RU2458413C2 - Audio encoding apparatus and audio encoding method - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: speech encoding apparatus includes a computing section which calculates correlation values in designated pulse positions using a target signal and a plurality of pulses which form a fixed codebook, and calculates on the basis, based on the pulse, representative values of pulses using maximum correlation values; a sorting section which sorts representative values obtained on the basis based on the pulse, groups pulses corresponding to the sorted representatives values into a plurality of predetermined subsets and determines a first subset to be searched first among the plurality of subsets; and a search section which performs search of the fixed codebook using the first subset and obtains code-indicative positions and polarity of the plurality of pulses to minimise coding distortion.

EFFECT: improved coding efficiency when performing separation search over an algebraic codebook in audio encoding.

9 cl, 9 dwg

Description

FIELD OF THE INVENTION

The present invention relates to a speech encoding apparatus and a speech encoding method. In particular, the present invention relates to a speech encoding apparatus and a speech encoding method for performing a fixed codebook search.

State of the art

In mobile communications, compression coding for digital information relating to speech and images is essential for efficient use of the transmission frequency band. In particular, the expectations are high for the speech codec (encoding and decoding) techniques that are widely used for mobile phones, and demands are made to further improve the sound quality for standard coding with high compression efficiency.

Recently, standardization of a scalable codec having a multilayer configuration is on the way to implementation by efforts, for example, ITU-T (telecommunication standardization sector of the international telecommunication association) and MPEG (moving picture expert group), and a speech codec with higher efficiency and higher quality.

The effectiveness of the speech coding technique, which has been significantly improved by the basic CELP (linear prediction with coding) scheme, modeling of the vocal speech system and the adoption of vector quantization in a qualified way, is further improved by means of fixed-excitation techniques using a small number of pulses, such as algebraic code book disclosed in non-patent document 1. ITU-T Recommendation G.729 and ETSI (European Telecommunications Standards Institute) tions) standard for the AMR (adaptive multiple speed) offers a representative CELP codec using an algebraic codebook, and is widely used throughout the world.

In the case of speech encoding using an algebraic codebook, taking into account the mutual influence between the pulses forming the algebraic codebook, it is desirable to search for all combinations of pulses (hereinafter “full search”). However, when the number of pulses increases, the amount of computation required for the search increases exponentially. By comparison, Non-Patent Document 2 discloses, for example, part search, abbreviated search and Viterbi search as algebraic code book search methods to significantly reduce the amount of computation and, in essence, maintain efficiency in the case of a full search at the same time.

Among them, especially, the search for parts is the simplest way, providing the effect of a significant reduction in the magnitude of the calculations. Here, searching for parts is a way of dividing a closed cycle into many smaller closed cycles and performing an open cycle search in many closed cycles. In this part search, it is possible to significantly reduce the amount of computation according to the number of partitions. Also, the search for parts is used in international standard schemes, and in the search with the algebraic codebook ETSI of the AMR standard, which is the standard codec of third-generation mobile phones, the search for parts is performed after dividing four pulses into two subsets.

For example, if there are four pulses having eight presumptive positions, there are 8 ⁴ (i.e. 4096) pulse combinations that must be evaluated in order to search for four pulses in one closed loop. Compared to this, the ETSI AMR standard splits four pulses into two subsets of two pulses and searches individually for their closed cycles. Therefore, the number of pulse combinations to be evaluated in the ETSI AMR standard is 2 · 8 ² (i.e. 128), which is one thirty-second calculation value in the case of a full search. Additionally, evaluation in the ETSI AMR standard is performed for two pulses, which is less than four pulses, so that the magnitude of the calculations is further reduced.

Non-Patent Document 1: Salami, Laflamme, Adoul, "8 kbit / s ACELP Coding of Speech with 10 ms Speech-Frame: a Candidate for CCITT Standardization", IEEE Proc. ICASSP94, p. II-97n.

Non-Patent Document 2: T. Nomura, K. Ozawa, M. Serizawa, "Efficient pulse excitation search methods in CELP", Proc. of the 1996 spring meeting of the Acoustic Society of Japan. 2-P-5, p. 311-312, March. 1996.

Disclosure of invention

Problems to be Solved by this Invention

However, in general, speech coding efficiency by searching for parts with an algebraic code book is lower than in the case of a full search, since the positions of the two pulses, which were determined first, are not always optimal.

Therefore, in the search for parts, the efficiency of speech coding can be further improved depending on which pulses to choose in order to form a subset to be searched first. For example, it is possible to adopt a method of selecting two pulses from four pulses at random and performing a search, and after this process is repeated several times, finding a pair of pulses, whereby the coding efficiency is the highest. For example, by providing four types of subset pairs and searching for these four pairs individually, it is possible to make speech coding efficiency close to coding efficiency in full search. In this case, 128 (8 ² × 2) × 4 (i.e. 512) calculation templates are required, which is one-eighth of the calculation value in the case of a full search. Here, in the examples described above, the subsets are formed in an arbitrary manner, and there is no particular reason for any pairs to be the first to search among the four types of pairs. Therefore, if the search is performed individually in many cases, the resulting coding efficiency shows large variations, and the full coding efficiency is insufficient.

Therefore, it is an object of the present invention to provide a speech encoding device and a speech encoding method for performing a search for parts with an algebraic codebook and improving encoding efficiency.

Means for solving the problem

The speech encoding apparatus of the present invention applies a configuration having: a calculation section that calculates correlation values at the estimated pulse positions using the target signal and a plurality of pulses forming a fixed codebook, and calculates, based on the pulse basis, representative pulse values using the maximum values correlations; a sorting section that sorts representative values obtained on the basis of the momentum, groups the impulses corresponding to the sorted representative values into a plurality of predetermined subsets and determines a first subset to be searched first among a plurality of subsets; and a search section that searches for a fixed codebook using the first subset and receives a code indicating the position and polarity of the plurality of pulses to minimize coding distortion.

The speech encoding method of the present invention includes the steps of: calculating correlation values at the estimated positions of the pulses using the target signal and a plurality of pulses forming a fixed codebook, and calculating, based on the pulse, representative pulse values using the maximum correlation values; sorting the representative values obtained on the basis of the momentum, grouping the pulses corresponding to the sorted representative values into a plurality of predetermined subsets and determining the first subset to be searched first among the plural subsets; and searching for a fixed codebook using a first subset and generating code-indicating positions and polarities of a plurality of pulses to minimize coding distortion.

The preferred effect of this invention

According to the present invention, when performing a search for parts with a fixed codebook in speech encoding, the subset to be searched first is determined using representative values related to pulses, such as maximum correlation values, so that it is possible to search for parts with an algebraic codebook and improve coding efficiency.

Brief Description of the Drawings

1 is a block diagram showing a configuration of a CELP encoding apparatus according to Embodiment 1 of the present invention;

FIG. 2 is a block diagram showing a configuration within a distortion minimizing section according to Embodiment 1 of the present invention; FIG.

FIG. 3 is a flowchart showing the steps of calculating the maximum correlation value of each pulse in the calculation section of the maximum correlation value according to Embodiment 1 of the present invention;

4 is a flowchart showing the steps of sorting processing on the maximum correlation value of each pulse in the sorting section according to Embodiment 1 of the present invention;

5 is a flowchart showing the steps of searching for parts with a fixed code book in a search section according to Embodiment 1 of the present invention;

6 is another flowchart showing the steps of searching for parts with a fixed codebook in a search section according to Embodiment 1 of the present invention;

7 is a flowchart showing the steps of sorting processing on the maximum correlation value of each pulse in the sorting section according to Embodiment 2 of the present invention;

FIG. 8 is a flowchart showing steps of sorting processing on the maximum correlation value of each pulse in the sorting section according to Embodiment 3 of the present invention; and

9 is a flowchart showing the steps of processing permutation over the order of pulses in the sorting section according to Embodiment 3 of the present invention.

Preferred Embodiment

Embodiments of the present invention will be explained in detail below with reference to the accompanying drawings.

Option 1 implementation

1 is a block diagram showing a configuration of a CELP encoding apparatus 100 according to Embodiment 1 of the present invention. Here, an illustrative case using the CELP encoding device as a speech encoding device according to the present invention will be explained.

1, for a speech signal S11 consisting of vocal tract information and excitation information, the CELP coding apparatus 100 encodes the vocal path information by calculating the LPC parameters (linear prediction coefficient) and encodes the excitation information by determining an index defining which speech model, saved in advance, use. That is, the drive information is encoded by determining an index defining which drive vector (code vector) to generate in the adaptive codebook 103 and the fixed codebook 104.

To be more specific, sections of the CELP encoding device 100 perform the following operations.

The LPC analysis section 101 performs linear prediction analysis of the speech signal S11, calculates the LPC parameters, which are spectral envelope information, and outputs the LPC parameters to the LPC quantization section 102 and the perceived weighting section 111.

The LPC quantization section 102 quantizes the LPC parameters derived from the LPC analysis section 101, and outputs the resulting quantized LPC parameters to the LPC synthesis filter 109 and the index of the quantized LPC parameters outside the CELP encoding device 100.

On the other hand, the adaptive codebook 103 stores past excitations used in the LPC synthesis filter 109 and generates an excitation vector of one subframe from the stored excitations according to the adaptive codebook delay associated with the index assigned from the distortion minimizing section 112 described later. This excitation vector is output to a factor 106 as an adaptive codebook vector.

The fixed codebook 104 stores in advance a plurality of excitation vectors of a predetermined shape and outputs the excitation vector associated with the index assigned from the distortion minimizing section 112 to a factor 107 as a fixed codebook vector. Here, the fixed codebook 104 is an algebraic excitation, and a case where an algebraic codebook is used is explained. Also, algebraic excitation is an excitation accepted in many standard codecs.

Additionally, the above adaptive codebook 103 is used to represent more periodic components as voice speech, while the fixed codebook 104 is used to represent less periodic components as white noise.

According to the purpose of the distortion minimization section 112, the gain codebook 105 generates an gain for the adaptive codebook vector that is derived from the adaptive codebook 103 (i.e., an adaptive codebook gain) and a gain for the fixed codebook vector that is derived from the fixed codebook books 104 (i.e., a fixed codebook gain), and outputs these amplifiers at factors 106 and 107, respectively.

A multiplier 106 multiplies the adaptive codebook vector derived from the adaptive codebook 103 by amplifying the adaptive codebook derived from the amplification codebook 105 and outputs the result to the addition unit 108.

A multiplier 107 multiplies the fixed codebook vector derived from the fixed codebook 104 by amplifying the fixed codebook derived from the amplification codebook 105 and outputs the result to the addition unit 108.

The adder 108 adds the adaptive codebook vector derived from the factor 106 and the fixed codebook vector derived from the factor 107 and outputs the resulting excitation vector to the LPC synthesis filter 109 as excitation.

The LPC synthesis filter 109 generates a synthesis signal using a filter function, including quantized LPC parameters derived from the LPC quantization section 102, as a filter coefficient and excitation vectors generated in the adaptive codebook 103 and the fixed codebook 104 as excitations, i.e. using an LPC synthesis filter. This synthesis signal is output to addition module 110.

The adder 110 calculates the error signal by subtracting the synthesis signal generated in the LPC synthesis filter 109 from the speech signal S11 and outputs this error signal to the sensed weighting section 111. Here, this error signal is equivalent to a coding distortion.

The perceptual weighting section 111 performs perceptual weighting to distort the coding output from the addition unit 110, and outputs the result to the distortion minimizing section 112.

The distortion minimizing section 112 finds the indices of the adaptive codebook 103, the fixed codebook 104 and the amplification codebook 105 on the basis of the subframe, so as to minimize the encoding distortion output from the sensed weighting section 111, and outputs these indices to the outside of the CELP encoding device 100 as encoded information. To be more specific, distortion minimizing section 112 generates a synthesis signal based on the above adaptive codebook 103 and fixed codebook 104. Here, the processing sequences to find the encoding distortion of this signal form a closed-loop control (feedback control). Further, the distortion minimizing section 112 searches for codebooks by variously changing the indices that denote codebooks in one subframe, and outputs the resulting codebook indices minimizing coding distortion.

Also, the excitation when coding distortion is minimized is fed back to the adaptive codebook 103 on a basis based on a subframe. Adaptive codebook 103 updates stored excitations through this feedback.

The method for searching for the fixed codebook 104 will be explained below. First, the search of the excitation vector and the code calculation are performed by searching the excitation vector in order to minimize the encoding distortion in the following equation 1.

[one]

E = | x- (pHa + qHs) | ² ... (Equation 1)

Where:

E: coding distortion;

x: encoding target

p: adaptive codebook vector gain;

H: perceptual weight synthesis filter;

a: adaptive codebook vector

q: fixed codebook vector gain; and

s: fixed codebook vector.

In general, the adaptive codebook vector and the fixed codebook vector are searched in open loops (i.e., in separate loops), and therefore, the adaptive codebook code 104 is output by searching for the fixed codebook vector to minimize coding distortion shown in the following equation 2.

[2]

y = x-pHa E = | y-qHs | ² ... (Equation 2)

Where:

E: coding distortion

x: encoding target (perceptually weighted speech signal);

p: optimal adaptive codebook vector gain

H: perceptual weight synthesis filter;

a: adaptive codebook vector

q: fixed codebook vector gain;

s: fixed codebook vector and

y: target vector in fixed codebook search.

Here, the gains p and q are determined after the drive code is detected so that the search is performed using the optimal gains. Then, the above equation 2 can be expressed by the following equation 3.

[3]

... (Equation 3)

Additionally, minimizing this distortion equation is equivalent to maximizing the function C of the following equation 4.

[four]

...(Уравнение 4)

... (Equation 4)

Therefore, in the case of searching for an excitation consisting of a small number of pulses, such as, for example, algebraic codebook excitation, by calculating yH and HH in advance, it is possible to calculate the above function C with a small amount of computation. Here, the elements of the vector yH correspond to momentum-specific correlation values. That is, the element of the vector yH obtained by performing the time-inverse synthesis of the target y is equivalent to the correlation value between the pulse synthesis signal that rises in this position and the target signal.

FIG. 2 is a block diagram showing a configuration within the distortion minimizing section 112 according to the present embodiment. Here, an illustrative case will be explained where, in a search for a fixed codebook in the distortion minimization section 112, the four pulses forming the algebraic codebook are divided into two subsets of two pulses and searched. It is also assumed that each impulse has eight putative positions.

2, a distortion minimization section 112 is provided by an adaptive codebook search section 201, a fixed codebook search section 202, and a gain codebook search section 203. Also, the fixed codebook search section 202 is provided with a maximum correlation value calculating section 221, a sorting section 222, a preprocessing section 223, and a search section 224.

The adaptive codebook search section 201 searches for the adaptive codebook 103 using coding distortion depending on perceived weighting in the perceived weighting section 111. The adaptive codebook search section 201 outputs the adaptive codebook vector code obtained in the search step to the adaptive codebook 103, outputs the adaptive codebook vector code obtained as a search result to the maximum correlation value calculating section 221 in the fixed codebook search section 202 and outside the CELP encoding device 100.

The fixed codebook search section 202 searches for adaptive codebook parts using coding distortion depending on perceived weighting in the perceived weighting section 111 and the adaptive codebook vector code received as input from the adaptive codebook search section 201. Additionally, the fixed codebook search section 202 outputs the fixed codebook vector code obtained in the search step to the fixed codebook 104 and outputs the fixed codebook vector code obtained as a search result outside the CELP encoding device 100 and to the search section 203 codebook gain.

The gain codebook search section 203 searches for the gain codebook based on the fixed codebook vector code, received as input from the search section 224 in the fixed codebook search section 202, coding distortion depending on perceived weighting in the perceived weighting section 111 and adaptive vector code a codebook received as input from an adaptive codebook search section 201. Additionally, the gain codebook search section 203 outputs the adaptive codebook gain and the fixed codebook gain obtained in the search step to the gain codebook 105 and outputs the adaptive codebook gain and the fixed codebook gain obtained as the search results to the outside of the device 100 CELP encoding.

The maximum correlation value calculation section 221 calculates an adaptive codebook vector using the adaptive codebook vector code received as input from the adaptive codebook search section 201 and calculates a target vector y shown in equation 2. Additionally, using the synthesis filter coefficient H in perceived weighing section 111, maximum correlation value calculating section 221 calculates and outputs a pulse-specific correlation value yH at each assumed position in the section Yu 223 pre-processing. Further, using the pulse-specific correlation value yH at each assumed position, the maximum correlation value calculating section 221 calculates and outputs the maximum correlation values of the individual pulses to the sorting section 222. Here, the calculation of the maximum correlation values in section 221 of calculating the maximum correlation value will be described later in detail.

The sorting section 222 sorts the maximum correlation values of the individual pulses received as input from the maximum correlation value calculating section 221, in order of the largest maximum correlation value (hereinafter “sorting processing”). Further, based on the sorting result, the sorting section 222 divides the four pulses into two subsets of the two pulses and outputs the separation results to the search section 224. The sorting processing in sorting section 222 will be described in detail later.

The pre-processing section 223 calculates the HH matrix using the synthesis synthesis filter coefficient H in the perceived weighting section 111. Additionally, from the polarities (+ and -) of the elements of the vector yH received as input from the maximum correlation value calculation section 221, the preprocessing section 223 determines and outputs the pulse polarity, pol, to the search section 224. To be more specific, in the pre-processing section 223, the polarities of the individual pulses that occur in the respective positions are coordinated with the polarities of the yH values at these positions and the polarities of the yH values are stored in a different sequence. After the polarities in these positions are stored in a different sequence, the preprocessing section 223 makes all of the yH values absolute, that is, the preprocessing section 223 converts the yH values to positive values. Additionally, in order to convert the polarity of the HH values, the preprocessing section 223 multiplies the HH values by the polarities in coordination with the stored polarities at these positions. The calculated yH and HH are output to the search section 224.

Search section 224 searches for parts of the fixed codebook using separation results received as input from sorting section 222, coding distortion depending on perceived weighting in perceived weighting section 111, and yH and HH received as input from preprocessing section 223. The search section 224 outputs the fixed codebook vector code obtained in the search step to the fixed codebook 104 and outputs the fixed codebook vector code obtained as a search result outside the CELP encoding device 100 and the amplification codebook search section 203. Also, the search for parts of the fixed codebook in the search section 224 will be described in detail later.

Next, the processing for calculating the maximum correlation value of each pulse in section 221 for calculating the maximum correlation value will be described in detail.

FIG. 3 is a flowchart showing the steps of calculating the maximum correlation value of each pulse in section 221 of calculating the maximum correlation value. An example of processing will be explained here, where the maximum correlation value calculating section 221 finds two estimated positions, where the pulse value 0 (yH) is the highest, and, based on these positions, calculates the maximum pulse correlation value 0.

First, the maximum correlation value calculating section 221 provides a sequence ici0 [8] of predetermined estimated pulse positions 0 and a sequence yH [32] obtained by converting the correlation value yH used for search to a positive value (ST 1010).

Next, the maximum correlation value calculation section 221 initializes the maximum value max00, the half-maximum value (i.e., the second highest value) max01 and the counter i (ST 1020), and the step proceeds to the cycle generated by ST 1030 to ST 1080.

In this cycle, when the value of counter i is equal to or greater than 8 (“Yes” in ST 1040), the maximum correlation value calculating section 221 determines that the cycle processing for each assumed position is complete and completes the processing. In contrast, when the counter value i is less than 8 (“No” on ST 1040), the maximum correlation value calculating section 221 decides that the loop processing is not fully completed, and the step proceeds to ST 1050.

Further, if the correlation value yH [ici0 [i]] in the position shown by the counter i is greater than the maximum value max00 (“Yes” in ST 1050), the maximum correlation value calculating section 221 stores the maximum value max00 as the half-maximum value max01, assigns the correlation value yH [ici0 [i]] in the position shown by counter i to the maximum value max00 (ST 1060) and returns the step to ST 1030. If the correlation value yH [ici0 [i]] in the position shown by counter i, equal to or less than the maximum value max00 ("No" on S T 1050), the maximum correlation value calculation section 221 transfers the step to ST 1070.

Further, if the correlation value yH [ici0 [i]] in the position shown by the counter i is greater than the half-maximum value max01 (“Yes” in ST 1070), the maximum correlation value calculating section 221 assigns the correlation value yH [ici0 [i] ] at the position shown by counter i to the half-maximum value max01 (ST 1060) and returns the step to ST 1030 (ST 1080). In contrast, if the correlation value yH [ici0 [i]] in the position shown by the counter i is equal to or less than the half-maximum value max01 (“No” in ST 1070), the maximum correlation value calculating section 221 returns the step to ST 1030.

Next, in ST 1030, the maximum correlation value calculating section 221 increments the counter i by one and returns the step to ST 1040.

Thus, the maximum correlation value calculating section 221 calculates the maximum value max00 and the half-maximum value max01 among the correlation values of a single pulse 0 at the assumed positions. Additionally, using the steps shown in FIG. 3, the maximum correlation value calculating section 221 finds two estimated positions where the correlation values (yH) of the individual pulses 1, 2 and 3 are the highest. That is, the maximum correlation value calculating section 221 finds max10, max11, max20, max21, max30, and max31, which represent the maximum values and half-maximum values of the individual pulses 1, 2, and 3.

Further, using maximum values and half-maximum values

 among the correlation values of the individual pulses 0, 1, 2 and 3, the maximum correlation value calculating section 221 calculates the maximum correlation values S [0], S [1], S2 [2] and S [3] of the individual pulses according to the following equation 5. How shown in FIG. 5, the maximum correlation value calculation section 221 finds stable maximum correlation values associated with individual pulses by adding the half-maximum value of the correlation value at a predetermined coefficient to the maximum value ju values of correlation on the basis of momentum.

S [0] = max00 + max01 × 0.05

S [1] = max10 + max11 × 0.05

S [2] = max20 + max21 × 0.05

S [3] = max30 + max31 × 0.05 ... (Equation 5)

Next, sorting processing over the maximum correlation value of each pulse in sorting section 222 will be described in detail.

4 is a flowchart showing the steps of sorting processing of maximum correlation values of individual pulses in sorting section 222.

First, the sorting section 222 accepts as input the maximum correlation value S [j] (j = 0, 1, 2, 3) of each pulse from the maximum correlation value calculating section 221 and resets counter i to 0, to which rank the sorting is completed (ST 2010).

Further, when the value of counter i is equal to or greater than 4 (“Yes” in ST 2030), the sorting section 222 decides that the sorting is complete and transfers the step to ST 2100. In contrast, when the value of the counter i is less than 4 ( “No” on ST 2030), sorting section 222 assigns 0 to pulse number N [i], resets counter j to 0 to calculate the number of cycles in which the ith maximum correlation value S [N (i)] is searched, and resets the variable "max", which stores the maximum value, at 0 (ST 2040).

Further, when the counter value j is less than 4 (“No” in ST 2060), the sorting section 222 transfers the step to ST 2070.

Further, when the maximum correlation value S [j] is greater than the variable "max" ("Yes" on ST 2070), the sorting section 222 assigns the maximum correlation value S [j] to the variable "max", assigns the counter value j to the pulse number N [ i] corresponding to the ith maximum correlation value S [N [i]] (ST 2080), and transfers the step to ST 2050. In contrast, when the maximum correlation value S [j] is equal to or less than the variable "max "(" No "on ST 2070), sorting section 222 moves the step to ST 2050. Next, to ST 2050, sorting section 222 increments counter j by one and rotates at step ST 2060.

In contrast, when the counter value j is equal to or greater than 4 on the ST 2060 (“Yes” on the ST 2060), the sorting section 222 decides that the cycle generated from ST 2050 to ST 2080 to search for the ith maximum correlation value S [N [i]], ends, and assigns “-1” to the i-th maximum correlation value S [N [i]] (ST 2090). Thereby, the ith maximum correlation value S [N [i]] is excluded from the loop processing purpose to search for the (i + 1) th maximum correlation value S [N [i + 1]]. Next, sorting section 222 increments counter i by one at ST 2020 and returns the step to ST 2030.

Thus, the sorting section 222 sorts the maximum correlation values S [0], S [1], S [2] and S [3] of the individual pulses in descending order and obtains N [i] showing the result of the sorting. In the following, an illustrative case will be explained where the sorting section 222 receives N [i] = {2, 0, 3, 1}. That is, it is assumed that the pulse number N [0] corresponding to the highest maximum correlation value S [N [0]] is 2, then 0, 3 and 1 are in order.

Next, in ST 2100, the sorting section 222 determines the search order of the pulses by grouping four pulse numbers, N [i], corresponding to the sorted maximum correlation values, into two separation patterns of the predefined subsets and outputs the resulting search order to the search section 224. That is, before searching for parts of the fixed codebook in the search section 224, the sorting section 222 determines the numbers of two pulses to be searched first and the numbers of two pulses to be searched further. In the sorting section 222, the three prospective search order patterns shown in the following equation 6 are set in advance.

{First subset} {Second subset} First prospective
template: {N [0], N [1]} {N [2], N [3]} Second alleged
template: {N [0], N [2]} {N [3], N [1]} Third alleged
template: {N [0], N [3]} {N [1], N [2]} ... (Equation 6)

In the search for parts, there are many types of separation patterns for a subset to be searched first (i.e., a first subset), and for a subset to be searched next (i.e., a second subset). For these separation patterns, as shown in Equation 6, by adopting a separation pattern where the pulse N [0] of the highest maximum correlation value is included in the subset to be searched first (ie, the first subset), it is possible to provide good coding efficiency.

In each prospective search order in equation 6, the search is performed in order from the subset to be searched by the first (first subset) to the subset to be searched by the second (second subset).

If N [i] in Equation 6 is expressed by specific values obtained by sorting, the following Equation 7 is obtained and the search is performed in order from the first candidate pattern to the second candidate pattern and the third candidate pattern.

{First subset} {Second subset} First prospective
template: {twenty} {3, 1} Second alleged
template: {2, 3} {10} Third alleged
template: {2, 1} {0, 3} ... (Equation 7)

The three search orders shown in equation 7 can be grouped in M [3] [4], shown in the following equation 8. Here, M [3] [4] represents the search order of pulses in the case of performing a part search for a set of four pulses of three times.

M [3] [4] = {{2,0,3,1}, {2,3,1,0}, {2,1,0,3}} ... (Equation 8)

That is, the sorting section 222 outputs M [3] [4] to the search section 224 as a search order.

Next, the search for parts of the fixed codebook in the search section 224 will be described in detail.

5 and 6 are flowcharts showing the steps of searching for parts of a fixed codebook in a search section 224. Here, the algebraic codebook parameters are shown below.

(1) number of bits: 16 bits (2) processing unit (subframe length): 32 (3) number of pulses: four

With these parameters, the following algebraic codebook is constructed.

ici0 [8] = {0, 4, 8, 12, 16, 20, 24, 28}

ici1 [8] = {1, 5, 9, 13, 17, 21, 25, 29}

ici2 [8] = {2, 6, 10, 14, 18, 22, 26, 30}

ici3 [8] = {3, 7, 11, 15, 19, 23, 27, 31}

First, at ST 3010, the search section 224 prepares the sequences ici0 [8], ici1 [8], ici2 [8] and ici3 [8] showing the estimated positions of the four pulses of the fixed codebook, and prepares the sequence yH [32] obtained by converting yH to positive values, the sequence HH [32] [32] obtained by adjusting the polarities HH, and the vector pol [32], which stores the polarity values (-1, +1) yH before yH is converted to positive values. Next, in ST 3020, variables that are used in a subsequent search loop are initialized.

Search section 224 compares “j” and the value “3” to ST 3030, and if “j” is equal to or greater than 3, transfers the step to ST 3250 to complete the search, and if “j” is less than 3, transfers the step to initializing ST 3050. At ST 3040, “j” is incremented by one. By this, the search section 224 searches for parts for a set of two subsets three times according to the three search orders shown in the search order M [3] [4] received as input from the sorting section 222.

ST 3050 through ST 3130 show the processing of the search cycle, the first subset. To be more specific, in ST 3050, a search cycle for a first subset is initialized. Next, in ST 3060, the search section 224 compares i0 and the value “8” and, if i0 is equal to or greater than 8, transfers the step to ST 3140 to initiate the next search cycle or, if i0 is less than 8, transfers the step to ST 3070 At ST 3070, the correlation value sy0 and the drive power sh0 of the pulse shown by M [j] [0] (j = 0, 1, 2) are calculated. Additionally, the counter i1 is initialized to zero. Next, on ST 3080, i0 is increased by one. By this, the search section 224 performs eight cycle processing for eight estimated pulse positions shown by M [j] [0] (J = 0, 1, 2). Similarly, in ST 3090 to ST 3130, the search section 224 performs loop processing eight times for eight estimated pulse positions shown by M [j] [1] (j = 0, 1, 2).

First, “i1” and the value “8” are compared in decision ST 3090, and if “i1” is equal to or greater than 8, the step goes to step ST 3080 increase, or, if “i1” is less than 8, the step goes to step ST 3100. At ST 3100, the search section 224 calculates the correlation value sy1 and the drive power sh1 of the pulse shown by M [j] [1] (j = 0, 1, 2) using the correlation value sy0 and the drive power sh0 calculated on ST 3070, in addition to yH and HH, accepted as input from the pre-processing section 223.

At ST 3120, the search section 224 calculates and compares the values of the function C according to equation 4, using the correlation values and the excitation powers of the individual pulses, which are the processing goals in the first subset, overwrites and stores i0 and i1 higher functional values in ii0 and ii1 and additionally overwrites and saves the numerator member and the denominator member of function C (ST 3130). Here, in ST 3120, a separation requiring some large amount of computation is avoided, and calculation and comparison are performed by cross-multiplying the denominator and the numerator. In the above solution, if the correlation values are lower or if the correlation values are higher and the processing is performed at ST 3130, the step proceeds to the increment step ST 3110. At step ST 3110 increase, "i1" increase by one.

ST 3140 to ST 3220 shows the processing of the search cycle of the second subset. Here, the processing of the search cycle of the second subset takes basically the same steps as in the processing of the search cycle of the first subset shown in ST 3050 to ST 3130. Here, only differences from the processing of the search cycle of the first subset will be described. First, in ST 3140, initialization of the processing of the search cycle of the second subset is performed using the result of the processing of the search cycle of the first subset. Also, the purpose of processing the search cycle of the second subset is pulses shown by M [j] [2] (j = 0, 1, 2) and M [j] [3] (j = 0, 1, 2). Also, in ST 3160, the correlation value sy2 and the drive power sh2 of pulse 2 are calculated using counter information ii0 and ii1, which are searched for and stored in the search loop for the first subset. Also, similarly, in ST 3190, the correlation value sy3 and the drive power sh3 of pulse 3 are calculated using counter information ii0 and ii1, which are searched and stored in the search loop for the first subset.

Further, on ST 3230 and ST 3240, the search section 224 finds a combination of impulse positions in which the value of the function C is the highest in the whole part search.

Next, in ST 3250, the search section 224 selects ii0, ii1, ii2 and ii3 as pulse position information. Also, the sequence value pol represents the polarity (± 1), and the search section 224 converts the polarity p0, p1, p2 and p3 to 0 or 1 according to the following equation 9 and encodes the results in one bit.

p0 = (pol [ichi0 [ii0]] + 1) / 2

p1 = (pol [ichi1 [ii1]] + 1) / 2

p2 = (pol [ichi2 [ii2]] + 1) / 2

p3 = (pol [ichi3 [ii3]] + 1) / 2 ... (Equation 9)

Here, as a method for decoding position and polarity information, pulse positions are decoded using ichi0 [ii0], ichi1 [ii1], ichi2 [ii2] and ichi3 [ii3] and the fixed codebook vector are decoded using decoded positions and polarities.

As shown in FIGS. 5 and 6, the search section 224 searches for parts for two subsets, so it is possible to reduce the amount of computation significantly, compared to the case of a full search. To be more specific, while loop processing is performed 4096 (8 ⁴ ) times in a full search, according to the method shown in FIG. 5 and FIG. 6, loop processing is performed 64 (8 ² ) times for a search in every two subsets . Additionally, according to M [3] [4], parts are searched for for a set of two subsets three times and, as a result, cycle processing is performed 384 (64 × 2 (subsets) × 3) times as a whole. This is one tenth of the calculation in the case of a full search.

Thus, according to the present embodiment, the parts of the fixed codebook are searched, so that it is possible to reduce the amount of computation compared to the case of performing a full search.

Further, according to the present embodiment, in the search for parts, when dividing the pulses forming the fixed codebook into a subset to be searched first, and a subset to be searched further, a subset to be searched first is generated using the pulse of the highest maximum correlation value, so that it is possible to suppress coding distortion caused by the search for parts. That is, even in the case of performing a full search, a pulse at a position of a higher maximum correlation value is most likely to be received, so that it is possible to suppress coding distortion by preliminary searching for a pulse in the part search.

Also, although the case of the present embodiment has been described above, where the number of pulses is four and the number of splits is two, the present invention does not depend on the number of pulses or the number of splits, and, by determining the order of pulses to be searched based on the result of sorting the maximum the correlation values of individual pulses, it is possible to provide the same effect as the present embodiment.

Also, an exemplary case has been described above with the present embodiment, where the maximum correlation value calculation section 221 calculates the maximum correlation value by adding a half-maximum value of the correlation value at a predetermined coefficient to the maximum value of the correlation value on the basis of the pulse. However, the present invention is not limited to this, and it is equally possible to calculate the maximum correlation values by adding the third highest correlation values at a predetermined coefficient to the above values in individual pulses, or it is equally possible to use the maximum value among the correlation values of individual pulses, as for the maximum values correlations.

Also, although an illustrative case has been described above with the present embodiment, where the estimated positions of each pulse are not preselected, the present invention is not limited thereto, and it is equally possible to sort after the estimated positions of each pulse are preselected. Through this, it is possible to improve the efficiency of sorting.

Also, although an illustrative case where an algebraic codebook is used as a fixed codebook has been described above with the present embodiment, the present invention is not limited to this, and it is equally possible to use a multi-pulse codebook as a fixed codebook. That is, it is possible to implement the present embodiment using position information and polarity information of multiple pulses.

Also, although an illustrative case has been described with the present embodiment above where the CELP encoding scheme is used as a speech encoding scheme, the present invention is not limited to this, and an essential requirement is to adopt a codebook encoding scheme that stores drive vectors where the number of vectors excitation is known. This is because the search for parts according to the present invention is performed only to search for a fixed codebook, and does not depend on whether or not an adaptive codebook is present and whether or not the spectral envelope analysis method is one of the LPC, FFT and filter block .

Option 2 implementation

Embodiment 2 of the present invention is basically the same as Embodiment 1, and differs from Embodiment 1 only in sorting processing in sorting section 222 (see FIG. 4). In the subsequent sorting section in this present embodiment, a reference position “422” is assigned and placed instead of the sorting section 222, and only sorting processing in the sorting section 422 (not shown) will be explained.

7 is a flowchart showing steps of sorting processing of a maximum correlation value of each pulse in sorting section 422 according to the present embodiment. Here, the steps shown in FIG. 7 include basically the same steps as in FIG. 4, and therefore, the same reference numbers will be assigned to the same steps and their descriptions will be omitted.

On ST 4040, the sorting section 422 assigns “0” to the pulse number N [i], resets the counter j, which counts the number of cycles to search for the i-th maximum correlation value S [N [i]], to “0”, resets the variable “ max ", which stores the maximum value, is in" 0 "and assigns" 0 "to the variable L [i] to store the ith maximum correlation value S [N [i]].

In ST 4090, the sorting section 422 assigns the ith maximum correlation value S [N [i]] for L [i] and assigns “-1” for S [N [i]]. Thereby, the ith maximum correlation value S [N [i]] is stored in L [i] and is also excluded from the loop processing purpose to search for the (i + 1) th maximum correlation value S [N [i + 1]].

By processing at ST 2010 to ST 4090, the sorting section 422 sorts the maximum correlation values S [0], S [1], S [2] and S [3] of individual pulses in descending order and receives N [i] and L [i ] showing the result of the sort.

At ST 4100, the sorting section 422 determines the pulse search order by grouping four pulse numbers, N [i], corresponding to the sorted maximum correlation values, into two separation patterns of the predefined subsets and outputs the resulting search order to the search section 224. That is, before searching for parts of the fixed codebook in the search section 224, the sorting section 422 determines the numbers of two pulses to be searched first and the numbers of two pulses to be searched further. In sorting section 422, three prospective search order patterns are set in advance. Here, the difference from the sorting section 222 of Embodiment 1 is that, for the third putative position, the search order is determined using L [i] storing the maximum correlation values.

To be more specific, the sorting section 422 first sets two prospective search orders for the first prospective position and the second prospective position shown in the following equation 10 using the sort result N [i]. That is, as shown in equation 10, the sorting section 422 includes a pulse of the highest maximum correlation value in the subset to be searched first, in the first putative position and second putative position, thereby improving coding efficiency.

{First subset} {Second subset} First prospective
position: {N [0], N [1]} {N [2], N [3]} Second alleged
position: {N [0], N [2]} {N [3], N [1]} ... (Equation 10)

Next, the sorting section 422 sets the third prospective search order using the sorting result N [i] and L [i] as follows. That is, the sorting section 422 decides whether L [2] + L [3] is equal to or greater than (L [0] + L [1]) × 0.91, and if L [2] + L [3] is equal to or greater than (L [0] + L [1]) × 0.91, takes {N [2], N [3]} {N [0], N [1]} as the third assumed position. If L [2] + L [3] is less than (L [0] + L [1]) × 0.91, the sorting section 422 then decides if L [1] + L [3] is equal to or greater than (L [ 0] + L [2]) × 0.94. If L [1] + L [3] is equal to or greater than (L [0] + L [2]) × 0.94, sorting section 422 receives {N [1], N [3]} {N [2] , N [0]} as the third alleged position. If L [1] + L [3] is less than (L [0] + L [2]) × 0.94, the sorting section 422 then decides if L [0] + L [3] is equal to or greater than L [1 ] + L [2]. If L [0] + L [3] is equal to or greater than L [1] + L [2], sorting section 422 generates {N [0], N [3]} {N [1], N [2]} as the third proposed position, or if L [0] + L [3] is less than L [1] + L [2], takes {N [1], N [2]} {N [3], N [0 ]} as the third alleged provision.

When adopting the third estimated search order, when the differences between the maximum pulse correlation values are small, the sorting section 422 generates a subset to be searched first, which does not always include the pulse of the highest maximum correlation value, in order to reduce the search redundancy in the subsequent search section 224. That is, sorting section 442 generates a plurality of combinations of maximum correlation values of individual pulses based on the sorting result N [i] and groups four pulses into two subsets based on a comparison result of the plurality of generated combinations multiplied by a certain coefficient.

For example, when N [i] = {2, 0, 3, 1} and L [i] = {9.5, 9.0, 8.5, 8.0} are found as a result of sorting, L [2] + L [3] is less than (L [0] + L [1]) × 0.91 and L [1] + L [3] is greater than or equal to (L [0] + L [2]) × 0.94. Therefore, the sorting section 422 accepts {N [1], N [3]} {N [2], N [0]} as the third proposed position.

When N [i] is represented by specific values, the first, second, and third putative positions are expressed in the following equation 11.

{First subset} {Second subset} First prospective
position: {twenty} {3, 1} Second alleged
position: {2, 3} {10} Third alleged
position: {0, 1} {3, 2} ... (Equation 11)

The three prospective search orders shown in equation 11 can be grouped in M [3] [4] shown in the following equation 12.

M [3] [4] = {{2,0,3,1}, {2,3,1,0}, {0,1,3,2}} ... (Equation 12)

Sorting section 422 outputs M [3] [4] to search section 224 as prospective search orders.

Thus, according to the present embodiment, when dividing the pulses forming the fixed codebook into a subset to be searched first and a subset to be searched further in the search for parts, a subset to be searched first that does not always include a pulse of the highest maximum correlation values are formed on the basis of not only the order of the maximum correlation values of individual pulses, but also the values of the maximum correlation values of these pulses. By this, it is possible to reduce search redundancy in the search for parts.

Also, although an illustrative case has been described above with the present embodiment, where coefficients such as 0.91 and 0.94 are used to take the third prospective search order, the present invention is not limited thereto and it is equally possible to use other coefficients determined statistically in advance.

Also, although an illustrative case has been described above with the present embodiment, where L [i] is additionally used in addition to N [i] to take a third prospective search order, the present invention is not limited to this, and it is equally possible to use as N [ i] and L [i] even if the first prospective search order or second prospective search order is adopted.

Option 3 implementation

Embodiment 3 is basically the same as Embodiment 1, and differs from Embodiment 1 only in that the pulses grouped into subsets are further rearranged according to a predetermined order. That is, the present embodiment differs from Embodiment 1 only in the sorting processing part shown in FIG. 4. In the subsequent sorting section in this present embodiment, the reference numeral “522” is assigned and placed instead of the sorting section 222, and only sorting processing in the sorting section 522 (not shown) will be explained.

FIG. 8 is a flowchart showing steps of sorting processing of maximum correlation values of individual pulses in sorting section 522 according to the present embodiment. Here, the steps shown in FIG. 8 include basically the same steps as in FIG. 4, and therefore, the same steps will be assigned the same reference numbers and descriptions thereof will be omitted.

In ST 5100 of FIG. 8, although sorting section 522 performs basically the same processing as processing in ST 2100 of FIG. 4 performed by sorting section 222 according to Embodiment 1, sorting section 522 is different from sorting section 222 in that there is no output of the resulting M [3] [4] as such, and its output to the search section 224 after subsequent processing on ST 5110 instead.

In ST 5110, the sorting section 522 generates M '[6] [2] by grouping the elements included in M [3] [4] into pairs of two pulses and adjusts the permutation order of the two pulses included in M' [6] [ 2], into one of {0, 1}, {1, 2}, {2, 3}, {3, 0}, {0, 2} and {1, 3}.

FIG. 9 is a flowchart showing steps in the sorting section 522 of ST 5110 shown in FIG.

First, in ST 6010, the sorting section 522 initializes the variable “i” to “0”.

Next, in ST 6020, the sorting section 522 decides whether the “i” is equal to “6”.

If it is decided at ST 6020 that “i” is “6” (“Yes” at ST 6020), the sorting section 522 ends the processing shown in FIG. 9 (i.e., processing at ST 5110).

In contrast, if it is decided on ST 6020 that “i” is not equal to “6” (“No” on ST 6020), the sorting section 522 transfers the step to ST 6030.

At ST 6030, the sorting section 522 decides whether it is true that M '[i] [1] = "2" and M' [i] [2] = "1".

If it is decided on ST 6030 that M '[i] [1] = "2" and M' [i] [2] = "1" ("Yes" on ST 6030), sorting section 522 sets M '[i] [1] to “1” and M '[i] [2] to “2” on ST 6040 and transfers the step to ST 6150.

In contrast, if it is decided on ST 6030 that the two conditions M '[i] [1] = "2" and M' [i] [2] = "1" do not occur at the same time ("No "at ST 6030), sorting section 522 transfers the step to ST 6050.

In ST 6050, the sorting section 522 decides whether it is true that M '[i] [1] = "3" and M' [i] [2] = "2".

If it is decided on ST 6050 that M '[i] [1] = "3" and M' [i] [2] = "2" ("Yes" on ST 6050), sorting section 522 sets M '[i] [1] to “2” and M '[i] [2] to “3” on ST 6060 and transfers the step to ST 6150.

In contrast, if it is decided on ST 6050 that the two conditions M '[i] [1] = "3" and M' [i] [2] = "2" are not satisfied at the same time ("No "at ST 6050), sorting section 522 transfers the step to ST 6070.

In ST 6070, the sorting section 522 decides whether it is true that M '[i] [1] = "4" and M' [i] [2] = "3".

If it is decided on ST 6070 that M '[i] [1] = "4" and M' [i] [2] = "3" ("Yes" on ST 6070), the sorting section 522 sets M '[i] [1] to “3” and M '[i] [2] to “4” on ST 6080 and transfers the step to ST 6150.

In contrast, if it is decided on ST 6070 that the two conditions M '[i] [1] = "4" and M' [i] [2] = "3" are not satisfied at the same time ("No "at ST 6070), sorting section 522 transfers the step to ST 6090.

At ST 6090, the sorting section 522 decides whether it is true that M '[i] [1] = "1" and M' [i] [2] = "4".

If it is decided on ST 6090 that M '[i] [1] = "1" and M' [i] [2] = "4" ("Yes" on ST 6090), sorting section 522 sets M '[i] [1] to “4” and M '[i] [2] to “1” to ST 6100 and transfers the step to ST 6150.

In contrast, if it is decided on ST 6090 that the two conditions M '[i] [1] = "1" and M' [i] [2] = "4" are not satisfied at the same time ("No "at ST 6090), sorting section 522 transfers the step to ST 6110.

In ST 6110, the sorting section 522 decides whether it is true that M '[i] [1] = "3" and M' [i] [2] = "1".

If it is decided on ST 6110 that M '[i] [1] = "3" and M' [i] [2] = "1" ("Yes" on ST 6110), sorting section 522 sets M '[i] [1] to “1” and M '[i] [2] to “3” on ST 6120 and transfers the step to ST 6150.

In contrast, if it is decided on ST 6110 that the two conditions M '[i] [1] = "3" and M' [i] [2] = "1" are not satisfied at the same time ("No "at ST 6110), sorting section 522 transfers the step to ST 6130.

In ST 6130, the sorting section 522 decides whether M '[i] [1] = "4" and M' [i] [2] = "2" is true.

If it is decided on ST 6130 that M '[i] [1] = "4" and M' [i] [2] = "2" ("Yes" on ST 6130), sorting section 522 sets M '[i] [1] to “2” and M '[i] [2] to “4” to ST 6140 and transfers the step to ST 6150.

In contrast, if it is decided on ST 6130 that the two conditions M '[i] [1] = "4" and M' [i] [2] = "2" are not satisfied at the same time ("No "at ST 6130), sorting section 522 transfers the step to ST 6150.

In ST 6150, sorting section 522 increments “i” by one and transfers the step to ST 6020.

For example, if the sorting section 522 generates M '[6] [2] = {{2, 0}, {3, 1}, {2, 3}, {1, 0}, {2, 1}, {0, 3}} using M [3] [4] = {{2, 0, 3, 1}, {2, 3, 1, 0}, {2, 1, 0, 3}} and further adjusts the order of the pairs of two pulses included in M '[6] [2], according to the steps shown in Fig.9, is M' [6] [2] = {{0, 2}, {1, 3}, {2, 3} , {0, 1}, {1, 2}, {3, 0}}. Additionally, sorting section 522 generates M [3] [4] = {{0, 2, 1, 3}, {2, 3, 0, 1}, {1, 2, 3, 0}} again using M ' [6] [2] = {{0, 2}, {1, 3}, {2, 3}, {0, 1}, {1, 2}, {3, 0}} obtained by adjusting, and displays it in a search section 224.

The effect of the adjustment processing in the sorting section 522 shown in FIG. 9 will be described below.

The search for the pulses forming the fixed codebook is performed by searching for the pulse positions and polarities by which the function C in the above equation 4 is maximized. Therefore, when searching, you need a memory (RAM: random access memory) for the matrix HH of the denominator in equation 4. For example, when the excitation vector is 32, a memory is needed that supports half the 32 × 32 matrix, including the diagonal vector. That is, a memory of (32 × 32/2 + 16) bytes = 528 bytes is needed. Here, a memory is needed that supports the full matrix (32 × 32 bytes = 1024 bytes) in order to reduce the amount of computation required to access the assigned index when calculating, and therefore a larger memory is needed.

In comparison with this, as the present invention, by preliminary dividing the pulses forming the fixed codebook into pairs of the subset to be searched first and the subset to be searched further, and by searching for pulses in the basis based on the pair, only an 8 × 8 matrix is required ( square of the number of entries per pair), so it is possible to save memory capacity up to 8 × 8 × 6 = 384 bytes. In this case, this matrix is not a symmetric matrix, and therefore, the matrix changes if the order of the momentum numbers is reversed. As a result, it is necessary to additionally prepare the inverse matrix (which doubles the memory capacity), change the access method during the search (which increases the amount of computation) or prepare the program based on a combination of a pair (which increases the memory capacity and the computation amount). Therefore, the present embodiment permutes the order of the pulses in a pair-based search and limits the full search to six pairs. By this means, it is possible to limit the memory capacity required for a pulse search to 384 bytes, as above, and to reduce the amount of computation.

Thus, according to the present embodiment, when grouping the pulses forming the fixed codebook into pairs, the pulses to be grouped are rearranged in a predetermined order and searched in the basis based on the pair, so that it is possible to reduce the memory capacity and the amount of computation that are required to search for a fixed codebook.

Also, although an illustrative case has been described above with the present embodiment, the pulse pairs to be searched are limited to six patterns {0, 1}, {1, 2}, {2, 3}, {3, 0}, {0 , 2} and {1, 3}, the present invention is not limited to this, and it is equally possible to reverse the order of the pulses included in the above pairs, which does not change the average search efficiency of the pulse.

Embodiments of the present invention have been described above.

Also, a fixed codebook according to the above described embodiments may be referred to as a “noise codebook”, “stochastic codebook” or “random codebook”.

Also, an adaptive codebook may be indicated as an “adaptive excitation codebook”, and a fixed codebook may be indicated as a “fixed excitation codebook”.

Also, the LSP may be indicated as “LSF (linear spectral frequency)”, and the LSF may be replaced by the LSP. Also, although it is possible that the ISP pairs (immittance spectral pairs) are encoded as spectral parameters instead of the LSP, in this case, by substituting the ISP instead of the LSP, it is possible to implement the above embodiments as an ISP encoding device.

Although the case of the above-described embodiments has been described above as an example where the present invention is implemented by hardware, the present invention can be implemented by software.

Additionally, each function block used in the description of each of the above embodiments may typically be implemented as an LSI implemented as an integrated circuit. They can be separate microcircuits or partially or completely contained on a single microcircuit. “LSI” is accepted here, but it may also be indicated as “IC”, “LSI system”, “super LSI” or “ultra LSI” depending on the varying degrees of integration.

Additionally, the circuit integration method is not limited to LSI, and an embodiment using dedicated circuits or general purpose processors is also possible. After the production of LSI, the use of an FPGA (Field Programmable Gate Array) or a processor with a configurable configuration where the connections and settings of the circuit cells in LSI can be reconfigured is also possible.

Additionally, if integrated circuit technology comes out to replace LSI as a result of the development of a semiconductor technology or a derivative of another technology, it is naturally also possible to integrate function blocks using this technology. The use of biotechnology is also possible.

Disclosures of Japanese Patent Application No. 2007-196782, filed July 27, 2007, Japanese Patent Application No. 2007-260426, filed October 3, 2007, and Japanese Patent Application No. 2008-007418, filed January 16, 2008, including descriptions, drawings and abstracts, are included here. link in their entirety.

Industrial applicability

The speech encoding device and the speech encoding method according to the present invention provide the ability to encode speech using a fixed codebook with efficient use of bits and are applicable, for example, to mobile phones in a mobile communication system.

Claims (9)

1. A speech encoding device comprising:
a calculation section that calculates the correlation values at the estimated pulse positions using the target signal and a plurality of pulses forming a fixed codebook, and calculates representative pulse values using the maximum correlation values based on the pulse;
a sorting section that sorts representative values obtained on the basis of the momentum, groups the impulses corresponding to the sorted representative values into a plurality of predetermined subsets, and determines a first subset to be searched first among a plurality of subsets and
a search section that searches for a fixed codebook using the first subset and receives a code indicating the position and polarity of the plurality of pulses to minimize coding distortion. 2. The speech encoding device according to claim 1, in which:
the calculation section calculates the maximum values of the correlation of pulses calculated using the maximum values of the values of the correlation of pulses as representative values and
the sorting section sorts the maximum correlation values. 3. The speech encoding device according to claim 1, wherein the sorting section sets a subset including a pulse corresponding to the highest representative value among the representative values obtained on the basis of the pulse as the first subset. 4. The speech encoding device according to claim 1, in which:
the sorting section groups the pulses corresponding to the sorted representative values into a plurality of combinations of a plurality of predetermined subsets and determines the first subsets in a plurality of combinations, respectively, and
the search section searches for a fixed codebook using the first subsets and obtains a code to minimize coding distortion. 5. The speech encoding device according to claim 2, wherein the calculation section calculates the maximum correlation value of each pulse by adding the second highest correlation value multiplied by a predetermined coefficient to the maximum value of the correlation value on the basis of the pulse. 6. The speech encoding device according to claim 1, wherein the sorting section determines a first subset using representative values corresponding to the grouped pulses. 7. The speech encoding device according to claim 1, wherein the sorting section generates a plurality of combinations of representative values corresponding to the grouped pulses, and determines a first subset based on a comparison result of the combinations multiplied by a predetermined value. 8. The speech encoding device according to claim 1, wherein the sorting section permutes the pulses to be grouped into a plurality of subsets in a predetermined order. 9. A method for encoding speech, comprising the steps of:
calculating the correlation values at the estimated positions of the pulses using the target signal and the set of pulses forming a fixed codebook, and calculating on the basis of the pulse representative values of the pulses using the maximum values of the correlation;
sorting representative values obtained on a basis based on the momentum, grouping the impulses corresponding to the sorted representative values into a plurality of predetermined subsets and determining a first subset to be searched first among the plurality of subsets; and
searching for a fixed codebook using a first subset and generating code-indicating positions and polarities of a plurality of pulses to minimize coding distortion.

Images