variable q _n as ideal gains g ^to _n ;
A distance calculation substep for recording a product of the ideal gains g ^to _n and the variable q _n as a distance d _s ;
If d _smax <d _s , a confirmation sub-step with d _s as d _smax , n as the optimal code i _opt , and g ^through _n as the optimal ideal gains g ^through _opt ,
Have
Repeating the inner product substep, the ideal gain calculation substep, the distance calculation substep, and the confirmation substep until n is the number N of weighted shape vectors in the weighted shape codebook,
The repetition at the end of the _{i opt} and shape sign _{I s,}
A coding method characterized in that the shape code _Is and the optimum ideal gain g ^to _opt at the end of repetition are output. An encoding method for converting an input signal into a code composed of a shape code indicating a shape vector and a gain code indicating a gain,
A weighting step of giving a weight to the input signal e and outputting a weighted noise signal We;
As the distance between the weighted noise signal We is minimized, seeking the ideal gain g ^~ _opt for suitable with selecting the weighted shape vector Wc _n weighted in shape codebook, the shape representing the weighted shape vector A shape calculation step for outputting a code _Is and an optimum ideal gain g ^to _opt ;
As the distance between the optimal ideal gain g ^~ _opt is minimized, and the gain calculation step of selecting a gain g _m in the gain codebook, and outputs the gain code I _g indicating the gain,
A multiplexing step of multiplexing the shape code I _s and the gain code _Ig and outputting the code I _e ,
The gain calculating step includes:
Let g _x be the gain of the xth smallest value in the gain codebook,
In order from a small gain of the value of the gain codebook ^searches for the first x satisfying _{g ~ opt <(g x +} g x + 1) / 2,
An encoding method, wherein x is output as a gain code _Ig . An encoder that converts an input signal into a code composed of a shape code indicating a shape vector and a gain code indicating a gain,
A weighted shape codebook that associates shape codes with weighted shape vectors;
A weighted shape power reciprocal book that records the reciprocal of the weighted shape vector power;
A gain codebook that associates the gain code with the gain;
A weighting unit that gives a weight to the input signal e and outputs a weighted noise signal We;
As the distance between the weighted noise signal We is minimized to obtain the optimal ideal gain g ^~ _opt with selecting the weighted shape vector Wc _n weighted in shape codebook, the shape representing the weighted shape vector A shape calculation unit for outputting a code _Is and an optimum ideal gain g ^to _opt ;
As the distance between the optimal ideal gain g ^~ _opt is minimized, and the gain calculation unit that selects the gain g _m in the gain codebook, and outputs the gain code I _g indicating the gain,
A multiplexing unit that multiplexes the shape code _Is and the gain code _Ig and outputs the code _Ie ;
The shape calculator is
initialization means for n = 1 and d _smax = 0,
Inner product means for recording the inner product result (Wc _n ) ^t (We) of the _nth weighted shape vector Wcn in the weighted shape codebook and the weighted noise signal We as a variable q _n ;
And n-th element of the weighted shapes power reciprocal within book (1 / ‖Wc _n ‖ ^2), the product of the variables q _n, and the ideal gain calculation means for recording as an ideal gain g ^~ _n,
A distance calculation means for recording a product of the ideal gains g ^to _n and the variable q _n as a distance d _s ;
If d _smax <d _s , confirmation means for d _s to be d _smax , n to be an optimal code i _opt , and g ^to _n to be optimal ideal gains g ^to _opt
Have
Causing the inner product means, the ideal gain calculation means, the distance calculation means, and the confirmation means to repeatedly perform processing until n becomes the number N of weighted shape vectors in the weighted shape codebook,
The repetition at the end of the _{i opt} and shape sign _{I s,}
An encoder characterized by having a shape code _Is and an optimum ideal gain g ^to _opt at the end of repetition as outputs. An encoder that converts an input signal into a code composed of a shape code indicating a shape vector and a gain code indicating a gain,
A weighted shape codebook that associates shape codes with weighted shape vectors;
A weighted shape power reciprocal book that records the reciprocal of the weighted shape vector power;
A gain codebook that associates the gain code with the gain;
A weighting unit that gives a weight to the input signal e and outputs a weighted noise signal We;
As the distance between the weighted noise signal We is minimized to obtain the optimal ideal gain g ^~ _opt with selecting the weighted shape vector Wc _n weighted in shape codebook, the shape representing the weighted shape vector A shape calculation unit for outputting a code _Is and an optimum ideal gain g ^to _opt ;
As the distance between the optimal ideal gain g ^~ _opt is minimized, and the gain calculation unit that selects the gain g _m in the gain codebook, and outputs the gain code I _g indicating the gain,
A multiplexing unit that multiplexes the shape code _Is and the gain code _Ig and outputs the code _Ie ;
The gain calculator is
Let g _x be the gain of the xth smallest value in the gain codebook,
In order from a small gain of the value of the gain codebook ^searches for the first x satisfying _{g ~ opt <(g x +} g x + 1) / 2,
encoder and outputs a x as a gain code I _g.   The program which makes a computer operate | move each step of the method in any one of Claim 1 to 7, 15, 16. A computer-readable recording medium on which the program according to claim 19 is recorded.

Images