JP4738260B2 - Prediction delay search method, apparatus using the method, program, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce processing time while maintaining long term prediction performance. <P>SOLUTION: A delay amount-gain calculation section comprises: a selection point evaluation calculating means; an optimum selection point search means; a hierarchical evaluation calculating means; an optimum evaluation value search means; and an output means. The selection point evaluation calculating means calculates an evaluation point (a selection point evaluation value) of a sample point (a selection point) of a selected delay amount. The optimum selection point search means calculates an optimum evaluation value and the selection point (an optimum selection point) from a plurality of selection point evaluation values. The hierarchical evaluation calculating means calculates an evaluation value (a hierarchical evaluation value) at the sample point between the optimum selection point and a selection point adjoining the optimum selection point. The optimum evaluation value search means calculates an optimum evaluation value (an optimum evaluation value) which is optimum and the delay amount from all the selection point evaluation values and the hierarchical evaluation value. The output means sets the delay amount obtaining the optimum evaluation value to the optimum delay amount, and an optimum gain and the optimum delay amount calculated from the optimum evaluation value are output. <P>COPYRIGHT: (C)2007,JPO&amp;INPIT

Description

  The present invention relates to a prediction delay search method for time-series signals such as acoustic signals, an apparatus using the method, a program, and a recording medium.

  Long-term prediction (or pitch prediction) may be used for compression coding of time-series signals such as acoustic signals (Non-Patent Document 1). In this method, a delay amount and a multiplier for the delay sample are used as parameters. Then, in order to obtain the optimum delay amount, the signal sequence is shifted by τ points, the difference is multiplied by the optimum multiplier, and τ with the smallest distortion is searched.

により、短期予測誤差ｙ（ｎ）を求める（Ｓ９２２）。ステップＳ９２２で用いるα１〜αＰは、ステップＳ９２１で符号化した符号により決まる値、すなわち、量子化済の値が好ましい。また、一般には短期予測分析部９２０を備えるが、必ずしも備える必要はなく、省略可能である。その場合は、フレーム化部９１０の出力ｘ（ｎ）が短期予測残差ｙ（ｎ）の代わりに用いられることになる。 The configuration of a conventional compression encoding apparatus is shown in FIG. FIG. 2 shows a processing flow of the conventional compression coding apparatus. A conventional compression coding apparatus includes a framing unit 910, a short-term prediction analysis unit 920, a delay amount / gain calculation unit 930, a delay unit 940, a gain multiplication unit 950, a subtraction unit 960, a long-term prediction error coding unit 970, a code string The generation unit 980 is configured. The short-term prediction analysis unit 920 includes a short-term prediction coefficient calculation unit 921 and a short-term prediction analysis filter 922. When the input signal x (n), which is a digitized time series signal, is input, the framing unit 910 collects the input signal x (n) every predetermined number N (S910). Examples of the predetermined number N include 1024 and 512. The short-term prediction coefficient calculation unit 921 obtains _P- order linear prediction coefficients α _{1 to} α _P from, for example, a P-order (eg, 10th-order) autocorrelation function, and outputs them to the short-term prediction analysis filter 922. Further, the obtained short-term prediction coefficients α _{1 to} α _P are encoded and output to the code string generation unit 980 (S921). In step S921, encoding may be performed after conversion to an LSP coefficient or an LSF coefficient instead of the short-term prediction coefficient itself. The short-term prediction analysis filter 922

Thus, a short-term prediction error y (n) is obtained (S922). Α1 to αP used in step S922 are preferably values determined by the code encoded in step S921, that is, quantized values. In general, the short-term prediction analysis unit 920 is provided, but is not necessarily provided and can be omitted. In that case, the output x (n) of the framing unit 910 is used instead of the short-term prediction residual y (n).

が最大となる遅延量τを探索し、最大となる利得γ（τ）を最適利得γ_ｏｐｔ、そのときの遅延量τを最適遅延量τ_ｏｐｔとすることにより、長期予測誤差を最小とする最適遅延量τ_ｏｐｔと最適利得γ_ｏｐｔとを求めることができる。 The delay amount / gain calculation unit 930 obtains the optimum delay amount τ _opt and the optimum gain γ _opt that minimize the long-term prediction error while shifting by one sample. Then, the obtained optimal delay amount τ _opt is output to the delay unit 940, the optimal gain γ _opt is output to the gain multiplier 950, and the delay code and the gain code are output to the code string generator 980 (S930). Specifically, the gain from the preset delay amount τ is selected.

Is searched for the delay amount τ that maximizes the maximum gain, and the maximum gain γ (τ) is the optimum gain γ _opt , and the delay amount τ at that time is the optimum delay amount τ _opt so that the long-term prediction error is minimized. The delay amount τ _opt and the optimum gain γ _opt can be obtained.

The delay unit 940 delays the short-term prediction error y (n) by the received optimum delay amount τ _opt and outputs y (n−τ _opt ) to the gain multiplication unit 950 (S940). The gain multiplication unit 950 multiplies the delayed short-term prediction error y (n−τ _opt ) by the optimum gain γ _opt (S950). The subtraction unit 960 subtracts γ _opt · y (n−τ _opt ) from the short-term prediction error y (n) to obtain a long-term prediction error y (n) −γ _opt · y (n−τ _opt ) (S960). . The long-term prediction error encoding unit 970 encodes the long-term prediction error y (n) −γ _opt · y (n−τ _opt ) (S970). The code string generation unit 980 generates a code string from the received long-term prediction error code, delay code, gain code, and short-term prediction code, and outputs the code string (S980).

The calculation amount of this method is proportional to the search range of τ and the number of samples in the frame. That is, when the sampling frequency is high, the search range of τ (the number of sample points to be searched) and the number of samples per frame increase. Therefore, it takes an enormous amount of time to search for the optimum delay amount τ _opt and the optimum gain γ _opt in step S930. For example, when the sampling frequency is 192 kHz, the number of samples in one frame is 8192, and there are 1024 candidates for the delay amount τ, the amount of calculation becomes very large.

FIG. 3 shows an example of the functional configuration of a conventional decoding device. The conventional decoding apparatus includes a code string decomposition unit 810, a long-term prediction error decoding unit 820, a delay unit 830, a gain multiplication unit 840, an addition unit 850, and a short-term prediction synthesis unit 860. The code string decomposing unit 810 decomposes the received code string into a long-term prediction error code, a delay code, a gain code, and a short-term prediction code. The long-term prediction error decoding unit 820 decodes the long-term prediction error code to obtain a long-term prediction error y (n) −γ _opt · y (n−τ _opt ). The delay unit 830 decodes the delay code and obtains a delayed short-term prediction error y (n−τ _opt ). The gain multiplication unit 840 decodes the gain code, and multiplies the delayed short-term prediction error y (n−τ _opt ) by the gain γ _opt . The adder 850 adds the output γ _opt · y (n−τ _opt ) of the gain multiplier 840 to the long-term prediction error y (n) −γ _opt · y (n−τ _opt ), and the short-term prediction error y ( n). When the short-term prediction analysis unit 920 is not provided in the encoding device, the above-described y (n) becomes the output signal x (n) as it is because the short-term prediction synthesis unit 860 is not provided in the decoding device. The short-term prediction synthesis unit 860 decodes the short-term prediction code to obtain a short-term prediction signal, and synthesizes the short-term prediction signal and the short-term prediction error y (n) to obtain an output signal x (n).

The above example is an example in which a delay time is given to one input signal and encoded, but a plurality of signals (multi-channel signal) may be encoded as in Patent Document 1. In the case of encoding a multi-channel signal, there is an example in which a weighted difference signal with respect to a signal obtained by delaying another channel signal is encoded. FIG. 4 shows a functional configuration example of a conventional multi-channel signal encoding apparatus. A multi-channel signal encoding apparatus 1800 having an input of M channels (M is an integer of 2 or more) includes a frame buffer 1810 _i (i = 1 to M), an encoding information determination unit 1820, an encoding target signal generation unit 1830, The signal encoding unit 1840 _i (i = 1 to M) and the synthesis unit 1850 are included. Also, the coding information determination unit 1820 performs coding independently for each channel (hereinafter referred to as “independent coding”) or a weighted difference from other channels (hereinafter referred to as “master channel”). An independent / difference / master channel determination unit 1821 that determines whether to encode a signal (hereinafter referred to as “differential encoding”), and a weight determination unit 1826 that determines the weight of the master channel in the case of differential encoding. It has. The encoding target signal generation unit 1830 is an encoding information processing unit 1832 _i that collects necessary information for each channel in accordance with the encoding information determined by the encoding information determination unit 1820, and weights in the case of differential encoding. A weighted addition unit 1833 _i for performing addition (subtraction).

FIG. 5 shows a processing flow of a conventional multi-channel signal encoding apparatus 1800. The frame buffer 1810 _i (i = 1 to M) stores an input signal (channel signal). Here, when the channel signal is simply a sequence of sample values, the channel signal is divided into a plurality of sample sequences (hereinafter referred to as “frames”), and when the channel signal is already divided into frames, it is a frame unit. (S1810). The encoding information determination unit 1820 uses information that approximates the correlation such as the energy of each channel signal and the difference energy between the channels, and encodes each channel's encoding information (independent encoding or differential encoding, Master channel number, weight, etc.) are determined (S1820). The encoding target signal generation unit 1830 generates a signal to be encoded according to the encoding information for each channel (S1830). The signal encoding unit 1840 _i (i = 1 to M) encodes the generated encoding target signal (S1840). The combining unit 1850 combines the code of each channel signal and the encoded information, and outputs a multi-channel code (S1850).

The detailed processing flow of step S1830 is as follows. The encoding information processing unit 1832 _i of the encoding target signal generation unit 1830 acquires the encoding information determined by the encoding information determination unit 1820. The encoded information processing unit 1832 _i acquires a sample sequence of the encoding target channel. In the case of differential encoding, the encoded information processing unit 1832 _i acquires information on the sample sequence referred to by the master channel. Note that there are one case and three cases of sample columns to be referred to. Details will be described later. In the case of independent encoding, the weighted addition unit 1833 _i outputs the sample sequence of the encoding target channel acquired by the encoding information processing unit 1832 _{i as it} is as the sample sequence of the encoding target signal. In the case of differential encoding, the weighted addition unit 1833 _i multiplies the sample sequence referred to by the master channel by the weight to the sample sequence of the channel to be encoded and adds (subtracts) the sample sequence of the encoding target signal. Output as.

FIG. 6 shows an image of the processing in step S1830 when there is one sample string (one tap). FIG. 7 shows an image of the processing in step S1830 when there are three sample rows (3 taps). Since one frame is composed of N samples, the sample sequence of the channel X to be encoded (sequence of N sample values) is the signal of the encoding channel. In the example of FIG. 6, the sample sequence X of the encoding target signal is subtracted by multiplying the sample sequence Y ₀ of the master channel at the same time (τ = 0) as the sample sequence of the encoding target channel by the weight γ (weight −γ The difference signal X ^ obtained by multiplying and adding) becomes the encoding target signal. Here, τ indicates the time difference (time position difference) between the frame signal of the channel to be encoded (sample sequence that matches the frame) and the sample sequence of the master channel. The subscript of the sample column Y indicates the value of τ. For example, Y _i indicates a sample string of the master channel Y with τ = i. In the example of FIG. 7, the sample sequence X to be coded channels, one sample was shifted forward (tau = -1) sample sequence _{Y -1,} sample sequence _{Y 0} in the same time (tau = 0), and 1 samples shifted backward (tau = 1) sample sequence _Y each weight gamma _-1 to _1, γ _0, subtraction is multiplied by the gamma ₁ (weight -γ _-1, -γ _0, adding multiplied by-gamma ₁₎ The difference signal X ^ is the signal to be encoded.

FIG. 8 shows a functional configuration example of a conventional multi-channel signal decoding apparatus. An M-channel multi-channel signal decoding apparatus 1900 includes an information acquisition / separation unit 1910, a signal decoding unit 1920 _i (i = 1 to M), and a channel signal output unit 1930. The channel signal output unit 1930 includes an encoded information processing unit 1932 _i and a weighted addition unit 1933 _i . The information acquisition / separation unit 1910 receives a multi-channel code, acquires encoded information, and separates the encoded signal for each encoded signal. The signal decoding unit 1920 _i (i = 1 to M) performs decoding for each signal. The channel signal output unit 1930 acquires encoded information for each channel from the information acquisition / separation unit 1910 by the encoded information processing unit 1932 _i , and collects information such as a master channel sample sequence. In addition, in the case of independent encoding, the weighted addition unit 1933 _i outputs the sample sequence of the signal to be decoded as it is, and in the case of differential encoding, the sample sequence of the signal to be decoded and the sample of the master channel Perform weighted addition with columns and output.

As an invention related to the present invention, there is an unpublished patent application filed by the applicant (Japanese Patent Application No. 2005-199163: application date July 7, 2005) (hereinafter referred to as “related invention”). . The differential coding of the related invention is a weighted differential coding with a plurality of sample values of the master channel, and a master channel at a time other than the same time, immediately before, or just after the sample of the channel signal to be coded Is a weighted differential encoding that may contain In the related invention, the master channel sample sequence Y _opt (encoding target sample sequence) having the greatest correlation with the encoding target channel signal sample sequence X (frame signal) within a predetermined time difference range. and using the time difference between the sample sequence of the master channel sample sequence of the master channel Y of (time difference position) tau is tau _opt) to differential coding.

Next, an outline of encoding in the related invention will be described. FIG. 9 shows an image when two sample rows (2 taps) are used. In this example, Y ₀ and Y _opt are used when τ _opt is other than _0, and only Y ₀ is used when τ _opt is 0. FIG. 10 shows an image when six sample rows (6 taps) are used. In this example, when τ _opt is 0, three sample values Y ₋₁ , Y ₀ , and Y ₁ are used for weighted differential encoding, and when τ _opt is −1 or −2, Y ₋₄ , Y ₋₃ , Y ₋₂ , Y ₋₁ , Y ₀ , Y ₁ are used for weighted differential encoding, and when τ _opt is 1 or 2, Y ₋₁ , Y ₀ , Y _When six sample values of ₁ , Y ₂ , Y ₃ , Y ₄ are used for weighted differential encoding and τ _opt is other than the above, Y _opt−1 , Y _opt , Y _{opt + 1} , Y ₋₁ , Y ₀ , Y ₁ are used for weighted differential encoding.

As shown in FIG. 11, when there is a microphone input A and a microphone input B, there is no phase difference between the input signal from the microphone A and the input signal from the microphone B in the sound from the front voice. However, in the piano sound, the input signal from the microphone B is delayed in phase with respect to the input signal from the microphone A. When sounds from different sound sources overlap in this way, the relationship between the input signal from the microphone A and the input signal from the microphone B can be efficiently encoded only by inter-channel prediction as in the related invention. .
JP 2005-115267 A NJ Jayant, P. Noll, “Digital Coding of Waveform”, pp.312-319.

In the prior art, when the sampling frequency is high or the number of samples per frame is large, the optimum delay amount τ _opt is obtained by performing enormous calculations. Alternatively, the calculation amount is reduced by narrowing the search range of τ. Therefore, in this case, there is a problem that the truly optimal optimum delay amount τ _opt cannot be obtained. An object of the present invention is to shorten the processing time while maintaining the long-term prediction performance of obtaining the optimum optimum delay amount τ _opt .

  The prediction delay search apparatus of the present invention includes a selection / evaluation calculation unit, an optimal selection point search unit, a hierarchical evaluation calculation unit, an optimal evaluation value search unit, and an output unit in a delay amount / gain calculation unit. The optimal delay amount when performing or the optimal delay amount when performing prediction from other channel signals is output. The selection point evaluation calculation means calculates an evaluation value (hereinafter referred to as “selection point evaluation value”) at a sample point (hereinafter referred to as “selection point”) of the delay amount selected by a predetermined method. The optimum selection point search means obtains an optimum evaluation value and a selection point (hereinafter referred to as “optimum selection point”) for obtaining the evaluation value from among a plurality of selection point evaluation values. The hierarchical evaluation calculation means calculates an evaluation value (hereinafter referred to as “hierarchical evaluation value”) at the sample point between the optimal selection point and the selection point adjacent to the optimal selection point. The optimum evaluation value search means calculates an optimum evaluation value (hereinafter referred to as “optimum evaluation value”) and a delay amount for obtaining the evaluation value among all the selected point evaluation values and the hierarchical evaluation values calculated. Ask. The output means determines the delay amount for obtaining the optimum evaluation value when the processing of the hierarchical evaluation calculation means and the optimum evaluation value search means at the optimum selection points obtained for all the selection points is the optimum delay. The optimum gain obtained from the optimum evaluation value and the optimum delay amount are output.

In addition, a low-pass filter that generates a band-limited sample sequence or / and a band-limited delay sample sequence is provided, and the selection point evaluation calculation means includes a sample sequence, a low-frequency delay sample sequence, a low-frequency sample sequence, An evaluation value (hereinafter referred to as “selected point evaluation value”) of any one of the delay sample sequence, the low-frequency sample sequence, and the low-frequency delay sample sequence may be calculated. Further, the processing of the hierarchical evaluation calculation means and the optimum delay amount searching means may be repeated using the new optimum selection point obtained by the optimum delay amount searching means until a preset condition is satisfied. At that time, in the hierarchical evaluation calculation means, except for the final iteration, the optimum selection point and the sampling point of the delay amount other than the selection point between the optimum selection point and the selection point adjacent to the optimum selection point (hereinafter referred to as “hierarchical”). The evaluation value of the similarity between the sample sequence, the low-frequency delay sample sequence, the low-frequency sample sequence and the delay sample sequence, and the low-frequency sample sequence and the low-frequency delay sample sequence. Is calculated as a hierarchical evaluation value, and in the final iteration, the evaluation value of the similarity between the sample sequence and the delayed sample sequence at the hierarchical selection point is calculated as the hierarchical evaluation value .

  According to the present invention, after evaluating the delay amount and gain at the selected sample points, only the sample points at which the optimum delay amount is likely to be obtained are evaluated. For evaluation of the selection point, a sample string with a limited band is used. Therefore, the amount of calculation can be greatly reduced as compared with the conventional method that avoids overlooking the optimum delay and evaluates all candidate delay amounts.

  In the following, first, the principle of the present invention will be described, and then embodiments will be described. In addition, in order to avoid duplication of description, the same number is given to the structural part which has the same function, and the process step which performs the same process, and description is abbreviate | omitted.

Principle 1
FIG. 12 shows the principle in the case of performing hierarchically subdivided calculations each time the maximum evaluation value is updated. A sample point is selected from the sample points by a predetermined method. The predetermined method is a method of selecting at regular intervals (once every S samples) or a function using the characteristics of the target time series signal (for example, when the delay amount is small, the interval is narrow and the delay amount is When there are many, a method of selecting according to a function that increases the interval) may be considered. When selecting a sample point, the highest frequency that can be expressed by the sequence at the selected point (half the sampling frequency of the sequence at the selected point) is matched with the highest frequency contained in the sample sequence and the delayed sample sequence. It is easier to obtain an optimal evaluation result. Therefore, means for limiting the band such as a low-pass filter is used to limit the band of the sample sequence and the delayed sample sequence to only a low frequency equal to or lower than the highest frequency that can be expressed by a sequence of selection points, for example. Here, it is preferable to limit the bands of both the sample string and the delayed sample string, but it is not always necessary to limit both bands. That is, the band of only one of the sample sequence and the delayed sample sequence may be limited. Even by this method of limiting the bandwidth, a more favorable evaluation result can be obtained than the method of limiting the bandwidth of both. Even when the bandwidth of both the sample sequence and the delay sample sequence is limited, if the sample sequence and the delay sample sequence are based on the same digital time series signal, the sample sequence and the delayed sample are processed by a single low-pass filter process. It is more efficient to limit the bandwidth of the section including both the columns.

  Next, evaluation values (hereinafter referred to as “selected point evaluation values”) at the selected sample points (hereinafter referred to as “selected points”) are sequentially calculated. In this calculation, it is optimal to use a band-limited sample string and a band-limited delay sample string, and at least one of the sample string and the delay sample string is a band-limited sample string. It is preferable to use it. Note that the correlation value may be used as the evaluation value, but other values may be used. Below, it demonstrates on the assumption which used the correlation value as an evaluation value. Also in the calculation of the correlation value, it is possible to use only the amplitude value of the selected point (downsampling), and the product calculation can be reduced.

When the maximum correlation value is updated, the information on the selected point (hereinafter referred to as “optimal selected point”) with the updated maximum correlation value and the maximum correlation value are rewritten. A correlation value (hereinafter referred to as “hierarchical evaluation value”) of sample points between the optimal selection point and the adjacent selection point is calculated. The calculation of the hierarchical evaluation value may be performed as follows. First, a correlation value between a sample point obtained by adding the delay amount S / 2 to the optimum selection point and a sample point obtained by subtracting the delay amount S / 2 is calculated. As a result, the maximum correlation value is obtained among the correlation values at the optimum selection point and the correlation values at two sample points shifted by S / 2. Then, the correlation value between the sample point obtained by adding the delay amount S / 4 to the sample point where the maximum correlation value is obtained and the sample point obtained by subtracting the delay amount S / 4 is calculated. Then, a sample point for obtaining the maximum correlation value from the three correlation values is obtained. Such a process is repeated until S / 2 ⁱ becomes the interval between the sample points of the sample sequence. Note that the number of repetition processes may be determined in advance. Note that if the amount of change in delay (the interval between adjacent sample points of the delay amount for which the correlation value is to be calculated) matches the sample interval of the sample sequence, the sample sequence that does not limit the band (original sample sequence) The correlation value may be calculated using a delay sample sequence (original delay sample sequence) that is not band-limited. Furthermore, when calculating the correlation value, the bandwidth of the sample sequence and the delayed sample sequence is adjusted to the maximum frequency that can be expressed by the sequence of sample points of the delay amount for which the correlation value is to be calculated. A low-pass filter may be used in accordance with the amount of change. In this way, the sample point that gives the maximum correlation value more accurately can be obtained. However, there is a problem that the amount of calculation processing increases (the time required for processing increases).

  The selection point evaluation value calculation and the hierarchical evaluation value calculation are repeated for all selection points. Then, the largest correlation value (hereinafter referred to as “optimum evaluation value”) among all the calculated selection point evaluation values and hierarchical evaluation values and the delay amount for obtaining the correlation value are obtained. The delay amount for obtaining the optimum evaluation value is the optimum delay amount, and the correlation value (optimum evaluation value) at that time is the optimum gain.

  For example, in the case of selecting once for S samples, the speed increases as S increases, but the possibility of missing the optimum delay amount increases. When the sampling frequency is as high as 192 kHz, even if S is set to a large value such as 12 or 16, the possibility of being overlooked is low. However, when the sampling frequency is as low as 48 kHz, it is better to set S to a small value such as 2 to 8.

Principle 2
FIG. 13 shows the principle in the case of calculating the evaluation values of all the selected points and performing calculation by subdividing the neighborhood of the selected point having the maximum evaluation value hierarchically. A sample point is selected from the sample points by a predetermined method. The predetermined method is the same as in Principle 1. Then, the correlation values of all selected points are calculated. The correlation value at the sample point between the selection point having the maximum correlation value and the selection point adjacent to the selection point is calculated. Then, the largest correlation value (optimum evaluation value) and the delay amount for obtaining the correlation value are obtained from the correlation values at the sample point between the correlation value at the selection point with the largest correlation value and the adjacent selection point. . The delay amount for obtaining the optimum evaluation value is the optimum delay amount, and the correlation value (optimum evaluation value) at that time is the optimum gain.

  In the following, an embodiment of an apparatus for compressing and encoding a signal using the prediction delay search apparatus according to the present invention will be described. Note that the predictive coding apparatus of the present invention is a delay amount / gain calculation unit (120 or 140) and a low-pass filter 110 in the following embodiments.

[First Embodiment]
FIG. 14 shows a functional configuration example of the compression encoding apparatus of the first embodiment. FIG. 15 is an overview of the processing flow of the first embodiment. In this embodiment, an example of Principle 1 will be described, but Principle 2 is also possible. The compression coding apparatus of the first embodiment includes a framing unit 910, a short-term prediction analysis unit 920, a low-pass filter 110, a delay amount / gain calculation unit 120, a delay unit 940, a gain multiplication unit 950, a subtraction unit 960, a long-term A prediction error encoding unit 970 and a code string generation unit 980 are included. The delay amount / gain calculation unit 120 includes a selection point evaluation unit 121, an optimum selection point search unit 122, a hierarchical evaluation unit 123, an optimum evaluation search unit 124, and an output unit 125.

Since the processing of the framing unit 910 and the short-term prediction analysis unit 920 performs the same processing as the prior art shown in FIG. 1, here, the processing after step S922 will be described. Here, a case will be described in which the delay amount search range is calculated for each S. The low-pass filter 110 performs low-pass filtering on the short-term prediction error signal y (n) (n = 0,..., N−1) for each frame, which is the output of the short-term prediction analysis unit 920, and the band is limited. A signal y _L (n) (n = 0,..., N−1) (with the amplitude flattened) is obtained (S110). It is preferable that the bandwidth of the sample sequence corresponds to the interval for selecting the delay amount. Theoretically, according to the Nyquist theorem, the frequency band of the sample sequence is limited to less than half of the frequency corresponding to the frequency of the sample (delay amount selection interval). This is because the error due to taking can be reduced. However, since it is not necessary to cope completely, the low-pass filter can be omitted. The delay amount / gain calculation unit 120 initializes parameters used for the calculation process (S1201). The selection point evaluation means 121 calculates a correlation value at the initially set selection point (initially set sample point of delay amount) (S1211). The optimum selection point search means 122 confirms whether the maximum correlation value has been updated (S1221).

  When step S1221 is Yes, the optimum selection point search means 122 updates the maximum correlation value and the sample point (optimum selection point) for obtaining the maximum correlation value (S1222). Next, the hierarchical evaluation means 123 calculates a correlation value (hierarchical evaluation value) of sample points that are sample points between the optimal selection point and the adjacent selection points and that meet a predetermined condition ( S1231). When the correlation value of the optimum selection point is maximum, the optimum evaluation search means 124 sets the optimum selection point as the optimum delay amount and the correlation value of the optimum selection point as the optimum gain. If the maximum correlation value is updated, the delay amount at that time is the optimum delay amount, and the correlation value at that time is the optimum gain (S1241). The delay amount / gain calculation unit 120 confirms whether the hierarchical calculation process is completed (S1202). If step S1202 is No, the condition for selecting the sample point in step S1231 is changed, and the process returns to step S1231 (S1203). When step S1202 is Yes, it progresses to step S1204.

  When step S1221 is No, it progresses to step S1204. In step S1204, the delay amount / gain calculation unit 120 confirms whether the calculation of correlation values at all selection points is completed (S1204). If step S1204 is No, the delay amount / gain calculation unit 120 changes the selection point for calculating the correlation value, and returns to step S1211 (S1205). If step S1204 is Yes, the process proceeds to step S1251. The output unit 125 encodes the optimum delay amount and the optimum gain, and outputs the optimum delay amount, optimum gain, delay code, and gain code (S1251). The subsequent steps S940 to S980 are the same as in the prior art.

を計算する（Ｓ１２１１’）。ここで、ｙ_Ｌ（ｎ−τ）の一部（例えば、ｙ_Ｌ（−τ））は現フレームの処理では求められていないが、過去のフレームの処理で求めたものを用いることができる。もちろん、現フレームにおける低域通過フィルタの処理対象をｙ（ｎ）（ｎ＝−τ，…，Ｎ−１）として現フレームの処理でｙ_Ｌ（ｎ）（ｎ＝−τ，…，Ｎ−１）を求めてもよい。なお、ステップＳ１２１１’では、正規化した相関値を計算したが、正規化していない相関値

を計算してもよい。また、全てのサンプルを用いて計算するのではなく、ｍ個に１つのサンプルを用いて、

のように計算してもよい。また、上記の３つの式では、２つの信号とも帯域制限したが、どちらか一方のみを帯域制限してもよい。帯域制限しない場合は、経路１１１を経由して短期予測誤差ｙ（ｎ）が遅延量・利得算出部１２０に入力される。 FIG. 16 shows an example of a specific processing flow of steps S1201 to S1251 described above. The following processing is performed as an initial setting. The minimum delay amount candidate τ _min is set as the delay amount candidate τ. The maximum correlation value γ _{max is set} to zero. The parameter i indicating the fineness of the delay amount search is set to 1 (S1201 ′). Correlation value between the signal y _L (n) (n = 0,..., N−1) whose band is limited and the signal y _L (n−τ) (n = 0,..., N−1) delayed by τ. (Normalized correlation value)

Is calculated (S1211 ′). Here, a part of y _L (n−τ) (for example, y _L (−τ)) is not obtained by processing of the current frame, but can be obtained by processing of the past frame. Of course, the processing object of the low-pass filter in the current frame is y (n) (n = −τ,..., N−1), and y _L (n) (n = −τ,. 1) may be obtained. In step S1211 ′, the normalized correlation value is calculated, but the correlation value is not normalized.

May be calculated. Also, instead of calculating using all samples, using one sample per m,

You may calculate as follows. In the above three formulas, both the two signals are band-limited, but only one of them may be band-limited. When the band is not limited, the short-term prediction error y (n) is input to the delay amount / gain calculation unit 120 via the path 111.

Next, it is confirmed whether γ _max > γ (τ) (S1221 ′). If step S1221 ′ is Yes, the process proceeds to step S1204 ′. If step S1221 'is No, substituting gamma (tau) in γ _max (S12221), substitutes tau to _τ γmax (12222).

In step S1231 ′ of the hierarchical evaluation, first, γ (τ _γmax −S / 2 ⁱ ) is calculated (S12311). _τ γmax + S / ^{2 i} to confirm that less than the maximum delay τ _max (S12312). If step S12312 is Yes, the calculating the ^{_{γ (τ γmax + S / 2}} i) (S12313). If step S12312 is No, the ^{_{γ (τ γmax + S / 2}} i) a 0 (S12314). The maximum value among γ _max , γ (τ _γmax + S / 2 ⁱ ), and γ (τ _γmax −S / 2 ⁱ ) is γ _max . Further, the delay amount having the maximum value and _τ γmax (S1241 '). It is confirmed that i is not more than a predetermined value I (S1202 ′). If Step S1202 ′ is Yes, i is incremented by 1, and the process returns to Step S12311 (S1203 ′). When step S1202 ′ is No, it is confirmed that τ is smaller than τ _max (S1204 ′). When step S1204 'is Yes, S is added to (tau) and it returns to step S1211' (S1205 '). If step S1204 ′ is No, the process proceeds to step S1251 ′. Then, τ _γmax is output as the optimal delay amount τ _opt , γ _max is output as the optimal gain γ _opt , and the delay amount τ _opt and the optimal gain γ _opt are also output.

  By processing in this way, the optimum delay amount and gain can be obtained without calculating correlation values at all delay amount candidate points. Therefore, it is possible to avoid overlooking the optimum delay due to the selected sample point, and to greatly reduce the amount of calculation compared to the conventional case.

により求める。ただし、τ_γｍａｘ＝τ_ｍａｘの場合は、γ（τ_γｍａｘ＋Ｓ/２^Ｉ−１）＝０とする。また、γ_ｍａｘ＝γ（τ_γｍａｘ）とする。そして、ステップＳ１２４１’へ進む。このように処理することで、最適な遅延量付近での遅延量と利得の計算には、帯域が制限されていない短期予測残差信号を用いることができる。これにより、選択されていない遅延量があるにもかかわらず、最適遅延量の見逃しを防ぎながら、従来に比べて大幅な演算量の低減を図ることができる。 In the present embodiment, in the processing of steps S1231 ′ to S1241 ′ (calculation of the correlation value of the final stage) when i is I−1, the band-limited signal y _L (n) (n = 0, .., N−1) may be used instead of the short-term prediction error signal y (n) (n = 0,..., N−1). In this case, the delay amount / gain calculation unit 120 uses the short-term prediction error signal y (n) input from the line 111 shown by the dotted line in FIG. Moreover, step S1231 'when i is I-1 is changed as follows. Correlation value

Ask for. However, when τ _γmax = τ _max , γ (τ _γmax + S / 2 ^I-1 ) = 0 is set. In addition, γ _max = γ (τ _γmax ). Then, the process proceeds to step S1241 ′. By processing in this way, a short-term prediction residual signal whose bandwidth is not limited can be used for calculation of the delay amount and gain near the optimum delay amount. As a result, despite the unselected delay amount, it is possible to significantly reduce the amount of calculation compared to the conventional method while preventing the optimum delay amount from being overlooked.

[Second Embodiment]
FIG. 17 illustrates a functional configuration example of the compression encoding apparatus according to the second embodiment. FIG. 18 is an overview of the processing flow of the second embodiment. In this embodiment, an example of principle 2 will be described, but principle 1 is also possible. The compression coding apparatus of the second embodiment includes a framing unit 910, a short-term prediction analysis unit 920, a low-pass filter 110, a delay amount / gain calculation unit 140, a delay unit 940, a gain multiplication unit 950, a subtraction unit 960, a long-term A prediction error encoding unit 970 and a code string generation unit 980 are included. The delay amount / gain calculation unit 140 includes a selection point evaluation unit 141, an optimum selection point search unit 142, a hierarchical evaluation unit 143, an optimum evaluation search unit 144, an output unit 145, and a recording unit 146.

  The processes of the framing unit 910, the short-term prediction analysis unit 920, and the low-pass filter 110 are the same as those in the first embodiment. Here, the process after step S110 is demonstrated. The recording unit 146 of the delay amount / gain calculation unit 140 records the sample value (S1461). The delay amount / gain calculation unit 140 initializes parameters used for the calculation process (S1401). The selection point evaluation means 141 calculates a correlation value at the initially selected selection point (S1411). The optimum selection point searching unit 142 confirms whether the maximum correlation value has been updated (S1421).

  If step S1421 is Yes, the optimum selection point searching unit 142 updates the maximum correlation value and the sample point (optimum selection point) for obtaining the maximum correlation value (S1422). The delay amount / gain calculation unit 140 checks whether the calculation of correlation values at all selection points is completed (S1402). If step S1402 is No, the delay amount / gain calculation unit 140 changes the selection point for calculating the correlation value, and returns to step S1411 (S1403). When step S1402 is Yes, it progresses to step S1431.

  Next, the hierarchical evaluation unit 143 calculates a correlation value (hierarchical evaluation value) of sample points that are sample points between the optimal selection point and the adjacent selection points and that meet a predetermined condition ( S1431). When the correlation value of the optimum selection point is maximum, the optimum evaluation search means 144 sets the optimum selection point as the optimum delay amount and the correlation value of the optimum selection point as the optimum gain. If the maximum correlation value is updated, the delay amount at that time is set as the optimum delay amount, and the correlation value at that time is set as the optimum gain (S1441). The delay amount / gain calculation unit 120 confirms whether the hierarchical calculation process is completed (S1404). If step S1404 is No, the condition for selecting the sample point in step S1431 is changed, and the process returns to step S1431 (S1405). If step S1404 is Yes, the process proceeds to step S1451. The output unit 145 encodes the optimum delay amount and the optimum gain, and outputs the optimum delay amount, the optimum gain, the delay code, and the gain code (S1451). The subsequent steps S940 to S980 are the same as in the prior art.

を計算する（Ｓ１４１１’）。上記の式では、２つの信号とも帯域制限したが、どちらか一方のみを帯域制限してもよい。帯域制限しない場合は、経路１１１を経由して短期予測誤差ｙ（ｎ）が遅延量・利得算出部１２０に入力される。なお、第１実施形態と同じように、正規化されていない相関値や、ｍ個に１つの割合でサンプルを用いて相関値を計算してもよい。γ_ｍａｘ＞γ（τ）かを確認する（Ｓ１４２１’）。ステップＳ１４２１’がＹｅｓの場合、ステップＳ１４０２’へ進む。ステップＳ１４２１’がＮｏの場合、γ_ｍａｘにγ（τ）を代入し（Ｓ１４２２１）、τ_γｍａｘにτを代入する（１４２２２）。τがτ_ｍａｘよりも小さいことを確認する（Ｓ１４０２’）。ステップＳ１４０２’がＹｅｓの場合、τにＳを加え、ステップＳ１４１１’へ戻る（Ｓ１４０３’）。ステップＳ１４０２’がＮｏの場合、ステップＳ１４３１’へ進む。 FIG. 19 shows an example of a specific processing flow of steps S1461 to S1451 described above. The following processing is performed as an initial setting. The minimum delay amount candidate τ _min is set as the delay amount candidate τ. The maximum correlation value γ _{max is set} to zero. The parameter i indicating the detail of the delay amount search is set to 1 (S1401 ′). Correlation value between the signal y _L (n) (n = 0,..., N−1) whose band is limited and the signal y _L (n−τ) (n = 0,..., N−1) delayed by τ. (Normalized correlation value)

Is calculated (S1411 ′). In the above equation, the band of both signals is limited, but only one of them may be band limited. When the band is not limited, the short-term prediction error y (n) is input to the delay amount / gain calculation unit 120 via the path 111. Note that, as in the first embodiment, correlation values that are not normalized or correlation values may be calculated using samples at a rate of one per m. It is confirmed whether γ _max > γ (τ) (S1421 ′). When step S1421 'is Yes, it progresses to step S1402'. If step S1421 'is No, substituting gamma (tau) in γ _max (S14221), substitutes tau to _τ γmax (14222). It is confirmed that τ is smaller than τ _max (S1402 ′). If Step S1402 ′ is Yes, S is added to τ, and the process returns to Step S1411 ′ (S1403 ′). If step S1402 ′ is No, the process proceeds to step S1431 ′.

In step S1431 'hierarchical evaluation, first, it calculates the ^{_{γ (τ γmax -S / 2 i}} ) (S14311). _τ γmax + S / ^{2 i} to confirm that less than the maximum delay τ _max (S14312). If step S14312 is Yes, the calculating the ^{_{γ (τ γmax + S / 2}} i) (S14313). If step S14312 is No, the ^{_{γ (τ γmax + S / 2}} i) a 0 (S14314). The maximum value among γ _max , γ (τ _γmax + S / 2 ⁱ ), and γ (τ _γmax −S / 2 ⁱ ) is γ _max . Further, the delay amount having the maximum value and _τ γmax (S1441 '). It is confirmed that i is not more than a predetermined value I (S1404 ′). If step S1404 ′ is Yes, i is incremented by 1, and the process returns to step S14311 (S1405 ′). Step S1404 'If No, the optimum delay amount _τ γmax τ _opt, and outputs the gamma _max as the optimal gain gamma _opt, also outputs the sign of the delay tau _opt and the optimum gain γ _opt (S1451') .

  By processing in this way, the optimum delay amount and gain can be obtained without calculating correlation values at all delay amount candidate points. Therefore, it is possible to avoid overlooking the optimum delay due to the selected sample point, and to greatly reduce the amount of calculation compared to the conventional case.

Similarly to the first embodiment, in the processing of steps S1431 ′ to S1441 ′ (calculation of the correlation value at the final stage) when i is I−1, the band-limited signal y _L (n) ( Instead of n = 0,..., N−1), a short-term prediction error signal y (n) (n = 0,..., N−1) may be used. The specific processing method is the same as step S1231 ′ of the first embodiment.

または、

または

となる。ただし、ｙ_１Ｌ（ｎ）は入力信号ｘ_１（ｎ）の短期予測誤差を帯域制限した信号（低域通過フィルタ１１０_１を通過した短期予測誤差ｙ_１（ｎ））、ｙ_２Ｌは入力信号ｘ_２（ｎ）の短期予測誤差を帯域制限した信号（低域通過フィルタ１１０_２を通過した短期予測誤差ｙ_２（ｎ））である。また、上記の３つの式では、２つの信号とも帯域制限したが、どちらか一方のみを帯域制限してもよい。帯域制限しない場合は、経路１１１_１または経路１１１_２を経由して短期予測誤差ｙ_１（ｎ）またはｙ_２（ｎ）が遅延量・利得算出部１２０”に入力される。 [Third Embodiment]
FIG. 20 shows a functional configuration example of the two-channel signal encoding apparatus. Two signals of x ₁ (n) and x ₂ (n) are input to the 2-channel signal encoding apparatus. Differences from FIG. 14 are as follows. In the first embodiment (FIG. 14), the short-term prediction error signal of one input signal is encoded by obtaining a weighted difference from the temporally shifted signal of the short-term prediction error signal of the same input signal. In the configuration of FIG. 20, the input signal x ₁ (n) is independently encoded, and the short-term prediction error signal y ₂ after the short-term prediction analysis is performed for the input signal x ₂ (n), although the short-term prediction analysis is performed independently. for (n) _encodes the weighted difference between the short-term prediction error signal _y 1 (n) of _x 1 (n). Since there are two input signals, there are two framing units 910 _i , short-term prediction analysis units 920 _i , low-pass filters 110 _i , and long-term prediction error encoding units 970 _i (i = 1, 2). The difference from the first embodiment is that the input to the delay amount / gain calculation unit 120 ″ is 2 channels. FIG. 21 shows a processing flow of the 2-channel signal encoding apparatus. Even in the processing flow, steps S910 _i , There are two channels of S920 _i , S110 _i , and S970 _i . The equation for calculating the correlation in step S1211 ″ performed by the selection point evaluation unit 121 ″ and step S1231 ″ performed by the hierarchical evaluation unit 123 ″ is:

Or

Or

It becomes. However, y _1L (n) is a signal obtained by band-limiting the short-term prediction error of the input signal x ₁ (n) (short-term prediction error y ₁ (n) passing through the low-pass filter 110 ₁ ), and y _2L is the input signal x a _second band limiting the signal short-term prediction error (n) (low-pass short-term prediction error _y 2 which has passed through the filter 110 ₂ (n)). In the above three formulas, both the two signals are band-limited, but only one of them may be band-limited. When the band is not limited, the short-term prediction error y ₁ (n) or y ₂ (n) is input to the delay amount / gain calculation unit 120 ″ via the path 111 ₁ or the path 111 ₂ .

となる。ただし、ｙ_１Ｌ（ｎ）は入力信号ｘ_１（ｎ）の短期予測誤差を帯域制限した信号（低域通過フィルタ１１０_１を通過した短期予測誤差ｙ_１（ｎ））、ｙ_２Ｌは入力信号ｘ_２（ｎ）の短期予測誤差を帯域制限した信号（低域通過フィルタ１１０_２を通過した短期予測誤差ｙ_２（ｎ））である。また、上記の式では、２つの信号とも帯域制限したが、どちらか一方のみを帯域制限してもよい。帯域制限しない場合は、経路１１１_１または経路１１１_２を経由して短期予測誤差ｙ_１（ｎ）またはｙ_２（ｎ）が遅延量・利得算出部１４０”に入力される。 [Modification]
The 2-channel signal encoding apparatus of FIG. 20 can also be realized by using the delay amount / gain calculation unit 140 ″ corresponding to 2 channels for the input of the delay amount / gain calculation unit 140 of the second embodiment. Is a structure in which numbers are given in parentheses in the figure, and the processing flow in this case is shown in Fig. 22. The difference between Fig. 18 and Fig. 22 is the same as the difference between Fig. 15 and Fig. 21. Also, the equation for calculating the correlation between step S1411 ″ performed by the selection point evaluation unit 141 ″ and step S1431 ″ performed by the hierarchical evaluation unit 143 ″ is:

It becomes. However, y _1L (n) is a signal obtained by band-limiting the short-term prediction error of the input signal x ₁ (n) (short-term prediction error y ₁ (n) passing through the low-pass filter 110 ₁ ), and y _2L is the input signal x a _second band limiting the signal short-term prediction error (n) (low-pass short-term prediction error _y 2 which has passed through the filter 110 ₂ (n)). In the above formula, the band of both the two signals is limited, but only one of them may be band limited. When the band is not limited, the short-term prediction error y ₁ (n) or y ₂ (n) is input to the delay amount / gain calculation unit 140 ″ via the path 111 ₁ or the path 111 ₂ .

[Fourth Embodiment]
FIG. 23 shows a functional configuration example of the multi-channel signal encoding apparatus by a solid line. The multi-channel signal encoding apparatus 2100 according to the present embodiment multiplies the master channel at one time position difference (delay amount) τ by a weight, obtains a difference from the signal of the encoding target channel, and encodes it. The difference between the multi-channel signal encoding device 2100 and the multi-channel signal encoding device 1800 shown in FIG. The encoding information determination unit 2120 includes an independent / difference / master channel determination unit 1821, a low-pass filter 110, and a delay / gain calculation unit 120 ″ (see FIG. 20). Note that the delay / gain calculation unit. Instead of 120 ″, a delay / gain calculation unit 140 ″ (see FIG. 20) may be used. In the following description, either the delay / gain calculation unit 120 ″ or the delay / gain calculation unit 140 ″ may be used. In this case, the delay amount / gain calculation unit 120 ″ (140 ″) is shown. The internal structure of the delay amount / gain calculation unit 120 ″ (140 ″) is the delay amount / gain calculation unit 120 ″ (140 ″) of FIG. ).

FIG. 24 shows a processing flow of the multi-channel signal encoding apparatus 2100 with a solid line. The frame buffer 1810 _i (i = 1 to M) stores an input signal (channel signal). Here, when the channel signal is simply a sequence of sample values, it is divided into a plurality of frames, and when the channel signal is already divided into frames, it is stored in units of frames (S1810). The encoding information determination unit 2120 uses the independent / difference / master channel determination unit 1821, the low-pass filter 110, and the delay amount / gain calculation unit 120 ″ (140 ″) to encode the encoding information (independent code) of each channel. Or the master channel number, delay amount, weight, etc.) are determined (S2120). In this embodiment, only one delay amount and one weight are output. The encoding target signal generation unit 1830 generates a signal to be encoded by subtracting a value obtained by multiplying the master channel signal delayed by the determined delay amount from the input signal for each channel by the calculated weight ( S1830). The signal encoding unit 1840 _i (i = 1 to M) encodes the generated encoding target signal (S1840). The combining unit 1850 combines the code of each channel signal and the encoded information, and outputs a multi-channel code (S1850).

Detailed processing in step S2120 is as follows. The independent / difference / master channel determination unit 821 determines whether to perform independent encoding or differential encoding for each channel signal, and in the case of differential encoding, which channel signal to use for the master channel (S8210). The independent / difference / master channel determination unit 821 confirms whether the encoding of the channel signal is independent encoding (S21220). In the case of independent coding, the process of S2120 for the channel signal is terminated, and step S2120 for the next channel signal is performed. In the case of differential encoding, the process proceeds to step S110. Low pass filter 110, each frame from the frame buffer 1810 m of coded channels _(m is the channel number of the coded channel.) And the frame buffer 1810 m of the master channel _(m is the channel number of the master channel.) Input signal y _m (n) (m is the channel number of the channel to be encoded and the master channel. N = 0,..., N−1) is low-pass filtered to limit the band (flatten the amplitude) Signal y _mL (n) (m is the channel number of the channel to be encoded and the master channel, n = 0,..., N−1) is obtained (S110). In the case of the delay amount / gain calculation unit 120 ″, the optimum delay amount and the optimum gain are obtained by the same method as step S120 ″ in FIG. 21 (third embodiment). In the case of the delay amount / gain calculation unit 140 ″, the optimum delay amount and the optimum gain are obtained by the same method as step S140 ″ in FIG. 22 (modified example of the third embodiment). Then, the delay amount / gain calculation section 120 ″ (140 ″) uses the obtained optimum delay amount as the delay amount (time difference) of the master channel signal with respect to the signal of the encoding target channel, and the optimum gain is the signal of the master channel. Is output as a weight multiplied by (S120 ″ or S140 ″).
By processing in this way, it is possible to obtain the optimum delay amount and gain without calculating correlation values at all delay amount candidate points of the master channel with respect to the signal of the encoding target channel. Therefore, it is possible to avoid overlooking the optimum delay due to the presence of a delay amount for which a correlation value has not been obtained, and to greatly reduce the amount of calculation compared to the conventional case.

[Fifth Embodiment]
An example of a functional configuration of the multi-channel signal encoding device is shown by a solid line and a dotted line in FIG. The multi-channel signal encoding device 2100 according to the present embodiment multiplies the master channel sample sequence at a plurality of time position differences (delay amounts) τ by a weight to obtain a difference from the signal of the encoding target channel. Turn into. The encoding information determination unit 2120 of the multi-channel signal encoding device 2100 according to this embodiment includes a τ determination unit 2125 and a weight determination unit 2126. The solid line and the dotted line in FIG. 24 indicate the processing flow of the multi-channel signal encoding apparatus 2100. The τ determination unit 2125 determines the time position difference (delay amount) τ of the master channel sample sequence used for the weighted differential encoding from the optimum delay amount τ _opt obtained in step S120 ″ or step S140 ″ (S2125). ). The weight determination unit 2126 calculates a weight for each time position difference in the sample sequence (S2126).

Details of step S2125 are shown in FIG. The τ determination unit 2125 confirms whether or not the optimum delay amount τ _opt obtained in step S120 ″ or step S140 ″ is 0 (S21251). When the optimum delay amount τ _opt is 0, the time difference (time position difference) τ between the frame signal (sample sequence) of the channel to be encoded and the sample sequence of the master channel to be referred to is set to only 0 (S21252). When the optimal delay amount τ _opt is not 0, the time difference (time position difference) τ between the frame signal (sample sequence) of the channel to be encoded and the sample sequence of the master channel to be referenced is 0 and the optimal delay amount τ _opt Two are assumed (S21253).

により算出する（Ｓ２１２６３）。 Details of step S2126 are shown in FIG. The weight determination unit 2126 confirms the number of τ (S21261). When the number of τ is 1, the weight coefficient γ ₀ is
γ ₀ = (Y ₀ ^T Y ₀ ) ⁻¹ X ^T Y ₀
(S21262). However, X ^T Y ₀ is an inner product and is Σx (i) y (i). When the number of τ is two, the weighting coefficients γ ₀ and γ _opt are

(S21263).

Steps S2125 and S2126 in the case of using a sample string of three or six master channels are as follows. FIG. 27 shows a processing flow of step S2125 ′. The τ determination unit 2125 first confirms the value of the obtained optimum delay amount τ _opt (S21251 ′). When the optimal delay amount τ _opt is 0, the time difference (time position difference) τ between the frame signal of the channel to be encoded (sample sequence matching the frame) and the sample sequence of the master channel to be referenced is set to −1, 0. 1 (S21254). When the optimum delay amount τ _opt is 1 or 2, the time difference (time position difference) τ between the frame signal (sample sequence) of the channel to be encoded and the sample sequence of the master channel to be referenced is set to −1, 0, 1 2, 3, 4 (S21255). When the optimum delay amount τ _opt is −1 or −2, the time difference (time position difference) τ between the sample sequence of the channel to be encoded and the sample sequence of the master channel to be referenced is set to −4, −3, −2. , -1, 0, 1 (S21256). When the optimum delay amount τ _opt is not −2, −1, 0, 1, or 2, the time difference (time position difference) τ between the sample sequence of the channel to be encoded and the sample sequence of the master channel to be referenced is −1. , 0, 1, τ _opt -1, τ _opt , τ _opt +1 (S21257).

により算出する（Ｓ２１２６４）。τの数が６個の場合には、重み係数γ_−１、γ_０、γ_１、γ_{ｏｐｔ−１}、γ_ｏｐｔ、γ_{ｏｐｔ＋１}を、

ただし、

により算出する（Ｓ２１２６５）。 FIG. 28 shows a processing flow of step S2126 ′. The weight determination unit 2126 first checks the number of τ (S21261 ′). When the number of τ is 3, the weight coefficients γ ₋₁ , γ ₀ , γ ₁ are

(S21264). When the number of τ is 6, weighting factors γ ₋₁ , γ ₀ , γ ₁ , γ _opt−1 , γ _opt , γ _{opt + 1} are

However,

(S21265).

  By processing in this way, it is possible to obtain the optimum delay amount and gain without calculating correlation values at all delay amount candidate points of the master channel with respect to the signal of the encoding target channel. Therefore, it is possible to avoid overlooking the optimum delay due to the presence of a delay amount for which a correlation value has not been obtained, and to greatly reduce the amount of calculation compared to the conventional case.

  The above embodiment can be implemented by causing the recording unit 3020 of the computer shown in FIG. 29 to read a program for executing each step of the above method and causing the control unit 3010, the input unit 3030, the output unit 3040, and the like to operate. . 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 figure which shows the structure of the conventional compression encoding apparatus. The figure which shows the processing flow of the conventional compression encoding apparatus. The figure which shows the function structural example of the conventional decoding apparatus. The figure which shows the function structural example of the conventional multi-channel signal encoding apparatus. The figure which shows the processing flow of the conventional multi-channel signal encoding apparatus. The figure which shows the image of the process of step S1830 in case a sample row | line | column is one (1 tap). The figure which shows the image of the process of step S1830 in case a sample row | line is three (3 taps). The figure which shows the function structural example of the conventional multi-channel signal decoding apparatus. The figure which shows the image in the case of using two sample rows (2 taps). The figure which shows the image in the case of using 6 sample rows (6 taps). The figure which shows the specific example in which the effect of this invention appears. The figure which shows the principle in the case of calculating hierarchically subdivided whenever the maximum value of an evaluation value is updated. The figure which shows the principle in the case of calculating the evaluation value of all the selection points, and subdividing hierarchically the vicinity of the selection point with the largest evaluation value. The figure which shows the function structural example of the compression encoding apparatus of 1st Embodiment. The figure which shows the outline | summary of the processing flow of 1st Embodiment. The figure which shows the example of the specific process flow of step S1201-step S1251. The figure which shows the function structural example of the compression encoding apparatus of 2nd Embodiment. The figure which shows the outline | summary of the processing flow of 2nd Embodiment. The figure which shows the example of the specific process flow of step S1461-step S1451. The figure which shows the function structural example of the 2 channel signal encoding apparatus of 3rd Embodiment. The figure which shows the outline | summary of the processing flow of 3rd Embodiment. The figure which shows the outline | summary of the processing flow of the modification of 3rd Embodiment. The figure which shows the function structural example of the multi-channel signal encoding apparatus of 4th Embodiment and 5th Embodiment. The figure which shows the processing flow of the multi-channel signal encoding apparatus of 4th Embodiment and 5th Embodiment. The figure which shows the processing flow of step S2125. The figure which shows the processing flow of step S2126. The figure which shows the processing flow of step S2125 '. The figure which shows the processing flow of step S2126 '. The figure which shows the function structural example of a computer.

Claims (10)

A digital time series signal in a certain interval (hereinafter referred to as “sample sequence”) and a sample sequence (hereinafter referred to as “delayed sample sequence”) obtained by delaying the sample sequence or another digital time series signal in the same interval. Prediction for estimating a delay amount (hereinafter referred to as an “optimal delay amount”) that is most similar between the sample sequence and the delay sample sequence from a predetermined search range of a time difference (hereinafter referred to as “delay amount”). A lazy search method comprising:
In the low-pass filter, a sample sequence (hereinafter referred to as “low-frequency sample sequence”) in which the band of the sample sequence is only a low frequency and a sample sequence (hereinafter, “ A bandwidth limiting step for generating at least one of “low-pass delay sample sequence”,
The sample string and the low-frequency delay samples at a plurality of delay amount sample points (hereinafter referred to as “selection points”) selected by the selection point evaluation unit by a predetermined method from the delay amount search range. An evaluation value (hereinafter referred to as “selection point evaluation value”) of any one of the sequence, the low-frequency sample sequence, the delay sample sequence, the low-frequency sample sequence, and the low-frequency delay sample sequence . A selected point evaluation calculation step for calculating
Optimal selection point search means for obtaining one or a plurality of selection points (hereinafter referred to as “optimal selection points”) that give a selection point evaluation value having a high similarity among the plurality of selection points by the optimal selection point search means. When,
In the hierarchical evaluation calculation means, except for the final iteration, the optimum selection point and a sample point of delay amount other than the selection point between the optimum selection point and the selection point adjacent to the optimum selection point (hereinafter referred to as “hierarchical selection”). Any one of the sample sequence, the low frequency delay sample sequence, the low frequency sample sequence, the delay sample sequence, the low frequency sample sequence, and the low frequency delay sample sequence. The similarity evaluation value is calculated to be a hierarchical evaluation value, and in the final iteration, the similarity evaluation value between the sample sequence and the delayed sample sequence at the hierarchical selection point is calculated and hierarchical evaluation is performed. a hierarchical evaluation calculation step of the value,
An optimum delay amount search means for obtaining a delay amount that gives an evaluation value having a high similarity among the optimum selection point and the hierarchical selection point as a new optimum selection point;
Output means, possess an output step for outputting the delay amount of the optimal selection point which has been determined by the optimum delay amount search step as the optimum delay amount,
A prediction delay search characterized by repeating the processing of the hierarchical evaluation calculation step and the optimum delay amount search step using a new optimum selection point obtained in the optimum delay amount search step until a preset condition is satisfied. Method. The prediction delay search method according to claim 1 ,
Each time the selected point evaluation calculation step is performed on one selected point, the optimum selected point search step is executed,
A predictive delay search method, wherein the hierarchical evaluation calculation step and the optimal evaluation value search step are executed when an optimal selection point is changed. The prediction delay search method according to claim 1 ,
After executing the selection point evaluation calculation step for all the selection points, executing the optimum selection point search step,
A prediction delay search method comprising: executing the hierarchical evaluation calculation step and the optimum evaluation value search step. A prediction delay search method according to any one of claims 1 to 3 ,
A prediction delay search method, wherein a correlation value or a normalized correlation value is used as the similarity evaluation value in the selection point evaluation calculation step and the hierarchical evaluation calculation step. A digital time series signal in a certain interval (hereinafter referred to as “sample sequence”) and a sample sequence (hereinafter referred to as “delayed sample sequence”) obtained by delaying the sample sequence or another digital time series signal in the same interval. Prediction for estimating a delay amount (hereinafter referred to as an “optimal delay amount”) that is most similar between the sample sequence and the delay sample sequence from a predetermined search range of a time difference (hereinafter referred to as “delay amount”). A delay search device comprising:
A sample sequence in which the band of the sample sequence is only a low frequency (hereinafter referred to as “low frequency sample sequence”) and a sample sequence in which the band of the delay sample sequence is only a low frequency (hereinafter referred to as “low frequency delay sample sequence”). A low-pass filter that generates at least one of
The sample sequence, the low-frequency delay sample sequence, and the low-frequency region at a plurality of delay amount sample points (hereinafter referred to as “selected points”) selected by a predetermined method from the delay amount search range. Selection point evaluation for calculating an evaluation value (hereinafter referred to as “selection point evaluation value”) of any one of the sample sequence, the delay sample sequence, the low frequency sample sequence, and the low frequency delay sample sequence Calculation means;
An optimum selection point search means for obtaining one or more selection points (hereinafter referred to as “optimal selection points”) that give a selection point evaluation value having a high similarity among the plurality of selection points;
At times other than the last iteration, the sample points (hereinafter referred to as “hierarchical selection points”) with a delay amount other than the selection point between the optimum selection point and the selection point adjacent to the optimum selection point. Calculating an evaluation value of similarity between the sample sequence, the low-frequency delay sample sequence, the low-frequency sample sequence, the delay sample sequence, the low-frequency sample sequence, and the low-frequency delay sample sequence and a hierarchical evaluation value, the repetition of the last round, hierarchical evaluation calculation to hierarchical evaluation value by calculating an evaluation value of the similarity in the hierarchical selection point, and the sample series and the delayed sample sequence Means,
An optimum delay amount search means for obtaining a delay amount that gives an evaluation value having a high similarity among the optimum selection point and the hierarchical selection point as a new optimum selection point;
Output means for outputting the delay amount of the optimum selection point obtained by the optimum delay amount search means as the optimum delay amount ;
A prediction delay search characterized by repeating the processing of the hierarchical evaluation calculation means and the optimum delay amount search means using a new optimum selection point obtained by the optimum delay amount search means until a preset condition is satisfied. apparatus. The prediction delay search device according to claim 5 ,
Each time the process of the selection point evaluation calculation unit is executed for one selection point, the process of the optimum selection point search unit is executed,
A prediction delay search device, wherein when the optimum selection point is changed, the processing of the hierarchical evaluation calculation means and the processing of the optimum evaluation value search means are executed. The prediction delay search device according to claim 5 ,
After executing the process of the selection point evaluation calculation means for all the selection points, execute the process of the optimum selection point search means,
A prediction delay search device, wherein the processing of the hierarchical evaluation calculation means and the processing of optimal evaluation value search are executed. A prediction delay search apparatus according to any one of claims 5 to 7 ,
Before SL as the evaluation value of the similarity of the selected point evaluation and calculation step the hierarchical evaluation calculation step, prediction delay search apparatus characterized by using the correlation values or normalized correlation value. Expected delay search program executed by a computer the steps of the method according to any one of claims 1 to 4. A computer-readable recording medium on which the prediction delay search program according to claim 9 is recorded.

Images