RU2464650C2 - Apparatus and method for encoding, apparatus and method for decoding - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: encoding apparatus has a section for converting an input speech/audio signal to the frequency domain to obtain a frequency domain parameter; a section for selecting as the target quantisation range, a subrange from a plurality of subranges, wherein the plurality of subranges is obtained by dividing the frequency domain parameter; a section for quantisation of the form of the frequency domain parameter in the target quantisation range; a section for quantisation of the gain of the frequency domain parameter in the target quantisation range in order to obtain encoded gain information; and a section for determining whether predictive coding should be performed, based on the number of first subranges of the target quantisation range, which are common for the second subranges of the target quantisation range previously selected.

EFFECT: reduced volume of encoded information and audio signal coding errors, and improved quality of the decoded audio signal.

20 cl, 29 dwg

Description

FIELD OF THE INVENTION

[0001] The present invention relates to an encoding device / decoding device and an encoding method / decoding method used in a communication system in which a message is encoded and transmitted and received and decoded.

State of the art

[0002] When a voice / audio signal is transmitted in a mobile communication system or a packet communication system symbolized by Internet connection, compression / coding technology is often used to increase the transmission efficiency of the speech / audio signal. In addition, in recent years, a scalable encoding / decoding method has been developed that allows a good quality decoded signal to be obtained from part of the encoded information, even if a transmission error occurs during transmission.

[0003] One compression / coding technique described above is a time domain prediction coding technique that improves compression efficiency by using temporal correlation of a speech signal and / or an audio signal (hereinafter referred to as “speech / audio signal”). For example, in Patent Document 1, a signal of a current frame is predicted from a signal of a past frame, and the prediction coding method is switched according to the prediction error. In addition, Non-Patent Document 1 describes a technology by which a prediction coding method is switched according to a degree of change in a time domain of a speech parameter such as an LSF (Linear Spectral Frequency) and a frame error state.

Patent Document 1: Japanese Patent Application Laid-Open No. HEI 8-211900.

Non-Patent Document 1: Thomas Eriksson, Jan Linden, and Jan Skoglund, "Exploiting Inter-frame Correlation In Spectral Quantization," "Acoustics, Speech, and Signal Processing," 1996. ICASSP-96. Conference Proceedings, May 7-10, 1996, pp. 765-768, Volume 2.

Disclosure of invention

Problems to be Solved by the Invention

[0004] However, in any of the above technologies, predictive coding (prediction coding) is performed based on a time-domain parameter on a frame-by-frame basis, and predictive coding based on a non-time domain parameter, such as a frequency domain parameter, is not mentioned. If a prediction coding method based on a time domain parameter such as described above simply applies to encoding a frequency domain parameter, there is no problem if the quantization target range is the same in the last frame and the current frame, but if the quantization target range is different in the last frame and the current frame, the encoding error and the degree of deterioration of the quality of the decoded audio signal is greatly increased, and the speech / audio signal may not be able to be decoded.

[0005] An object of the present invention is to provide an encoding device, etc., capable of reducing the amount of encoded information of a speech / audio signal, and also capable of reducing encoding errors of a speech / audio signal and deterioration of the quality of the decoded audio signal when a frequency component of an excellent range is made a target quantization in each frame.

Means for solving the mentioned problems

[0006] An encoding device according to the present invention uses a configuration having: a conversion section that converts an input signal to a frequency domain to obtain a frequency domain parameter; a selection section that selects a quantization target range from a plurality of subbands obtained by dividing a frequency domain, and generates range information indicating a quantization target range; a quantization section of the shape (signal), which quantizes the shape (signal) of the frequency domain parameter in the quantization target range; and a gain quantization section that encodes (coefficient or value) a gain of a frequency domain parameter in a target quantization range to obtain encoded gain information.

[0007] A decoding apparatus according to the present invention uses a configuration having: a receiving section that receives information indicating a quantization target range selected from a plurality of subbands obtained by dividing the frequency domain of the input signal; a shape (signal) dequantization section that decodes shape encoding information in which the frequency domain parameter shape in the quantization target range is quantized to generate a decoded shape (decoded signal); a gain dequantization section that decodes encoded gain information in which a gain of a frequency domain parameter in a target quantization range is encoded to generate a decoded gain, and decodes a frequency parameter using a decoded shape and a decoded gain to generate a decoded frequency domain parameter; and a time-domain conversion section that converts the decoded frequency-domain parameter to the time-domain to obtain a decoded time-domain signal.

[0008] The encoding method according to the present invention has: a step of converting an input signal into a frequency domain to obtain a frequency domain parameter; a step of selecting a quantization target range from a plurality of subbands obtained by dividing a frequency domain, and generating range information indicating a quantization target range; and a step of quantizing a shape of a frequency domain parameter in a quantization target range to obtain a form of encoded information; and encoding a gain of a frequency domain parameter in a target quantization range to obtain encoded gain information.

[0009] The decoding method according to the present invention has: a step of receiving information indicative of a quantization target range selected from a plurality of subbands obtained by dividing the frequency domain of the input signal; a step of decoding the encoded form information in which the frequency domain parameter shape is quantized in the quantization target range to generate a decoded shape; the step of decoding encoded gain information in which the gain of the frequency domain parameter in the target quantization range is quantized to generate a decoded gain, and decoding the frequency domain parameter using the decoded shape and the decoded gain to generate the decoded frequency domain parameter; and the step of converting the decoded parameter of the frequency domain to the time domain to obtain a decoded signal of the time domain.

Advantages of the Invention

[0010] The present invention reduces the amount of encoded information of a speech / audio signal or the like, and can also prevent a sharp deterioration in the quality of a decoded signal, decoded speech, etc., and can reduce the encoding error of a speech / audio signal or the like and a deterioration in the quality of the decoded signal. .

Brief Description of the Drawings

[0011] FIG. 1 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 1 of the present invention;

FIG. 2 is a drawing illustrating an example of a configuration of regions obtained by a range selection section according to Embodiment 1 of the present invention;

3 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 1 of the present invention;

4 is a block diagram illustrating a basic configuration of an embodiment of a speech encoding apparatus according to Embodiment 1 of the present invention;

5 is a block diagram illustrating a basic configuration of an embodiment of a speech decoding apparatus according to Embodiment 1 of the present invention;

6 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 2 of the present invention;

7 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section according to Embodiment 2 of the present invention;

8 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 2 of the present invention;

9 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section according to Embodiment 2 of the present invention;

10 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 3 of the present invention;

11 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 3 of the present invention;

12 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 4 of the present invention;

13 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 4 of the present invention;

14 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 5 of the present invention;

15 is a block diagram illustrating a basic configuration of an inside of a range extension coding section according to Embodiment 5 of the present invention;

16 is a block diagram illustrating a basic configuration of an inside of a correction scale factor encoding section according to Embodiment 5 of the present invention;

17 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section according to Embodiment 5 of the present invention;

Fig. 18 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 5 of the present invention;

FIG. 19 is a block diagram illustrating a basic configuration of an inside of a band extension decoding section according to Embodiment 5 of the present invention;

20 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section according to Embodiment 5 of the present invention;

21 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 6 of the present invention;

FIG. 22 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section according to Embodiment 6 of the present invention; FIG.

23 is a drawing illustrating an example of a configuration of regions obtained by a band selection section according to Embodiment 6 of the present invention;

24 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 6 of the present invention;

25 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section according to Embodiment 6 of the present invention;

26 is a block diagram illustrating a basic configuration of a speech encoding apparatus according to Embodiment 7 of the present invention;

27 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section according to Embodiment 7 of the present invention;

28 is a block diagram illustrating a basic configuration of a speech decoding apparatus according to Embodiment 7 of the present invention; and

29 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section according to Embodiment 7 of the present invention.

The best mode of carrying out the invention

[0012] As a brief overview of an example of the present invention, when quantizing a frequency component of a different range in each frame, if the number of subbands common to the quantization target range of the past frame and the quantization target range of the current frame is determined to be greater than or equal to a predetermined value, a predictive encoding with respect to a frequency domain parameter, and if the number of common subbands is determined to be less than a predetermined value, the parameter is often This area is encoded directly. By this, the amount of encoded information of a speech / audio signal or the like is reduced, as well as a severe deterioration in the quality of the decoded signal, decoded speech, etc. can be prevented, and an encoding error of a speech / audio signal or the like, and a deterioration in the quality of the decoded signal and a deterioration in the quality of the decoded audio speech, in particular, can be reduced.

[0013] Embodiments of the present invention are described below in detail with reference to the accompanying drawings. In the following descriptions, a speech encoding device and a speech decoding device are used as examples of an encoding device and a decoding device according to the present invention.

[0014] Embodiment 1

1 is a block diagram illustrating a basic configuration of a speech encoding apparatus 100 according to Embodiment 1 of the present invention.

[0015] In this figure, the speech coding apparatus 100 is provided with a frequency domain conversion section 101, a band selection section 102, a shape (signal) quantization section 103, predictive coding execution / non-execution determining section 104, and gain quantization (coefficient or value) section 105 and multiplexing section 106.

[0016] The frequency domain transform section 101 performs Modified Discrete Cosine Transform (MDCT) using an input signal to calculate the MDCT coefficient, which is a frequency domain parameter, and outputs it to the band selection section 102.

[0017] The range selection section 102 divides the input MDCT coefficient from the frequency domain conversion section 101 into a plurality of subbands, selects a range as a quantization target range from a plurality of subbands, and outputs the range information indicating the selected range to a shape quantization section 103, a determination section 104 performing / not performing predictive coding and multiplexing section 106. In addition, the range selection section 102 outputs the MDCT coefficient to the shape quantization section 103. The input of the MDCT coefficient to the shape quantization section 103 can also be performed directly from the frequency domain conversion section 101 separately from the input from the frequency domain conversion section 101 to the range selection section 102.

[0018] The shape quantization section 103 quantizes the shape (signal) using an MDCT coefficient corresponding to the range indicated by the range information inputted from the range selection section 102 from among the MDCT coefficients inputted from the range selection section 102, and outputs the obtained encoded information forms into multiplexing section 106. In addition, the shape quantization section 103 finds the ideal shape quantization gain value and outputs the obtained ideal gain value to the gain quantization section 105.

[0019] Predictive encoding execution / non-execution determining section 104 finds the number of subbands common to the quantization target range of the current frame and the quantization target range of the past frame using the range information inputted from the range selection section 102. Then, the predictive encoding execution / non-execution determining section 104 decides that the predictive encoding should be performed with respect to the MDCT coefficient of the quantization target range indicated by the range information if the number of common subbands is greater than or equal to a predetermined value, or decides that the predictive encoding should not be performed in relation to the MDCT coefficient of the quantization target range indicated by the range information, if the number of total subbands is less than dawn its specific value. Predictive encoding run / fail determination section 104 outputs the result of this determination to gain quantization section 105.

[0020] If the determination result inputted from the predictive coding fulfillment / failure determination section 104 indicates that predictive coding is to be performed, the gain quantization section 105 performs predictive encoding of the gain of the quantization target range of the current frame using the quantization gain value of the past frame stored in an internal buffer, and an internal gain codebook to obtain encoded gain information. On the other hand, if the determination result inputted from the predictive coding execution / non-execution determining section 104 indicates that predictive coding should not be performed, the gain quantization section 105 obtains encoded gain information by directly quantizing the ideal gain value input from the section 103 quantization forms. The gain quantization section 105 outputs the obtained encoded gain information to the multiplexing section 106.

[0021] The multiplexing section 106 multiplexes the range information inputted from the range selection section 102, the encoded shape information inputted from the shape quantization section 103, and the encoded gain information inputted from the gain quantization section 105, and transmits the received bit stream to the speech decoding apparatus.

[0022] A speech encoding apparatus 100 having a configuration as described above splits the input signal into sections of N samples (where N is a natural number) and performs frame-by-frame encoding with N samples as one frame. The operation of each section of the speech encoding apparatus 100 is described in detail below. In the following description, the input signal of the frame, which is the encoding target, is represented by x _n (where n = 0, 1 ..., N-1). Here, n indicates the index of each sample in the frame, which is the encoding target.

[0023] The frequency domain transform section 101 has N internal buffers, and first initializes each buffer using the value 0 in accordance with Equation (1) below.

buf _n = 0 (n = 0, 1, ..., N-1) Equation 1

[0024] In this equation, buf _n (n = 0 ..., N-1) indicates the (n + 1) th of N buffers in the frequency domain transform section 101.

[0025] Then, the frequency domain transform section 101 finds the MDCT coefficient X _k by performing a modified discrete cosine transform (MDCT) of the input signal x _n in accordance with Equation (2) below

,

, ( k = 0, ... N-1) Equation 2

[0026] In this equation, k indicates the index of each sample in one frame, and x ' _n is the vector linking the input signal x _n and buf _n in accordance with Equation (3) below.

Equation 3

[0027] Then, the frequency domain transform section 101 updates buf _n (n = 0 ..., N-1), as shown in Equation (4) below.

buf _n = x _n (n = 0, ..., N-1) Equation 4

[0028] Then, the frequency domain transform section 101 provides MDCT coefficients X _k to the band selection section 102.

[0029] The band selection section 102 first divides the MDCT coefficient X _k into a plurality of subbands. Here, a description will be given taking the case in which the MDCT coefficient X _k is equally divided by J subbands (where J is a natural number) as an example. Then, the range selection section 102 selects L consecutive subbands (where L is a natural number) from among J subbands, and obtains M kinds of subband groups (where M is a natural number). Below, these M kinds of subband groups are called regions.

[0030] FIG. 2 is a drawing illustrating an example of a configuration of regions obtained by a range selection section 102.

[0031] In this drawing, the number of subbands is 17 (J = 17), the number of kinds of regions is eight (M = 8), and each region consists of five consecutive subbands (L = 5). Of these, for example, region 4 consists of subbands 6-10.

[0032] Then, the range selection section 102 calculates the average energy E (m) of each of the M kinds of regions in accordance with Equation (5) below.

(m = 0, ..., M-1) Equation 5

[0033] In this equation, j indicates the index of each of the J subbands, m indicates the index of each of the M kinds of regions, S (m) indicates the minimum value among the indices of the L subbands making up the region m, B (j) indicates the minimum value among the indices of the set of coefficients MDCT constituting subband j, and W (j) indicates the bandwidth of subband j. In the following description, the case in which the values of the bandwidth J of the subbands are all equal, that is, the case in which W (j) is a constant, is described as an example.

[0034] Then, the range selection section 102 selects a region, for example, a range consisting of subbands j ″ - j ″ + L-1, for which the average energy E (m) is maximum, as the range that is the target of quantization (target quantization range), and outputs an index m_max indicating this region as range information to the shape quantization section 103, the predictive encoding execution / non-execution determining section 104, and the multiplexing section 106. The range selection section 102 also outputs the MDCT coefficient X _k to the shape quantization section 103. In the following description, range indices indicating the quantization target range selected by the range selection section 102 are assumed to be j ″ - j ″ ″ + L − 1.

[0035] The shape quantization section 103 quantizes the shape (signal) based on a sub-band-by-subband with respect to the MDCT coefficient corresponding to the range indicated by the range information m_max input from the range selection section 102. More specifically, the shape quantization section 103 searches for an internal shape codebook, consisting of the number of SQ shape code vectors, for each of the L subbands, and finds the shape code vector index for which the result of Equation (6) below is maximum.

Уравнение 6

Equation 6

[0036] In this equation, SC ⁱ _k indicates the shape code vector constituting the shape codebook, i indicates the index of the shape code vector, and k indicates the index of the shape code vector element.

[0037] The shape quantization section 103 outputs the shape code vector index S_max, for which the result of Equation (6) is above the maximum, to the multiplexing section 106 as encoded shape information. The shape quantization section 103 also calculates the ideal gain value Gain_i (j) in accordance with Equation (7) below, and outputs it to the gain quantization section 105.

Gain_i (j) =

Equation 7

[0038] The predictive coding fulfillment / failure determination section 104 has an internal buffer that stores range information m_max entered from the range selection section 102 in the last frame. A case is described below by way of example in which the predictive coding fulfillment / failure determination section 104 has an internal buffer that stores m_max range information for three past frames. Predictive encoding execution / non-execution determining section 104 first finds the number of subbands common for the quantization target range of the past frame and the quantization target range of the current frame using the m_max range information input from the shape quantization section 103 and the m_max range information input from the section 103 quantization forms in the current frame. Then, the predictive coding execution / non-execution determining section 104 decides that the predictive coding should be performed if the number of common subbands is greater than or equal to a predetermined value, or decides that the predictive coding should not be performed if the number of common subbands is less than the predetermined value . More specifically, the L subbands indicated by the m_max range information inputted from the band selection section 102 one frame back in time are compared with the L subbands indicated by the m_max range information entered from the m_max range section 102 from the range selection section 102 in the current frame, and it is determined that the prediction coding should be performed if the number of common subbands is P or greater, or it is determined that predictive coding should not be performed if the number of common subbands is less than P. The prediction coding result / non-execution outputs the result of this determination to the gain quantization section 105. Then, the predictive coding fulfillment / failure determination section 104 updates the internal buffer storing the range information using the range information m_max input from the range selection section 102 in the current frame.

[0039] The gain quantization section 105 has an internal buffer that stores the quantization gain value obtained in the last frame. If the determination result inputted from the predictive coding fulfillment / failure determination section 104 indicates that predictive coding is to be performed, the gain quantization section 105 quantizes by predicting the gain value of the current frame using the gain frame quantization value C ^t _j stored in the internal buffer. More specifically, the gain quantization section 105 searches for an internal gain codebook consisting of the number of GQ gain code vectors for each of the L subbands, and finds the gain code vector index for which the result of Equation (8) below is a minimum.

Уравнение 8

Equation 8

[0040] In this equation, GC ⁱ _j indicates the gain code vector constituting the gain codebook, i indicates the index of the gain code vector, and j indicates the index of the gain code vector element. For example, if the number of subbands making up the region is five (L = 5), j has a value from 0 to 4. Here, C ^t _j indicates the gain value of t frames backward in time, so that when t = l, for example, C ^t _j indicates the gain value of one frame back in time. In addition, α is a fourth-order linear prediction coefficient stored in gain quantization section 105. The gain quantization section 105 processes the L subbands within the same region as an L-dimensional vector, and performs vector quantization.

[0041] The gain quantization section 105 outputs the gain code vector index G_min, for which the result of Equation (8) above is a minimum, to the multiplexing section 106, as encoded gain information. If the internal buffer does not have a subband gain value corresponding to the previous frame, the gain quantization section 105 replaces the gain value of the nearest frequency subband in the internal buffer in Equation (8) above.

[0042] On the other hand, if the determination result inputted from the predictive coding execution / non-execution determination section 104 indicates that the predictive coding should not be performed, the gain quantization section 105 directly quantizes the ideal gain value Gain_i (j) input from section 103 quantization of the form, in accordance with Equation (9) below. Here, the gain quantization section 105 processes the ideal gain value as an L-dimensional vector, and quantizes the vector.

Уравнение 9

Equation 9

[0043] Here, the codebook index that makes Equation (9) above the minimum is denoted by G_min.

[0044] The gain quantization section 105 outputs G_min to the multiplexing section 106 as encoded gain information. The gain quantization section 105 also updates the internal buffer in accordance with Equation (10) below using the encoded gain information G_min and the gain quantization value C ^t _j obtained in the current frame.

(j=0,…, L-1)

(j = 0, ..., L-1) Equation 10

[0045] The multiplexing section 106 multiplexes the range information m_max inputted from the range selection section 102, the encoded shape information S_max inputted from the shape quantization section 103, and the encoded gain information G_min inputted from the gain quantization section 105, and transmit the received bitstream to the device speech decoding.

[0046] FIG. 3 is a block diagram illustrating a basic configuration of a speech decoding apparatus 200 according to this embodiment.

[0047] In this drawing, the speech decoding apparatus 200 is provided with a demultiplexing section 201, a shape dequantization section 202, predictive decoding execution / non-execution determining section 203, a gain dequantization section 204 and a time domain conversion section 205.

[0048] The demultiplexing section 201 demultiplexes the range information, encoded form information and encoded gain information from a bit stream transmitted from the speech encoding apparatus 100, outputs the obtained range information to the form dequantization section 202 and the predictive decoding execution / non-execution determination section 203, outputs the obtained encoded form information to the form dequantization section 202, and outputs the obtained encoded gain information to the gain dequantization section 204.

[0049] The shape dequantization section 202 finds the shape value of the MDCT coefficient corresponding to the quantization target range indicated by the range information inputted from the demultiplexing section 201 by dequantizing the encoded form information inputted from the demultiplexing section 201 and outputs the found shape value to the dequantization section 204 gain.

[0050] Predictive decoding run / fail determination section 203 finds the number of subbands common to the current frame quantization target range and past frame quantization range using the range information input from demultiplexing section 201. Then, the predictive decoding run / fail determining section 203 determines that predictive decoding should be performed with respect to the MDCT coefficient of the quantization target range indicated by the range information if the number of common subbands is greater than or equal to a predetermined value, or decides that predictive decoding should not be performed with respect to the MDCT coefficient of the quantization target range indicated by the range information, if the number of total subbands is less than m predefined value. Predictive decoding run / fail determination section 203 outputs the result of this determination to gain dequantization section 204.

[0051] If the determination result inputted from the predictive decoding run / fail determination section 203 indicates that predictive decoding is to be performed, the gain dequantization section 204 performs predictive decoding on the encoded gain information inputted from the demultiplexing section 201 using the past gain value the frame stored in the internal buffer and the internal gain codebook to obtain the gain value. On the other hand, if the determination result inputted from the predictive decoding run / fail determination section 203 indicates that predictive decoding should not be performed, the gain dequantization section 204 obtains the gain value by directly dequantizing the encoded gain information inputted from the demultiplexing section 201, using the internal gain codebook. The gain dequantization section 204 outputs the obtained gain value to the time domain conversion section 205. The gain dequantization section 204 also finds the MDCT coefficient of the quantization target range using the obtained gain value and the shape value input from the form dequantization section 202, and outputs it to the time domain converting section 205 as a decoded MDCT coefficient.

[0052] The time domain converting section 205 performs the Inverse Modified Discrete Cosine Transform (IMDCT) on the decoded MDCT coefficient input from the gain dequantization section 204 to generate a time domain signal and outputs it as a decoded signal.

[0053] A speech decoding apparatus 200 having a configuration such as described above performs the following operations.

[0054] The demultiplexing section 201 demultiplexes the information of the m_max range, the encoded form information S_max, and the encoded gain information G_min from the bit stream transmitted from the speech encoding apparatus 100, outputs the obtained information of the m_max range to the form dequantization section 202 and the predictive execution / non-fulfillment determining section 203 decoding, outputs the obtained encoding information of the form S_max to the form dequantization section 202 and outputs the obtained encoded gain information G_min to the dequantization section 204 amplification.

[0055] The form dequantization section 202 has an internal form codebook similar to the form codebook provided with the form quantization section 103 of the speech encoding apparatus 100, and searches for the form code vector for which the encoded form information S_max input from the demultiplexing section 201 is an index . The shape dequantization section 202 outputs the found code vector to the gain dequantization section 204 as the shape value of the MDCT coefficient of the quantization target range indicated by the m_max range information inputted from the demultiplexing section 201. Here, the shape code vector found as the shape value is denoted as Shape_q (k) (k = B (j``) ..., B (j '' + L) -1).

[0056] The predictive decoding run / fail determination section 203 has an internal buffer that stores information of the m_max range entered from the demultiplexing section 201 in the last frame. Here, a case is described by way of example in which the predictive decoding run / fail determination section 203 has an internal buffer that stores m_max range information for the past three frames. Predictive decoding execution / non-execution determining section 203 first finds the number of subbands common to the quantization target range of the last frame and the quantization target range of the current frame using the m_max range information input from the last frame demultiplexing section 201 and the m_max range information input from the section 201 demultiplexing in the current frame. Then, predictive decoding run / fail determining section 203 decides that predictive decoding should be performed if the number of common subbands is greater than or equal to a predetermined value, or decides that predictive decoding should not be performed if the number of common subbands is less than a predetermined value. More specifically, the predictive decoding run / fail determination section 203 compares the L subbands indicated by the m_max range information entered from the demultiplexing section 201 one frame back in time with the L subbands indicated by the m_max range information entered from the m_max range information entered from the demultiplexing section 201 in the current frame, and decides that predictive decoding should be performed if the number of common subbands is P or greater, or decides that predictive decoding should not be performed but if the number of common subbands is less than P. execution determination section 203 / failure prediction decoding outputs the result of this determination to gain dequantization section 204. Then, the predictive decoding run / fail determination section 203 updates the internal buffer storing range information using the range information m_max input from the demultiplexing section 201 in the current frame.

[0057] The gain dequantization section 204 has an internal buffer that stores the gain value obtained in the last frame. If the determination result inputted from the predictive decoding run / fail determination section 203 indicates that predictive decoding is to be performed, the gain dequantization section 204 dequantizes by predicting the gain value of the current frame using the gain value of the past frame stored in the internal buffer. More specifically, the gain dequantization section 204 has the same kind of internal gain codebook as the gain quantization section 105 of the speech encoding apparatus 100, and obtains the gain value Gain_q ′ by performing gain dequantization in accordance with Equation (11) below. Here, C ″ ^t _j indicates the gain value of t frames backward in time, so that when t = 1, for example, C ″ ^t _j indicates the gain value of one frame backward in time. In addition, α is a fourth order linear prediction coefficient stored in gain dequantization section 204. Gain dequantization section 204 processes the L subbands within the same region as an L-dimensional vector, and dequantizes the vector.

Equation 11

[0058] If there is no subband gain value corresponding to the last frame in the internal buffer, the gain dequantization section 204 replaces the gain value of the nearest subband in frequency in the internal buffer in Equation (11) above.

[0059] On the other hand, if the determination result inputted from the predictive decoding execution / non-execution determining section 203 indicates that the predictive decoding should not be performed, the gain dequantization section 204 dequantizes the gain value in accordance with Equation (12) below using the above gain codebook. Here, the gain value is processed as an L-dimensional vector, and the vector is quantized. It should be said that when predictive decoding is not performed, the gain code vector GCj ^G_min corresponding to the encoded gain information G_min is taken directly as the gain value.

Gain_q '(j + j ”) =

(j = 0, ..., L-1) Equation 12

[0060] Then, the gain dequantization section 204 calculates the decoded MDCT coefficient in accordance with Equation (13) below, using the gain value obtained by dequantizing the current frame and the shape value input from the form dequantization section 202, and updates the internal buffer in accordance with the Equation (14) below. Here, the calculated decoded MDCT coefficient is denoted by X " _k . In addition, when dequantizing the MDCT coefficient, if k is present in B (j") - B (j "+1) -1, the gain value Gain_q '(j) takes the value Gain_q' (j ").

Уравнение 13 X ” _k = Gain_q '(j) · Shape_q' (k)

Equation 13

(j=j”,…, j”+L-1)
Уравнение 14

(j = j ”, ..., j” + L-1)
Equation 14

[0061] The gain dequantization section 204 outputs the decoded coefficient MDCT X ' _k calculated in accordance with Equation (13) above to the time domain transform section 205.

[0062] The time domain converting section 205 first initializes the internal buffer buf ' _{k with a} value of zero in accordance with Equation (15) below.

buf ' _k = 0 (k = 0, ..., N-1) Equation 15

[0063] Then, the time domain converting section 205 finds the decoded signal Y _n in accordance with Equation (16) below using the decoded coefficient MDCT X ″ _k introduced from the gain dequantization section 204.

(n = 0, ..., N-1) Equation 16

In this equation, X2 '' _k is the vector linking the decoded MDCT coefficient X '' _k and the buffer buf ' _k .

Equation 17

[0064] Then, the time domain converting sections 205 updates the buffer buf ' _k in accordance with Equation (18) below.

buf _k = X ” _k  (k = 0, ..., N-1) Equation 18

[0065] The time domain converting section 205 outputs the obtained decoded signal Y _n as an output signal.

[0066] Thus, according to this embodiment, the high energy range is selected in each frame as the quantization target range, and the frequency domain parameter is quantized, allowing an offset in the distribution of the quantized gain value to be created, and to improve the quantization efficiency of the vector.

[0067] Furthermore, according to this embodiment, when quantizing the frequency domain parameter of the different quantization target range of each frame, predictive coding is performed on the frequency domain parameter if the number of subbands common to the quantization target range of the past frame and the quantization target range of the current frame, is defined as greater than or equal to a predetermined value, and the frequency domain parameter is encoded directly if the number of common subbands is is defined as less than a predetermined value. Therefore, the amount of encoded information in speech encoding is reduced, and also a sharp deterioration in speech quality can be prevented, and an encoding error in the speech / audio signal and degradation in the quality of the decoded audio signal can be reduced.

[0068] Furthermore, according to this embodiment, on the encoding side, the quantization target range can be determined, and the quantization of the frequency domain parameter is performed in units of regions, each consisting of a plurality of subbands, and information about the frequency domain parameter, which region became the target of quantization, can be transmitted to the decoding side. Therefore, the quantization efficiency can be improved, and the amount of encoded information transmitted to the decoding side can be further reduced compared to deciding whether predictive encoding should be used based on “subband by subband” and transmitting information regarding which subband has become the target quantization, to the decoding side.

[0069] In this embodiment, a case has been described by way of example where gain quantization is performed in units of regions, each of which consists of a plurality of subbands, but the present invention is not limited thereto, and a quantization target may also be selected based on “subband beyond subband” ”, That is, determining whether predictive quantization should be performed can also be performed based on“ subband by subband ”.

[0070] In this embodiment, a case has been described by way of example in which a predictive gain quantization method must perform linear time domain prediction to amplify the same frequency range, but the present invention is not limited to this, and linear prediction can also be performed in time areas for amplification of excellent frequency ranges.

[0071] In this embodiment, a case has been described in which a conventional speech / audio signal is taken as an example of a signal that becomes a quantization target, but the present invention is not limited thereto, and an excitation signal obtained by processing the speech / audio signal by an inverse LPC filter (linear prediction coefficient) can also be used as a quantization target.

[0072] In this embodiment, a case has been described by way of example in which the region for which the energy value of a particular region — that is, the perceptual value — is largest is selected as a reference in order to select a quantization target range, but the present invention is not limited by this, and in addition to the perceptual value, the frequency correlation with the range selected in the last frame can also be taken into account at the same time. That is, if there are candidate ranges for which the number of subbands common for the quantization target range selected in the past is greater than or equal to a predetermined value and the energy is greater than or equal to a predetermined value, the range with the highest energy among the above candidate ranges can be selected as the target quantization range, and if no such candidate ranges exist, the range with the highest energy among all frequency ranges can be selected as stve quantization target band. For example, if the subband common to the region with the highest energy and the range selected in the last frame does not exist, the number of subbands common for the region with the second highest energy and the range selected in the last frame is greater than or equal to a predetermined threshold value, and the energy of the region with the second highest energy is greater than or equal to a predetermined threshold value, the region with the second highest energy is selected, and not the region with the highest energy. In addition, the range selection section according to this embodiment selects a region closest to the quantization target range selected in the past from among regions whose energy is greater than or equal to a predetermined value as the quantization target range.

[0073] In this embodiment, quantization of the MDCT coefficient can be performed after interpolation is performed using the last frame. For example, the case will be described with reference to FIG. 2, in which the quantization target range of the past frame is region 3 (i.e., subbands 5-9), the quantization target range of the current frame is region 4 (i.e., subbands 6-10), and predictive coding of the current frame is performed using the quantization result of the past frame. In this case, predictive coding is performed on subbands 6-9 of the current frame using subbands 6-9 of the past frame, and for subband 10 of the current frame, subband 10 of the past frame is interpolated using subbands 6-9 of the past frame, and then predictive coding is performed using subrange 10 of the past frame obtained by interpolation.

[0074] In this embodiment, an example has been described of a case in which quantization is performed using the same codebook regardless of whether predictive coding is performed, but the present invention is not limited thereto, and various codebooks can also be used according to whether predictive coding is performed or not when quantizing the gain and when quantizing the shape.

[0075] In this embodiment, a case has been described by way of example in which all the sizes of the subbands are the same, but the present invention is not limited thereto, and the sizes of the individual subbands may also be different.

[0076] In this embodiment, a case has been described by way of example in which the same codebook is used for all subbands in gain quantization and shape quantization, but the present invention is not limited thereto, and different codebooks can also be used based on “ subband by subband ”in gain quantization and shape quantization.

[0077] In this embodiment, a case has been described by way of example in which consecutive subbands are selected as the quantization target range, but the present invention is not limited thereto, and an inconsistent plurality of subbands can also be selected as the quantization target range. In such a case, the speech coding efficiency can also be improved by interpolating the values of the unselected subband using the values of adjacent subbands.

[0078] In this embodiment, an example has been described of a case in which the speech encoding device 100 is provided with the predictive encoding execution / non-execution determining section 104, but the speech encoding device according to the present invention is not limited thereto, and may also have a configuration in which a predictive encoding run / fail determination section 104 is provided, and predictive quantization is not always performed by the gain quantization section 105, as illustrated by the apparatus tion 100a speech coding shown in Figure 4. In this case, as shown in FIG. 4, the speech encoding apparatus 100a is provided with a frequency domain conversion section 101, a band selection section 102, a shape quantization section 103, a gain quantization section 105 and a multiplexing section 106. 5 is a block diagram illustrating a configuration of a speech decoding apparatus 200a corresponding to a speech encoding apparatus 100a, wherein the speech decoding apparatus 200a is provided with a demultiplexing section 201, a shape dequantization section 202, an gain dequantization section 204, and a time domain conversion section 205. In such a case, the speech encoding apparatus 100a partially chooses the range to be quantized from among all the ranges, further divides the selected range into a plurality of subbands, and quantizes the gain of each subband. Thus, quantization can be performed at a lower bit rate than the method in which the components of all ranges are quantized, and coding efficiency can be improved. In addition, coding efficiency can also be improved by quantizing the gain vector using gain correlation in the frequency domain.

[0079] The speech encoding apparatus according to the present invention may also have a configuration in which predictive encoding run / fail determination section 104 is not provided, and predictive quantization is always performed by gain quantization section 105, as illustrated by speech encoding apparatus 100a shown in FIG. 4. The configuration of the speech decoding apparatus 200a corresponding to this kind of speech encoding apparatus 100a is shown in FIG. In such a case, the speech encoding apparatus 100a partially chooses the range to be quantized from among all the ranges, further divides the selected range into a plurality of subbands, and performs gain quantization for each subband. Thus, quantization can be performed at a lower bit rate than in the method in which components of all ranges are quantized, and coding efficiency can be improved. In addition, coding efficiency can also be improved by predictively quantizing the gain vector using time-domain gain correlation.

[0080] In this embodiment, a case has been described by way of example in which the method of selecting the quantization target range in the range selection section is to select the region with the highest energy in all ranges, but the present invention is not limited thereto, and the selection can also be made using information of the range selected in the preceding time frame in addition to the aforementioned criterion. For example, a possible method would be to select the region to be quantized after multiplying by the weighting factor so that the region that includes the range near the range selected in the previous frame in time becomes more preferable for selection. In addition, if there are many levels in which the range to be quantized is selected, the range quantized in the upper level can be selected using the information of the range selected in the lower level. For example, a possible method would be to select a region that should be quantized after multiplying by a weight factor such that a region that includes a range near a range selected in the lower level becomes more preferred.

[0081] In this embodiment, a case has been described by way of example in which the method of selecting the target quantization range is to select the region with the highest energy in all ranges, but the present invention is not limited thereto, and a certain range can also be preselected in advance, after which the quantization target range is finally selected in a preselected range. In such a case, the preselected range may be determined according to the sampling rate of the input signal, bit rate coding, or the like. For example, one way is to pre-select a low range when the bit rate or sample rate is low.

[0082] For example, it is possible to use the method in the range selection section 102 by which the region to be quantized is determined by calculating the energy of the region after restricting the selected regions to regions with lower ranges from among all candidate candidate regions. As an example of this, a possible method is to fulfill the restriction of five candidates from lower ranges from among the total number of eight regions — candidates shown in FIG. 2, and select the region with the highest energy among them. Alternatively, the range selection section 102 may compare the energies after multiplying the energy by the weighting factor so that the region from the lower zone is proportionally more preferred for selection. Another possibility for the range selection section 102 is to select a fixed subband from lower ranges. A feature of the speech signal is that the structure of the harmonics becomes proportionally stronger towards the side of the lower ranges, as a result of which a strong peak is present on the side of the lower ranges. Since this strong peak is difficult to mask, this leads to what is perceived as noise. Here, by increasing the probability of selection toward the side of the lower range, rather than simply selecting an area based on the amount of energy, the possibility of a region that includes a strong selectable peak is increased, and the noise perception is reduced as a result. Thus, the quality of the decoded signal can be improved by restricting the selected areas to the side of the lower range, or by performing the multiplication by the weight factor so that the probability of selection increases towards the side of the lower range, thus.

[0083] The speech encoding apparatus according to the present invention has been described in terms of a configuration whereby quantization of a shape (waveform information) is first performed with respect to a component of a range to be quantized, followed by quantization of a gain (gain information), but the present invention is not limited this, and a configuration can also be used whereby gain quantization is performed first, followed by quantization of the shape (signal).

[0084] Embodiment 2

6 is a block diagram illustrating a basic configuration of a speech encoding apparatus 300 according to Embodiment 2 of the present invention.

[0085] In this drawing, the speech encoding device 300 is provided with a downsizing section 301, a first level encoding section 302, a first level decoding section 303, a upsampling section 304, a first frequency domain conversion section 305, a delay section 306, and a second conversion section 307 in the frequency domain, by the second layer encoding section 308 and the multiplexing section 309, and has a scalable configuration comprising two levels. In the first level, the CELP (code-excited linear prediction) speech coding method is applied, and in the second level, the speech coding method described in Embodiment 1 of the present invention is applied.

[0086] The downsampling section 301 performs downsampling processing on the input speech / audio signal to convert the sampling frequency of the speech / audio signal from frequency 1 to frequency 2 (where frequency 1> frequency 2), and outputs this signal to section 302 first level coding.

[0087] The first level encoding section 302 performs speech CELP encoding for the speech / audio signal after the down-sampling signal (input) from the down-sampling section 301, and outputs the encoded received first-level information to the first level decoding section 309 and the multiplexing section 303 . More specifically, the first level encoding section 302 encodes a voice message containing vocal section information and excitation information, by finding the LPC parameter for the vocal section information and for excitation information, performs encoding by finding an index that identifies which previously stored speech model should be used, that is, an index that identifies which excitation vector of the adaptive codebook and fixed codebook should be generated.

[0088] The first level decoding section 303 performs CELP speech decoding on the first level encoded information input from the first level encoding section 302, and outputs the obtained first level decoded signal to the upsample section 304.

[0089] The upsampling section 304 performs upsampling processing on a first-level decoded signal input from a first-level decoding section 303 to convert a sampling frequency of a first-level decoded signal from a frequency of 2 to a frequency of 1, and outputs this signal to the first section 305 conversion to the frequency domain.

[0090] The first frequency domain conversion section 305 performs an MDCT on a decoded first level signal after upsampling introduced from the upsampling section 304, and outputs the first level MDCT coefficient obtained as a parameter of the frequency domain to the second level encoding section 308. The actual conversion method used in the first frequency domain conversion section 305 is similar to the conversion method used in the frequency domain conversion section 101 of the speech encoding apparatus 100 according to Embodiment 1 of the present invention, and therefore, a description thereof is omitted here.

[0091] The delay section 306 outputs the delayed speech / audio signal to the second frequency domain conversion section 307 by issuing an input speech / audio signal after storing this input signal in an internal buffer for a predetermined time. The predefined delay time here is the time that takes into account the algorithm delay that occurs in the downsizing section 301, the first level encoding section 302, the first level decoding section 303, the upsizing section 304, the first frequency domain conversion section 305, and the second section 307 conversions to the frequency domain.

[0092] The second frequency domain conversion section 307 performs an MDCT on the delayed speech / audio signal input from the delay section 306, and outputs the second level MDCT coefficient obtained as a parameter of the frequency domain to the second level encoding section 308. The actual conversion method used in the second frequency domain conversion section 307 is similar to the conversion method used in the frequency domain conversion section 101 of the speech encoding apparatus 100 according to Embodiment 1 of the present invention, and therefore, a description thereof is omitted here.

[0093] The second level encoding section 308 performs second level encoding using the first level MDCT coefficient inputted from the first frequency domain conversion section 305 and the second level MDCT coefficient introduced from the second frequency domain conversion section 307, and outputs the obtained encoded information second level to multiplexing section 309. The main internal configuration and actual operation of the second layer encoding section 308 is described below.

[0094] The multiplexing section 309 multiplexes the first level encoded information input from the first level encoding section 302 and the second level encoded information input from the second level encoding section 308, and transmits the received bit stream to the speech decoding apparatus.

[0095] FIG. 7 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section 308. The second level encoding section 308 has a basic configuration similar to that of the speech encoding apparatus 100 according to Embodiment 1 (see FIG. 1), and therefore the same reference elements are assigned the same reference position, and their descriptions are omitted here.

[0096] The second level encoding section 308 differs from the speech encoding device 100 in that it is equipped with a MDCT differential coefficient calculation section 381 instead of the frequency domain transform section 101. The processing by the multiplexing section 106 is similar to the processing by the multiplexing section 106 of the speech encoding apparatus 100, and for description purposes, the name of the output from the multiplexing section 106 according to this embodiment is given as “second level encoded information”.

[0097] The range information, encoded shape information, and encoded gain information can also be input directly to the multiplexing section 309 and multiplexed with the first level encoded information without passing through the multiplexing section 106.

[0098] The MDCT differential coefficient calculation section 381 finds the remainder of the first level MDCT coefficient input from the first frequency domain conversion section 305 and the second level MDCT coefficient introduced from the second frequency domain conversion section 307, and outputs them to the range selection section 102 as a difference coefficient MDCT.

[0099] FIG. 8 is a block diagram illustrating a basic configuration of a speech decoding apparatus 400 according to Embodiment 2 of the present invention.

[0100] In this drawing, the speech decoding apparatus 400 is provided with a control section 401, a first level decoding section 402, a upsampling section 403, a frequency domain conversion section 404, a second level decoding section 405, a time domain conversion section 406, and a switch 407.

[0101] The control section 401 analyzes the configuration elements of the bitstream transmitted from the speech encoding apparatus 300, and according to these configurations of the bitstream adaptively outputs the corresponding encoded information to the first level decoding section 402 and the second level decoding section 405, and also outputs control information to switch 407. More specifically, if the bitstream contains first level encoded information and second level encoded information, control section 401 outputs encoded the first level information to the first level decoding section 402, and outputs the second level encoded information to the second level decoding section 405, whereas if the bitstream contains only the first level encoded information, the control section 401 outputs this first level encoded information to the first decoding section 402 level.

[0102] The first layer decoding section 402 performs CELP decoding on the first level encoded information inputted from the control section 401, and outputs the obtained first level decoded signal to the upsizing section 403 and the switch 407.

[0103] The upsampling section 403 performs upsampling processing on a decoded first level signal input from a first level decoding section 402 to convert the first sampling frequency of the decoded level signal from frequency 2 to frequency 1, and outputs this signal to conversion section 404 in the frequency domain.

[0104] The frequency domain transform section 404 performs an MDCT on the decoded first level signal after upsampling introduced from the upsample section 403, and outputs the decoded first level MDCT coefficient obtained as a frequency domain parameter to the second level decoding section 405. The actual conversion method used in the frequency domain conversion section 404 is similar to the conversion method used in the frequency domain conversion section 101 of the speech encoding apparatus 100 according to Embodiment 1, and therefore, a description thereof will be omitted here.

[0105] The second level decoding section 405 dequantizes the gain and dequantizes the shape using the encoded second level information input from the control section 401 and the decoded first level MDCT coefficient input from the frequency domain transform section 404 to obtain a decoded second MDCT coefficient level. The second level decoding section 405 sums together the obtained decoded second level MDCT coefficient and the decoded first level MDCT coefficient, and outputs the obtained summing result to the time domain converting section 406 as a total MDCT coefficient. The main internal configuration and the actual operation of the second level decoding section 405 are described below.

[0106] The time domain transform section 406 performs an IMDCT with respect to the total MDCT coefficient input from the second layer decoding section 405, and outputs the decoded second level signal obtained as a component of the time domain to the switch 407.

[0107] Based on the control information inputted from the control section 401, if the bitstream inputted to the speech decoding apparatus 400 contains encoded first level information and encoded second level information, the switch 407 outputs a decoded second level signal input from the conversion section 406 into time domain, as an output signal, whereas if the bitstream contains only encoded first-level information, the switch 407 outputs a decoded first-level signal inputted from ktsii 402 decoding the first level as the output signal.

[0108] FIG. 9 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section 405. The second level decoding section 405 has a basic configuration similar to that of the speech decoding apparatus 200 according to Embodiment 1 (see FIG. 3), and therefore, the same reference elements are assigned the same reference position, and their description is omitted here.

[0109] The second level decoding section 405 differs from the speech decoding apparatus 200 in that it is also equipped with a total MDCT coefficient calculation section 452. In addition, the processing is different, in particular, between the processing of the demultiplexing section 451 of the second layer decoding section 405 and the processing of the demultiplexing section 201 of the speech decoding apparatus 200, and various reference positions indicate this.

[0110] The demultiplexing section 451 demultiplexes the range information, the encoded form information and the encoded gain information from the second level encoded information from the control section 401, and outputs the obtained range information to the form dequantization section 202 and the predictive decoding execution / non-execution determining section 203, the obtained encoded information form per form dequantization section 202, and the received encoded gain information to gain dequantization section 204.

[0111] The total MDCT coefficient calculation section 452 sums together the first level decoded MDCT coefficient inputted from the frequency domain conversion section 404 and the second level decoded MDCT coefficient inputted from the gain dequantization section 204 and outputs the obtained summation result to the conversion section 406 into time domain as the total MDCT coefficient.

[0112] Thus, according to this embodiment, when a frequency component of an excellent range is made to be quantized in each frame, predictive encoding of a non-temporal parameter is performed adaptively in addition to applying scalable encoding, thereby allowing to reduce the amount of encoded information in speech encoding, to reduce encoding error speech / audio signal and deterioration in the quality of the decoded audio signal.

[0113] In this embodiment, a case has been described by way of example in which the second level encoding section 308 receives the difference component of the first level MDCT coefficient and the second level MDCT coefficient as the encoding target, but the present invention is not limited thereto, and the second level encoding section 308 may also take the difference component of the first level MDCT coefficient and the second level MDCT coefficient as an encoding target for a range of a predetermined frequency or lower, or may take the MDCT coefficient of the input signal directly as an encoding target for the range is higher than a predetermined frequency. That is, switching can be performed between using or not using a difference component according to a range.

[0114] In this embodiment, a case has been described by way of example in which a method for selecting a quantization target range in second level encoding is to select a region for which the energy of the differential component of the first level MDCT coefficient and the second level MDCT coefficient is the highest, but the present invention is not limited to this, and an area for which the energy of the first level MDCT is the highest can also be selected. For example, the energy of each subband of the first level MDCT coefficient can be calculated, after which the energies of each subband are summed together region-by-region, and the region for which the energy is the highest is selected as the quantization target range when encoding the second level. On the side of the decoding apparatus, the region for which the energy is the highest among the regions of the decoded first level MDCT coefficient obtained by decoding the first level is selected as the dequantization target range when decoding the second level. By this, the transmission rate in coding bits can be reduced, since the range information regarding the quantization range in second-level coding is not transmitted from the side of the encoding device.

[0115] In this embodiment, a case has been described by way of example where the second level encoding section 308 selects and performs quantization with respect to the quantization target range for the difference component of the first level MDCT coefficient and the second level MDCT coefficient, and the present invention is not limited thereto, and the second level coding section 308 may also predict the second level MDCT from the first level MDCT, and select and perform quantization with respect to the target quantum range tions for the differential component of the predicted MDCT coefficient and an actual second layer MDCT coefficient. This allows, allows to further improve the coding efficiency when using the correlation between the first level MDCT coefficient and the second level MDCT coefficient.

[0116] Embodiment 3

10 is a block diagram illustrating a basic configuration of a speech encoding apparatus 500 according to Embodiment 3 of the present invention. The speech encoding device 500 has a basic configuration similar to that of the speech encoding device 100 shown in FIG. 1, and therefore the same reference numbers are assigned to the identical configuration elements, and its description is omitted here.

[0117] The speech encoding device 500 differs from the speech encoding device 100 in that it is further provided with an interpolation value calculation section 504. Moreover, the processing is different, in particular, between the gain quantization section 505 of the speech encoding device 500 and the gain quantization section 105 of the speech encoding device 100, and excellent reference codes indicate this.

[0118] The interpolation value calculation section 504 has an internal buffer that stores range information indicating a quantization target range of a past frame. Using the quantization gain value of the quantization target range of the past frame read from the gain quantization section 505, the interpolation value calculation section 504 interpolates a gain value of a range that was not quantized in the last frame among the quantization target ranges of the current frame indicated by the range information inputted from section 102 range selection. The interpolation value calculation section 504 outputs the obtained gain interpolation value to the gain quantization section 505.

[0119] The gain quantization section 505 is different from the gain quantization section 105 of the speech encoding apparatus 100 using the gain interpolation value input from the interpolation value calculation section 504, in addition to the last frame quantization gain value stored in the internal buffer and the internal gain codebook by performing predictive coding.

[0120] The gain value interpolation method used by the interpolation value calculation section 504 is described in more detail below.

[0121] The interpolation value calculation section 504 has an internal buffer that stores range information m_max entered from the range selection section 102 in the last frame. Here is described by way of example a case in which an internal buffer is provided that stores m_max range information for the last three frames.

[0122] The interpolation value calculation section 504 first calculates a gain value of a range other than the range indicated by the range information m_max for the last three frames by performing linear interpolation. The interpolation value is calculated in accordance with Equation (19) for a gain value of a lower range than the range indicated by the m_max range information, and the interpolation value is calculated in accordance with Equation (19) for a gain value of a lower range than the range indicated by the m_max range information.

β ₀ Q ₀ + β _one Q _one + β ₂ Q ₂ + β ₃ G = 0 Equation 19 β ₀ 'Q ₀ + β _one 'Q _one + β ₂ 'Q ₂ + β ₃ 'G = 0 Equation 20

In Equation (19) and Equation (20), β _i indicates the interpolation coefficient, q _i indicates the gain of the quantization target range indicated by the range information m_max of the previous frame, and g indicates the gain interpolation value of the non-quantized range adjacent to the quantization range specified by the range information m_max of the last frame. Here, a lower value of i indicates a proportionally lower frequency range, and in Equation (19) g indicates the interpolation value of the gain of the adjacent range from the side of the higher range from the quantization target range indicated by the information of the m_max range of the previous frame, while in Equation (20 ) g indicates the gain interpolation value of the adjacent range on the lower side of the quantization target range indicated by the information of the m_max range of the past frame. For the interpolation coefficient β _i, it is assumed to use a value that has been found statistically in advance so as to satisfy Equation (19) and Equation (20). Here, a case is described in which different interpolation coefficients β _i are used in Equation (19) and Equation (20), but a similar set of prediction coefficients α _i can also be used in Equation (19) and Equation (20).

[0123] As shown in Equation (19) and Equation (20), it is possible to interpolate a gain of one range from a higher range side or a lower range side adjacent to a quantization target range indicated by a range information of a past frame m_max of a past frame, in a section 504 calculating the interpolation value. The interpolation value calculation section 504 sequentially interpolates the gain values of adjacent non-quantized ranges by repeating the operations in Equation (19) and Equation (20), using the results obtained from Equation (19) and Equation (20).

[0124] Thus, the interpolation value calculation section 504 interpolates the gain values of ranges other than the range indicated by the m_max range information of the past three frames from the quantization target ranges of the current frame indicated by the range information inputted from the range selection section 102 using the quantized gain values the last three frames read from gain quantization section 505.

[0125] The prediction encoding operation in gain quantization section 505 is described below.

[0126] The gain quantization section 505 quantizes by predicting the current frame gain value of the past frame, the gain interpolation value input from the interpolation value calculation section 504, and the internal gain codebook. More specifically, gain quantization section 505 searches for an internal gain codebook consisting of the number of GQ gain code vectors for each of the L subbands, and finds the gain code vector index for which the result of Equation (21) below is a minimum.

Уравнение 21

Equation 21

[0127] In Equation (21), GC ⁱ _j indicates the gain code vector constituting the gain codebook, i indicates the index of the gain code vector, and j indicates the index of the gain code vector element. Here, C ^t _j indicates the quantization gain value of t frames backward in time so that when t = 1, for example, C ^t _j indicates the quantization gain value of one frame backward in time. In addition, α is a fourth-order linear prediction coefficient stored in gain quantization section 505. The gain interpolation value calculated in accordance with Equation (19) and Equation (20) by the interpolation value calculation section 504 is used as a gain value of a range not selected as the quantization target range in the last three frames. The gain quantization section 505 processes the L subbands within the same region as an L-dimensional vector, and performs vector quantization.

[0128] The gain quantization section 505 outputs the gain code vector index G_min, for which the result of Equation (21) above is a minimum, to the multiplexing section 106 as encoded gain information. The gain quantization section 505 also updates the internal buffer in accordance with Equation (22) below using the encoded gain information, G_min and the quantization gain value C ^t _j obtained in the current frame.

(j=0,…, L-1)

(j = 0, ..., L-1) Equation 22

[0129] FIG. 11 is a block diagram illustrating a basic configuration of a speech decoding apparatus 600 according to Embodiment 3 of the present invention. The speech decoding apparatus 600 has a basic configuration similar to that of the speech decoding apparatus 200 shown in FIG. 3, and therefore, identical configuration items are denoted by the same reference numerals, and their description is omitted here.

[0130] The speech decoding apparatus 600 is different from the speech decoding apparatus 200 in that it is further provided with an interpolation value calculation section 603. Moreover, the processing is different, in particular, between the gain dequantization section 604 of the speech decoding apparatus 600 and the gain dequantization section 204 of the speech decoding apparatus 200, and various reference positions indicate this.

[0131] The interpolation value calculation section 603 has an internal buffer that stores range information indicating range information dequantized in the last frame. Using the gain value of a past-quantized range read from gain gain de-quantization section 604, interpolation value calculation section 603 interpolates a gain of a range that was not dequantized in the past frame among the quantization target ranges of the current frame indicated by the range information inputted from demultiplexing section 201 . The interpolation value calculation section 603 outputs the obtained gain interpolation value to the gain dequantization section 604.

[0132] The gain dequantization section 604 is different from the gain dequantization section 204 of the speech decoding apparatus 200 using the gain interpolation value input from the interpolation value calculation section 603, in addition to the stored dequantized gain value of the past frame, and the internal gain codebook when performing predictive coding.

[0133] The gain value interpolation method used by the interpolation value calculation section 603 is similar to the gain value interpolation method used by the interpolation value calculation section 504, and therefore, a detailed description thereof will be omitted here.

[0134] The prediction decoding procedure in gain dequantization section 604 is described below.

[0135] The gain dequantization section 604 dequantizes by predicting the gain value of the current frame using the stored gain value dequanted in the last frame, the interpolation gain value input from the interpolation value calculation section 603, and the internal gain codebook. More specifically, the gain dequantization section 604 obtains the gain value Gain_q ′ by performing dequantization of the gain in accordance with Equation (23) below.

Equation 23

[0136] In Equation (23), C ″ ^t _j indicates the gain value of t frames backward in time, so that when t = 1, for example, C ″ ^t _j indicates the gain value of one frame earlier. In addition, α is a 4th order linear prediction coefficient stored in gain dequantization section 604. Again, the interpolation value calculated by the interpolation value calculation section 603 is used as a gain value of a range not selected as a quantization target in the last three frames. Gain dequantization section 604 processes the L subbands within the same region as an L-dimensional vector, and dequantizes the vector.

[0137] Then, the gain dequantization section 604 calculates the decoded MDCT coefficient in accordance with Equation (24) below using the gain value obtained by dequantizing the current frame and the shape value input from the form dequantization section 202, and updates the internal buffer in accordance with the Equation (25) below. Here, the calculated decoded MDCT coefficient is denoted by X ' _k . Also, when dequantizing the MDCT coefficient, if k is present in B (j '') - B (j '' + 1) -1, the gain value Gain_q '(j) takes the value Gain_q' (j '').

Уравнение 24 X ” _k = Gain_q '(j) · Shape_q' (k)

Equation 24

(j = j ”, ..., j” + L-1) Equation 25

Thus, according to this embodiment, when quantizing the frequency domain parameter of the different quantization target range of each frame, the values of neighboring non-quantized ranges are sequentially interpolated from the quantized value in the last frame, and predictive quantization is performed using the interpolation value. Therefore, the coding accuracy of the speech coding can be further improved.

[0138] In this embodiment, a case has been described by way of example in which a fixed interpolation coefficient β, found in advance, is used in calculating the gain interpolation value, but the present invention is not limited thereto, and interpolation can also be performed after adjusting the previously found interpolation coefficient β . For example, the prediction coefficient may be adjusted according to the distribution of the gain of the range quantized in each frame. More specifically, it is possible to improve the coding accuracy of speech coding by adjusting so that the prediction coefficient is reduced and the gain weight of the current frame is increased when the change in gain quantized in each frame is large.

[0139] In this embodiment, a case has been described by way of example in which a consecutive plurality of ranges (one region) containing a range quantized in each frame is targeted, but the present invention is not limited thereto, and a plurality of regions can also be made a quantization target . In this case, it is possible to improve the encoding accuracy in speech encoding by using a method by which linear prediction of the final values of the respective regions is performed for the range between the selected regions in addition to the interpolation method according to Equation (19) and Equation (20).

[0140] Embodiment 4

12 is a block diagram illustrating a basic configuration of a speech encoding apparatus 700 according to Embodiment 4 of the present invention. The speech encoding device 700 has a basic configuration similar to that of the speech encoding device 100 shown in FIG. 1, and therefore the same reference elements are assigned the same configuration elements, and their description is omitted here.

[0141] The speech encoding device 700 is different from the speech encoding device 100 in that it is further provided with a prediction coefficient determining section 704. Furthermore, the processing is different, in particular, between the gain quantization section 705 of the speech encoding device 700 and the gain quantization section 105 of the speech encoding device 100 and various reference positions indicate this.

[0142] The prediction coefficient determination section 704 has an internal buffer that stores range information indicating a quantization target range of the past frame, determines a prediction coefficient to be used when quantizing the gain quantization section 705 based on the range information of the past frame, and outputs the determined prediction coefficient to gain quantization section 705.

[0143] The gain quantization section 705 is different from the gain quantization section 105 of the speech encoding apparatus 100 using the prediction coefficient inputted from the prediction coefficient determination section 704 instead of the prediction coefficient determined in advance when performing predictive encoding.

[0144] The prediction coefficient determination operation in the prediction coefficient determination section 704 is described below.

[0145] The prediction coefficient determination section 704 has an internal buffer that stores range information m_max entered from the range selection section 102 in the last frame. Here, by way of example, a case is described in which an internal buffer is provided that stores m_max range information for the past three frames.

[0146] Using the m_max range information stored in the internal buffer and the m_max range information inputted from the range selection section 102 in the current frame, the prediction coefficient determining section 704 finds the number of subbands common to the quantization target range of the current frame and the quantization target range of the past frame . Prediction coefficient determination section 704 determines the prediction coefficients as set A, and outputs it to gain quantization section 705 if the number of common subbands is greater than or equal to a predetermined value, or determines prediction coefficients as set B and outputs them to gain quantization section 705 if the number total subbands are less than a predetermined value. Here, the set A of prediction coefficients is a set of parameters that focuses more on the value of the past frame and makes the weight of the gain value of the past frame larger than in the case of set B of prediction coefficients. For example, in the case of 4th order prediction coefficients, it is possible that set A was defined as (αa0 = 0.60, αa1 = 0.25, αa2 = 0.10, αa3 = 0.05), and that set B was defined as (αb0 = 0.80, αb1 = 0.10, αb2 = 0.05, αb3 = 0.05).

[0147] Then, the prediction coefficient determination section 704 updates the internal buffer using the range information m_max input from the range selection section 102 in the current frame.

[0148] The prediction encoding operation in gain quantization section 705 is described below.

[0149] The gain quantization section 705 has an internal buffer that stores the quantization gain value obtained in the last frame. The gain quantization section 705 quantizes by predicting the gain value of the current frame using the prediction coefficient inputted from the prediction coefficient determination section 704 and the quantization gain value of the past frame C ^t _j stored in the internal buffer. More specifically, gain quantization section 705 searches for an internal gain codebook consisting of the number of GQ gain code vectors for each of the L subbands, and finds the gain code vector index for which the result of Equation (26) below is a minimum if the prediction coefficient is set A , or finds the index of the gain code vector for which the result of Equation (27) below is the minimum if the prediction coefficient is set B.

Уравнение 26

Equation 26

Уравнение 27

Equation 27

[0150] In Equation (26) and Equation (27), GC ⁱ _j indicates the gain code vector constituting the gain codebook, i indicates the index of the gain code vector, and j indicates the index of the gain code vector element. Here, C ^t _j indicates the gain value of t frames backward in time, so that when t = 1, for example, C ^t _j indicates the gain value of one frame backward in time. In addition, a 4th-order linear prediction coefficient is stored in gain quantization section 705. Gain quantization section 705 processes the L subbands within the same region as an L-dimensional vector, and performs vector quantization. If there is no sub-band gain value corresponding to the last frame in the internal buffer, the gain quantization section 705 replaces the gain of the nearest sub-band in the frequency in the internal buffer in Equation (26) or Equation (27) above.

[0151] FIG. 13 is a block diagram illustrating a basic configuration of a speech decoding apparatus 800 according to Embodiment 4 of the present invention. The speech decoding apparatus 800 has a basic configuration similar to that of the speech decoding apparatus 200 shown in FIG. 3, and therefore the same reference numbers are assigned to the same configuration elements, and their description is omitted here.

[0152] The speech decoding apparatus 800 is different from the speech decoding apparatus 200 in that it is further provided with a prediction coefficient determining section 803. In addition, the processing is different, in particular, between the gain dequantization section 804 of the speech decoding apparatus 800 and the gain dequantization section 204 of the speech decoding apparatus 200, and various reference positions indicate this.

[0153] The prediction coefficient determination section 803 has an internal buffer that stores range information input from the demultiplexing section 201 in the last frame, determines the prediction coefficient to be used when quantizing the gain dequantization section 804 based on the information of the range of the last frame, and outputs a specific prediction coefficient per gain dequantization section 804.

[0154] The gain dequantization section 804 is different from the gain dequantization section 204 of the speech decoding apparatus 200 using the prediction coefficient inputted from the prediction coefficient determination section 803 instead of the prediction coefficient determined in advance when performing predictive decoding.

[0155] The prediction coefficient determination method used by the prediction coefficient determination section 803 is similar to the prediction coefficient determination method used by the prediction coefficient determination section 704 of the speech encoding apparatus 700, and therefore, a detailed description of the operation of the prediction coefficient determination section 803 is omitted here.

[0156] The prediction decoding operation in gain dequantization section 804 is described below.

[0157] The gain dequantization section 804 has an internal buffer that stores the gain value obtained in the last frame. The gain dequantization section 804 dequantizes by predicting the gain value of the current frame using the prediction coefficient inputted from the prediction coefficient determination section 803 and the past frame gain value stored in the internal buffer. More specifically, the gain dequantization section 804 has the same kind of internal gain codebook as the gain quantization section 705 of the speech encoding device 700, and obtains the gain value Gain_q 'by performing gain dequantization in accordance with Equation (28) below, if the prediction coefficient introduced from the prediction coefficient determination section 803, there is a set A, or according to Equation (29) below, if the prediction coefficient is a set B.

Equation 28

Equation 29

[0158] In Equation (28) and Equation (29), C ″ ^t _j indicates the gain value of t frames backward in time, so that when t = 1, for example, C ″ ^t _j indicates the gain value of one frame before. In addition, αa _i and αb _i indicate a set A and a set B of prediction coefficients introduced from the prediction coefficient determination section 803. Gain dequantization section 804 processes the L subbands within the same region as an L-dimensional vector, and dequantizes the vector.

[0159] Thus, according to this embodiment, when quantizing a frequency domain parameter of an excellent quantization target range of each frame, predictive coding is performed by selecting from a plurality of sets of prediction coefficients a set of prediction coefficients that makes the weight of a gain value of a past frame proportionally greater than the greater the number of subbands common to the quantization target range of the last frame and the quantization target range of the current frame. Therefore, the coding accuracy of the speech coding can be further improved.

[0160] In this embodiment, a case has been described by way of example in which two kinds of sets of prediction coefficients are provided in advance, and the prediction coefficient used in predictive coding is switched according to the number of subbands common to the quantization target range of the past frame and the target quantization range of the current frame , but the present invention is not limited to this, and three or more kinds of sets of prediction coefficients can also be provided in advance.

[0161] In this embodiment, a case has been described by way of example in which if the quantization target range in the current frame has not been quantized in the last frame, the closest range value in the last frame is replaced, but the present invention is not limited to this, and if the value of the target range quantization in the current frame was not quantized in the last frame, predictive coding can also be performed by taking the corresponding prediction coefficient of the last frame equal to zero, summing the coefficient the prediction rate of this frame with the prediction coefficient of the current frame, calculating a new set of prediction coefficients, and using these prediction coefficients. Thus, the effect of predictive coding can be switched more flexibly, and the coding accuracy of the speech coding can be further improved.

[0162] Option exercise 5

14 is a block diagram illustrating a basic configuration of a speech encoding apparatus 1000 according to Embodiment 5 of the present invention. The speech encoding device 1000 has a basic configuration similar to that of the speech encoding device 300 shown in FIG. 6, and therefore, the same reference elements are assigned the same configuration elements, and their description is omitted here.

[0163] The speech encoding apparatus 1000 is different from the speech encoding apparatus 300 in that it is further provided with a range enhancement (gain) encoding section 1007. Furthermore, the processing is different, in particular, between the second level encoding section 1008 and the multiplexing section 1009 of the speech encoding apparatus 1000 and the second level encoding section 308 and the multiplexing section 309 of the speech encoding apparatus 300, and various reference codes indicate this.

[0164] The range enhancement coding section 1007 performs range enhancement coding using a first level MDCT coefficient inputted from the first frequency domain conversion section 305 and an MDCT input coefficient introduced from the second frequency domain conversion section 307 and outputs the obtained encoded enhancement information range to multiplexing section 1009.

[0165] The multiplexing section 1009 differs from the multiplexing section 309 only by multiplexing the encoded range improvement information in addition to the encoded information of the first layer and the encoded information of the second layer.

[0166] FIG. 15 is a block diagram illustrating a basic configuration of an interior of a range enhancement encoding section 1007.

[0167] In FIG. 15, the range improvement coding section 1007 is provided with a higher range spectrum estimation section 1071 and a correction scale factor encoding section 1072.

[0168] The higher range spectrum estimation section 1071 estimates the spectrum of the higher range of the ranges of the FL-FH signals using the spectrum of the lower range of the ranges of 0-FL signals of the input MDCT coefficient introduced from the second frequency domain transform section 307 to obtain an estimated spectrum. The method of obtaining the estimated spectrum is to find the estimated spectrum in such a way that the degree of similarity with the spectrum of the higher range becomes maximum by converting the spectrum of the lower range based on this spectrum of the lower range. The higher spectrum range estimation section 1071 encodes information regarding this estimated spectrum (estimation information), outputs the obtained encoding parameter, and also provides the estimated spectrum itself directly to the correction scale factor encoding section 1072.

[0169] In the following description, the estimated spectrum derived from the higher spectrum range estimation section 1071 is called the first spectrum, and the first level MDCT coefficient (higher range spectrum) derived from the first frequency domain conversion section 305 is called the second spectrum.

[0170] The above types of spectra and corresponding signal ranges can be summarized as follows.

[0171]

Narrowband spectrum (lower range spectrum) 0-fl Broadband spectrum 0-fh The first spectrum (estimated spectrum) Fl-fh The second spectrum (spectrum of a higher range) Fl-fh

The correction scale factor encoding section 1072 corrects the scale factor of the first spectrum so that the scale factor of the first spectrum approaches the scale factor of the second spectrum, and encodes and outputs information regarding this correction scale factor.

[0172] The encoded range enhancement information inputted from the range enhancement coding section 1007 to the multiplexing section 1009 includes an encoding parameter of the estimation information inputted from the higher spectrum spectrum estimation section 1071, and a correction scale factor encoding parameter inputted from the encoding section 1072 correction scale factor.

[0173] FIG. 16 is a block diagram illustrating a basic configuration of an interior of a correction scale factor encoding section 1072.

[0174] The correction scale factor encoding section 1072 is provided with scale factor calculation sections 1721 and 1722, a correction coefficient coefficient codebook 1723, a multiplier 1724, a subtractor 1725, a determination section 1726, a weighting error calculation section 1727, and a search section 1728. These sections perform the following operations.

[0175] The scale factor calculation section 1721 divides the second-spectrum input signal ranges FL-FH into a plurality of subbands, finds the size of the spectrum included in each subband, and outputs it to a subtractor 1725. More specifically, the subband division is performed associated with the critical range, and the division is performed at equal intervals of the Bark scale. In addition, the scale factor calculation section 1721 finds the average amplitude of the spectra included in these subbands and takes it as the second scale factor SF2 (k) {0≤k <NB}, where NB represents the number of subbands. The maximum amplitude value or the like can be used instead of the average amplitude.

[0176] The scale factor calculation section 1722 divides the input ranges of the first-spectrum FL-FH signals into a plurality of subbands, calculates the first subband subband factor SF1 (k) {0≤k <NB} and outputs it to a multiplier 1724. As with the calculation section 1721 a scale factor, a maximum amplitude value or the like can be used instead of the average amplitude.

[0177] In the subsequent processing, the parameters in a plurality of subbands are combined into a single vector value. For example, the number of NB scale factors is represented as a single vector. The description below is given for the case in which each processing operation is performed for each of these vectors — that is, the case in which quantization of the vector is performed — as an example.

[0178] The corrective scale factor codebook 1723 stores a plurality of corrective scale factors — candidates, and subsequently provides one of the stored corrective scale factors — candidates for a multiplier 1724 in accordance with a directive from a search section 1728. A plurality of correction scale factors — candidates stored in the codebook 1723 of the correction scale factor — is represented by a vector.

[0179] A multiplier 1724 multiplies the first scale factor derived from the scale factor calculation section 1722 by a correction scale factor, a candidate derived from the correction scale factor codebook 1723, and outputs the result of the multiplication to a subtractor 1725.

[0180] The subtractor 1725 subtracts the output of the multiplier 1724, that is, the product of the first scale factor and the correction scale factor, from the second scale factor derived from the scale factor calculation section 1721, and outputs an error signal thus obtained to the error calculation section 1727 weighing and determination sections 1726.

[0181] The determining section 1726 determines the weighting vector to be supplied to the weighing error calculation section 1727 based on the sign of the error signal issued from the subtractor 1725. More specifically, the error signal d (k) issued from the subtractor 1725 is represented Equation (30) below.

d (k) = SF2 (k) -v _i (k) SF1 (k) (0≤k <NB} Equation 30

[0182] Here, v _i (k) represents the i-th correction scale factor — the candidate. The determination section 1726 checks the sign of d (k), selects w _pos as the weight if d (k) is positive, or selects w _neg as the weight if d (k) is negative, and outputs the weighting vector w (k) consisting of these, to a weighting error calculation section 1727. These weights have relative aspect ratios shown in Equation (31) below.

0 <w _pos <w _mg Equation 31

[0183] For example, if the number of subbands is NB = 4 and the signs d (k) are {+, -, -, +}, the weighting vector w (k) derived to the weighting error calculation section 1727 is represented by w (k) = {w _pos , w _neg , w _neg , w _pos }.

[0184] The weighting error calculation section 1727 first calculates the square of the error signal output from the subtractor 1725, and then multiplies the weighting vector w (k) output from the determination section 1726 by the square of the error signal to calculate the weighted error E of the square, and outputs the result of this calculation on the search section 1728. Here, the weighted error E of the square is represented as shown in Equation (32) below.

Equation 32

[0185] The search section 1728 controls the correction scale factor codebook 1723 and sequentially outputs the stored correction scale factors as candidates, and through closed-loop processing finds the correction scale factor as a candidate for which the weighted square error E derived from the weighting error calculation section 1727 is minimal. Search section 1728 displays the index iopt of the found correction scale factor — a candidate as an encoding parameter.

[0186] When the weight used in calculating the weighted error E of the square is set according to the sign of the error signal, and the form of the relations shown in Equation (30) refers to this weight, as described above, the following kind of effect is obtained, namely, the case in wherein the error signal d (k) is positive, is a case in which a decoded value generated on the decoding side (in terms of the encoding side, a value obtained by multiplying the first scale factor by a correction scale factor), less than the second scale factor, which is the target value. Also, the case in which the error signal d (k) is negative is the case in which the decoded value generated on the decoding side is larger than the second scale factor, which is the target value. Therefore, by setting the weight when the error signal d (k) is positive, so as to be less than the weight when the error signal d (k) is negative, when the square error values are of the same order, the correction factor is the candidate that generates a decoded value smaller than the second scale factor becomes preferred for selection.

[0187] The following kind of enhancement effect is obtained by processing the range enhancement coding section 1007. For example, when a spectrum of a higher range is estimated using a spectrum of a lower range, as in this embodiment, a lower bit rate can usually be achieved. However, while a lower bit rate can be achieved, the accuracy of the estimated spectrum — that is, the similarity between the estimated spectrum and the spectrum of a higher range — cannot be said to be sufficiently high, as described above. In this case, if the decoded scale factor value becomes larger than the target value, and the post-quantization scale factor works in the direction of amplification of the estimated spectrum, the low accuracy of the estimated spectrum tends to be visible to the human ear as a deterioration. Conversely, when the decoded scale factor value becomes smaller than the target value, and the post-quantization scale factor works in the direction of decreasing this estimated spectrum, the low accuracy of the estimated spectrum ceases to be noticeable, and an effect of improving the quality of the audio decoded signal is obtained. This trend has also been confirmed in computer modeling.

[0188] FIG. 17 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section 1008. The second level encoding section 1008 has a similar basic configuration to that of the second level encoding section 308 shown in FIG. 7, and therefore the same reference elements are assigned the same reference numerals, and their description is omitted here. The processing is different, in particular, between the MDCT difference coefficient calculation section 1081 of the second level encoding section 1008 and the MDCT difference coefficient calculation section 381 of the second level encoding section 308, and various reference numbers indicate this.

[0189] The MDCT differential coefficient calculating section 1081 calculates the differential MDCT, which should be the quantization target in the second level coding section, from the input MDCT input coefficient and the first level improvement MDCT coefficient. The MDCT difference coefficient calculation section 1081 is different from the MDCT difference coefficient calculation section 381 according to Embodiment 2 by taking the remainder of the input MDCT coefficient and the first level difference coefficient of improvement as the MDCT difference coefficient for a range not improved by the range improvement coding section 1007 and accepting the input MDCT coefficient directly, and not the remainder, as the MDCT difference coefficient for the range improved by the range improvement coding section 1007.

[0190] FIG. 18 is a block diagram illustrating a basic configuration of a speech decoding apparatus 1010 according to Embodiment 5 of the present invention. The speech decoding apparatus 1010 has a basic configuration similar to that of the speech decoding apparatus 400 shown in FIG. 8, and therefore the same reference elements are assigned the same configuration elements, and their description is omitted here.

[0191] The speech decoding apparatus 1010 differs from the speech decoding apparatus 400 in that it is further provided with a range enhancement decoding section 1012 and a time domain converting section 1013. Furthermore, the processing is different, in particular, between the control section 1011, the second level decoding section 1015, and the switch 1017 from the speech decoding apparatus 1010 and the control section 401, the second level decoding section 405, and the switch 407 from the speech decoding apparatus 400, and various reference numbers indicate this.

[0192] the Control section 1011 analyzes the configuration elements of the bit stream transmitted from the speech encoding device 1000, and according to these configuration elements of the bit stream adaptively outputs the corresponding encoded information to the first level decoding section 402, the range enhancement decoding section 1012, and the second level decoding section 1015 , and also outputs control information to the switch 1017. More specifically, if the bitstream contains first level encoded information, the encoded information is improved. tions range, and second level encoded information, control section 1011 outputs the encoded information to first layer decoding section 402 of the first level, outputs encoded information to improve band decoding section 1012 improve range and outputs the encoded information to second layer decoding section 1015 of the second level. If the bitstream contains only the encoded first level information and the encoded range enhancement information, the control section 1011 outputs the encoded first level information to the first layer decoding section 402, and outputs the encoded range enhancement information to the range enhancement decoding section 1012. If the bitstream contains only first level encoded information, the control section 1011 outputs this first level encoded information to the first level decoding section 402. In addition, the control section 1011 outputs control information that controls the switch 1017 to the switch 1017.

[0193] The range enhancement decoding section 1012 performs range enhancement processing using the encoded range enhancement information from the control section 1011, and the decoded first level MDCT coefficient introduced from the frequency domain transform section 404 to obtain the first level MDCT coefficient. Then, the range enhancement decoding section 1012 outputs the obtained first level improvement coefficient MDCT to the time domain converting section 1013 and the second level decoding section 1015. The main internal configuration and the actual operation of the range enhancement decoding section 1012 are described below.

[0194] The time domain converting section 1013 performs an IMDCT with respect to the first level improvement MDCT coefficient inputted from the range improvement decoding section 1012, and outputs the decoded first level improvement signal obtained as a time domain component to the switch 1017.

[0195] The second level decoding section 1015 dequantizes gain and dequantizes the shape using the encoded second level information entered from the control section 1011 and the first level improvement MDCT introduced from the range improvement decoding section 1012 to obtain a decoded second level MDCT . The second level decoding section 1015 sums together the obtained decoded second level MDCT coefficient and the decoded first level MDCT coefficient, and outputs the obtained summing result to the time domain converting section 406 as the total MDCT coefficient. The main internal configuration and the actual operation of the second level decoding section 1015 are described below.

[0196] Based on the control information inputted from the control section 1011, if the bitstream supplied to the speech decoding apparatus 1010 contains encoded first level information, encoded range improvement information, and encoded second level information, the switch 1017 outputs a decoded second level signal, introduced from the time domain conversion section 406 as an output signal. If the bitstream contains only encoded first level information and encoded range improvement information, the switch 1017 outputs a decoded first level enhancement signal inputted from the time domain conversion section 1013 as an output signal. If the bitstream contains only encoded first-level information, the switch 1017 outputs the decoded first-level signal inputted from the first-level decoding section 402 as an output signal.

[0197] FIG. 19 is a block diagram illustrating a basic configuration of an interior of a range enhancement decoding section 1012. The range enhancement decoding section 1012 comprises a higher range spectrum decoding section 1121, a correction scale factor decoding section 1122, a multiplier 1123, and a communication section 1124.

[0198] The higher range spectrum decoding section 1121 decodes the estimated spectrum (accurate spectrum) of the FL-FH bands using the encoding parameter of the estimation information and the first spectrum included in the encoded range enhancement information inputted from the control section 1011. The resulting estimated spectrum is fed to a multiplier 1123.

[0199] The correction scale factor decoding section 1122 decodes the correction scale factor using the correction scale factor encoding parameter included in the encoded range enhancement information inputted from the control section 1011. More specifically, the correction scale factor decoding section 1122 accesses the internal correction scale factor codebook (not shown) and outputs the corresponding correction scale factor to a multiplier 1123.

[0200] A multiplier 1123 multiplies the estimated spectrum derived from the higher range spectrum decoding section 1121 by a correction scale factor derived from the correction scale factor decoding section 1122, and outputs the multiplication result to the communication section 1124.

[0201] The communication section 1124 couples the first spectrum and the estimated spectrum extracted from the frequency domain multiplier 1123 to generate a broadband decoded spectrum of the 0-FH signal ranges, and outputs it to the time domain conversion section 1013 as an MDCT coefficient of improvement of the first level.

[0202] Through the range enhancement decoding section 1012, when an input signal is converted to a frequency domain coefficient and the scale factor is quantized when encoding the upper level frequency domain, a scale factor is quantized using a weighted distortion scale such that a quantization candidate for which the scale factor becomes small, becoming preferred for choice. Thus, a quantization candidate by which the scaling factor after quantization is smaller than the scaling coefficient before quantization is more likely to be selected. Thus, deterioration in perceptual subjective quality can be suppressed even when the number of bits allocated to quantize the scale factor is insufficient.

[0203] FIG. 20 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section 1015. The second level decoding section 1015 has a basic configuration similar to that of the second level decoding section 405 shown in FIG. 9, and therefore the same reference elements are assigned to the same configuration elements, and their description is omitted here.

[0204] The processing differs, in particular, between the second level decoding section MDCT coefficient calculation section 1151 and the second level decoding section calculation section 452 and the second level decoding section MDCT coefficient calculation section 452, and various reference positions indicate this.

[0205] The total MDCT coefficient calculation section 1151 has a first level improvement MDCT coefficient as input from a range improvement decoding section 1012, and a second level decoded MDCT coefficient as input from a gain dequantization section 204. Section 1151 calculates the total coefficient MDCT summarizes the decoded coefficient MDCT of the first level and the decoded coefficient MDCT of the second level, and outputs the summed coefficient MDCT. For a range with an improved range, the first level improvement MDCT coefficient value is added up to zero in the total MDCT coefficient calculation section 1151. That is, for a range with an improved range, the value of the MDCT coefficient of improvement of the second level is taken as the value of the total coefficient MDCT.

[0206] Thus, according to this embodiment, when a frequency component of a different range is made a quantization target in each frame, predictive encoding of a non-temporal parameter is performed adaptively in addition to applying scalable encoding using range enhancement technology. Therefore, the amount of encoded information in speech encoding can be reduced, and the encoding error of the speech / audio signal and the decoded signal and quality degradation can also be reduced.

[0207] Furthermore, since the remainder is not calculated for the range component improved by the range enhancement coding method, the energy of the target quantization component does not increase in the upper level, and the quantization efficiency can be improved.

[0208] In this embodiment, an example has been described of a case in which a method is used by which encoded range enhancement information is computed in an encoding device using the correlation between a low range component, a decoded first layer decoding section and a higher range component of an input signal, but the present invention is not limited to this, and can also be similarly applied to a configuration that uses the method by which encoded range improvement information is not calculated, and higher range pseudo-generation is performed by a noise component, as with AMR-WB (Adaptive MultiRate - Wideband). Alternatively, the range selection method of the present invention can be similarly applied to the range enhancement encoding method described in this example, or a scalable encoding / decoding method that does not use the higher range component generation method also used in AMR-WB.

[0209] Option exercise 6

21 is a block diagram illustrating a basic configuration of a speech encoding apparatus 1100 according to Embodiment 6 of the present invention.

[0210] In this drawing, the speech encoding device 1100 is provided with a downsizing section 301, a first level encoding section 302, a first level decoding section 303, a upsampling section 304, a first frequency domain conversion section 305, a delay section 306, and a second conversion section 307 in frequency domain, second layer encoding section 1108, and multiplexing section 309, and has a scalable configuration comprising two layers. In the first level, the speech CELP encoding method is used, and in the second level, the speech encoding method described in Embodiment 1 of the present invention is applied.

[0211] With the exception of the second level encoding section 1108, the configuration elements in the speech encoding device 1100 shown in FIG. 21 are identical to the configuration elements of the speech encoding device 300 shown in FIG. 6, and therefore, the same reference numbers are assigned to the identical configuration elements, and their description is omitted here.

[0212] FIG. 22 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section 1108. The second level encoding section 1108 mainly comprises a MDCT differential coefficient calculation section 381, a range selection section 1802, a shape quantization section 103, a predictive encoding execution / non-execution determining section 104, a gain quantization section 1805, and a multiplexing section 106. With the exception of the range selection section 1802 and the gain quantization section 1805, the configuration elements in the second level encoding section 1108 are identical to the configuration elements of the second level encoding section 308 shown in FIG. 7, and therefore the same reference elements are assigned the same reference position, and their description here falls.

[0213] The band selection section 1802 first divides the MDCT coefficient X _k into a plurality of subbands. Here, the description refers to a case in which the MDCT coefficient X _k is divided evenly into J subbands (where J is a positive integer), as an example. Then, the range selection section 1802 selects L subbands (where L is a natural number) from among J subbands, and obtains M kinds of regions (where M is a natural number).

[0214] FIG. 23 is a drawing showing an example of a configuration of regions obtained by a range selection section 1802.

[0215] In this drawing, the number of subbands is 17 (J = 17), the number of kinds of regions is eight (M = 8), and each region consists of two groups of subbands (the number of ranges that make up these two groups of subbands is three and two, respectively ) Of these two groups of subbands, a group of subbands containing two ranges located on the higher side is fixed across all frames, subband indices, for example, are 15 and 16. For example, region 4 consists of subbands 6-8, 15 and 16.

[0216] Then, the range selection section 1802 calculates the average energy E (m) of each of the M kinds of regions in accordance with Equation (33) below.

Equation 33

[0217] In this equation, j 'indicates the index of each of the J subbands, and m indicates the index of each of the M kinds of regions. Region (m) means a collection of indices L of the subbands making up the region m, and B (j ') indicates the minimum value among the indices of the set of MDCT coefficients making up the subband j'. W (j) indicates the bandwidth of the subband j ′, and in the following description, the case in which the bandwidths of each of the J subbands are equal, that is, the case in which W (j ′) is constant, is described as an example.

[0218] Then, when a region is selected for which the average energy E (m) is a maximum — for example, a region m_max, the band selection section 1802 selects a band consisting of j'∈Region (m_max) subbands as a quantization target range and outputs an index m_max indicating this area as range information to the shape quantization section 103, the predictive coding fulfillment / failure determination section 104, and the multiplexing section 106. The range selection section 1802 also outputs a difference coefficient MDCT X _k to the shape quantizing section 103.

[0219] The gain quantization section 1805 has an internal buffer that stores the quantization gain value obtained in the last frame. If the determination result inputted from the predictive coding fulfillment / failure determination section 104 indicates that predictive coding is to be performed, the gain quantization section 1805 quantizes by predicting the gain value of the current frame using the past frame quantization gain value C ^t _j stored in the internal buffer. More specifically, gain quantization section 1805 searches for an internal gain codebook consisting of the number of GQ gain code vectors for each of the L subbands, and finds the gain code vector index for which the result of Equation (34) below is a minimum.

Уравнение 34

Equation 34

[0220] In this equation, GC ⁱ _k indicates the gain code vector constituting the gain codebook, i indicates the index of the gain code vector, and k indicates the index of the gain code vector element. For example, if the number of subbands that make up the region is five (L = 5), k has a value from 0 to 4. Here, the gain values of the subbands of the selected region are connected so that the subband indices are in ascending order, the subsequent gain values are processed as one L- dimensional vector of the gain code, and vector quantization is performed. Therefore, to give a description using FIG. 23, in the case of region 4, the gain values of the subband indices 6, 7, 8, 15, and 16 are associated and processed as a 5-dimensional gain code vector. In addition, C ^t _{j ′} indicates the gain value of t frames backward in time, so that when t = 1, for example, C ^t _{j ′} indicates the gain value of one frame backward in time, and α is a fourth-order linear prediction coefficient, stored in gain quantization section 1805.

[0221] The gain quantization section 1805 provides the gain code vector index G_min, for which the result of Equation (34) above is a minimum, to the multiplexing section 106 as encoded gain information. If there is no gain value of the subband corresponding to the last frame in the internal buffer, the gain quantization section 1805 replaces the gain value of the nearest subband in the frequency in the internal buffer in Equation (34) above.

[0222] On the other hand, if the determination result inputted from the predictive coding execution / non-execution determining section 104 indicates that the predictive coding should not be performed, the gain quantization section 1805 directly quantizes the ideal gain value Gain_i (j ′) input from the section 103 form quantization, in accordance with Equation (35) below. Here, the gain quantization section 1805 processes the ideal gain value as an L-dimensional vector, and quantizes the vector.

Equation 35

[0223] Here, the codebook index that makes the result of Equation (35) above the minimum is denoted by G_min.

[0224] The gain quantization section 1805 outputs G_min to the multiplexing section 106 as encoded gain information. Gain quantization section 1805 also updates the internal buffer in accordance with Equation (36) below using the encoded gain information G_min and quantization gain value C ^t _{j ′} obtained in the current frame. That is, in Equation (36), the value C ^l _{j 'is} updated by the index j of the element GC ^G - ^min _j by the gain code vector and j' satisfies the condition j'∈Region (m_max), respectively, associated in ascending order.

Equation 36

[0225] FIG. 24 is a block diagram illustrating a basic configuration of a speech decoding apparatus 1200 according to this embodiment.

[0226] In this drawing, the speech decoding apparatus 1200 is provided with a control section 401, a first level decoding section 402, a upsampling section 403, a frequency domain conversion section 404, a second level decoding section 1205, a time domain conversion section 406, and a switch 407.

[0227] With the exception of the second level decoding section 1205, the configuration elements in the speech decoding apparatus 1200 shown in FIG. 24 are identical to the configuration elements of the speech decoding apparatus 400 shown in FIG. 8 and therefore the same reference numbers are assigned to the same configuration elements, and their description is omitted here.

[0228] FIG. 25 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section 1205. The second level decoding section 1205 mainly comprises a demultiplexing section 451, a shape dequantization section 202, a predictive decoding execution / non-execution determination section 203, a gain dequantization section 2504, and a total MDCT coefficient calculation section 452. With the exception of gain dequantization section 2504, the configuration elements in the second level decoding section 1205 are identical to the configuration elements of the second level decoding section 405 shown in FIG. 9, and therefore the same reference elements are assigned the same reference numerals, and their description is omitted here.

Region(m_max) соответственно ассоциированном в порядке возрастания.[0229] The gain dequantization section 2504 has an internal buffer that stores the gain value obtained in the last frame. If the determination result inputted from the predictive decoding run / fail determination section 203 indicates that predictive decoding should be performed, the gain dequantization section 2504 dequantizes by predicting the gain value of the current frame using the gain value of the past frame stored in the internal buffer. More specifically, the gain dequantization section 2504 has the same kind of internal gain codebook (GC ^G - ^min _k , where k indicates the index of the element) as the gain quantization section 105 of the speech encoding apparatus 100, and obtains the gain value Gain_q ′ by performing dequantization gain in accordance with Equation (37) below. Here, C ″ ^t _{j ′} indicates the gain value of t frames backward in time, so that when t = 1, for example, C ″ ^t _{j ′} indicates the gain value of one frame backward in time. In addition, α is a fourth order linear prediction coefficient stored in gain dequantization section 2504. Gain dequantization section 2504 processes the L subbands within the same region as an L-dimensional vector, and dequantizes the vector. That is, in Equation (37), the value Gain_q '(j') is calculated with the index k of the element GC ^G - ^min _k the gain code vector and j 'satisfies the condition j'

Region (m_max), respectively, associated in ascending order.

Equation 37

[0230] If there is no subband gain value corresponding to the last frame in the internal buffer, the gain dequantization section 2504 replaces the gain value of the nearest subband in the frequency in the internal buffer in Equation (37) above.

[0231] On the other hand, if the determination result inputted from the predictive decoding run / fail determination section 203 indicates that the predictive decoding should not be performed, the gain dequantization section 2504 quantizes the gain in accordance with Equation (38) below using the above codebook gain. Here, the gain value is processed as an L-dimensional vector, and the vector is quantized. That is, when predictive decoding is not performed, the gain dequantization section 2504 takes the ^gain code vector GC _k ^G_min corresponding to the encoded gain information G_min directly as the gain value. In Equation (38), k and j 'are respectively associated in ascending order in the same manner as in Equation (37).

Gain_q '(j') =

Equation 38

Region(m_max) соответственно, ассоциированном в порядке возрастания. Здесь вычисленный декодированный коэффициент MDCT обозначается X''_k. Кроме того, при деквантовании коэффициента MDCT, если k присутствует в B(j')-B(j'+1)-1, значение усиления принимает значение Gain_q' (j')[0232] Then, the gain dequantization section 2504 calculates the decoded MDCT coefficient in accordance with Equation (39) below using the gain value obtained by dequantizing the current frame and the shape value input from the form dequantization section 202, and updates the internal buffer in accordance with Equation ( 40) below. In Equation (40), the value C ″ ¹ _{j is} updated from j dequantized gain value Gain_q ′ (j), and j ′ satisfying condition j ′

Region (m_max), respectively, associated in ascending order. Here, the calculated decoded MDCT coefficient is denoted by X ' _k . In addition, when dequantizing the MDCT coefficient, if k is present in B (j ') - B (j' + 1) -1, the gain value takes on the value Gain_q '(j')

Equation 39

Equation 40

[0233] The gain dequantization section 2504 outputs the decoded MDCT coefficient X _{k k} calculated in accordance with Equation (39) above to the total MDCT coefficient calculation section 452.

[0234] Thus, according to this embodiment, compared with the selection of one region composed of adjacent subbands from among all ranges as the target quantization range, a plurality of ranges for which it is desirable to improve audio quality are set in advance over a wide range, and an inconsistent set of ranges covering a wide range is selected as the target quantization ranges. Therefore, the quality of both low range and high range can be improved simultaneously.

[0235] In this embodiment, the reason for always locking the subbands included in the target quantization range on the high range side, as shown in FIG. 23, is that the encoding distortion is still large for the high range in the first layer of the scalable codec. Therefore, audio quality is also improved steadily by selecting a high range that has not been encoded with very high accuracy by the first level as a quantization target, in addition to selecting a low or medium range having a perceptual value for selecting as a quantization target in the second level.

[0236] In this embodiment, a case has been described by way of example where the range that becomes the target of quantizing a high range is fixed by including the same subbands of a higher subband (more specifically, subbands with indices 15 and 16) over all frames, but the present invention is not limited to this, and the range that becomes the target of high range quantization can also be selected from a plurality of target quantization range candidates for the high subband Range same manner as for the low-range sub-band. In this case, the selection can be made after multiplying by the greater the weight, the higher the region of the subrange. It is also possible for ranges that become candidates for adaptive variation according to the sampling frequency of the input signal, the bit rate, and the spectral characteristics of a decoded first level signal, or the spectral characteristics of a differential signal for an input signal and a decoded first level signal, or the like. For example, a possible method is to prioritize the quantization target range, the part where the spectrum energy distribution (MDCT difference coefficient) of the differential signal for the input signal and the first decoded level signal is high.

[0237] In this embodiment, a case has been described by way of example where the group of subbands on the high band side constituting the region is fixed and whether predictive coding should be applied to the gain quantization section is determined according to the number of subbands common to the quantization target range selected in the current frame and the target quantization range selected in the last frame, but the present invention is not limited thereto, and predictive coding may also be entirely Always apply to the amplification of a group of subbands from the high range side constituting the region, with the determination of whether predictive coding should be performed having been performed only for the group of subbands from the low range side. In this case, the number of subbands common to the quantization target range selected in the current frame and the quantization target range selected in the last frame is taken into account only for the group of subbands on the low range side. That is, in this case, the quantization vector is quantized after dividing by the part for which predictive coding is performed, and the part for which predictive coding is not performed. Thus, since the determination of whether predictive coding is necessary for a fixed group of subbands on the high-band side constituting the region is not performed, and predictive coding is always performed, the gain can be quantized more efficiently.

[0238] In this embodiment, a case has been described by way of example where switching is performed between applying and not applying predictive coding in a gain quantization section according to the number of subbands common to the quantization target range selected in the current frame and the quantization target range selected one frame back in time, but the present invention is not limited to this, and many subbands common to the quantization target range selected in the current frame and the target range for quantization of the selected two or more frames back in time it may also be used. In this case, even if the number of subbands common for the quantization target range selected in the current frame and the quantization target range selected one frame back in time is less than or equal to a predetermined value, predictive coding can be applied in the gain quantization section according to the number subbands common to the quantization target range selected in the current frame and the quantization target range selected two or more frames backward in time.

[0239] In this embodiment, a case has been described by way of example in which the region consists of a group of subbands on the low side and a group of subbands on the high side, but the present invention is not limited thereto, and for example, the group of subbands can also be set to mid range, and the region may be composed of three or more groups of subbands. The number of groups of subbands constituting the region can also be adapted adaptively according to the sampling frequency of the input signal, the coding rate in bits, and the spectral characteristics of the decoded signal of the first level, or the spectral characteristics of the differential signal for the input signal and the decoded signal of the first level, or the like.

[0240] In this embodiment, a case has been described by way of example in which a group of subbands on the high range side constituting a region is captured in all frames, but the present invention is not limited thereto, and a group of subbands on the low side constituting the region may also be fixed across all frames. In addition, both the subband groups from the high band and the low band side constituting the region can also be fixed across all frames, or the subband group the group of subbands from both the high band and the low band side can be found and selected on the frame-by-frame basis. In addition, the various methods described above can be applied to three or more subband groups among the subband groups constituting the region.

[0241] In this embodiment, a case has been described by way of example in which of the subbands making up the region, the number of subbands making up the group of subbands on the high band side is less than the number of subbands making up the group of subbands on the low band side (the number of subbands of the group of subbands on the high side is two, and the subbands of the group of subbands on the low side is three), but the present invention is not limited thereto, and count honors subbands constituting the group of subbands from the high range may also be equal to or greater than the number of subbands composing a group of subbands from the low range. The number of subbands making up each group of subbands can also be adapted adaptively according to the sampling frequency of the input signal, the coding rate in bits, the spectral characteristics of the decoded signal of the first level, the spectral characteristics of the differential signal for the input signal and the decoded signal of the first level, or the like.

[0242] In this embodiment, a case has been described by way of example in which encoding using the CELP encoding method is performed by the first layer encoding section 302, but the present invention is not limited thereto, and encoding using a different encoding method than CELP (such as conversion encoding , for example) can also be performed.

[0243] Embodiment 7

26 is a block diagram illustrating a basic configuration of a speech encoding apparatus 1300 according to Embodiment 7 of the present invention.

[0244] In this drawing, the speech encoding device 1300 is provided with a downsizing section 301, a first level encoding section 302, a first level decoding section 303, a upsizing section 304, a first frequency domain conversion section 305, a delay section 306, and a second conversion section 307 in frequency domain, by second layer encoding section 1308, and multiplexing section 309, and has a scalable configuration comprising two levels. In a first layer, a speech CELP encoding method is used, and a second layer uses a speech encoding method described in Embodiment 1 of the present invention.

[0245] With the exception of the second level encoding section 1308, the configuration elements in the speech encoding device 1300 shown in FIG. 26 are identical to the configuration elements of the speech encoding device 300 shown in FIG. 6, and therefore, the same reference numbers are assigned to the same configuration elements , and their description is omitted here.

[0246] FIG. 27 is a block diagram illustrating a basic configuration of an interior of a second layer encoding section 1308. The second level coding section 1308 mainly comprises a MDCT differential coefficient calculation section 381, a range selection section 102, a shape quantization section 103, a predictive coding execution / non-execution determination section 3804, a gain quantization section 3805, and a multiplexing section 106. With the exception of the predictive coding execution / non-execution determination section 3804 and the gain quantization section 3805, the configuration elements in the second level encoding section 1308 are identical to the configuration elements of the second level encoding section 308 shown in FIG. 7 and therefore the same reference elements are assigned the same configuration elements positions, and their description is omitted here.

[0247] The prediction encoding execution / non-execution determination section 3804 has an internal buffer that stores range information m_max entered from the range selection section 102 in the last frame. Here, an example is described by a case in which the predictive coding execution / non-execution determination section 3804 has an internal buffer that stores m_max range information for the past three frames. Predictive encoding execution / non-execution determining section 3804 first detects a subband common to the quantization target range of the past frame and the quantization target range of the current frame using the m_max range information input from the range selection section 102 in the last frame and the m_max range information input from the section 102 range selection in the current frame. From the L subbands indicated by the m_max range information inputted from the range selection section 102, the prediction encoding execution / not determining section 3804 determines that the prediction coding should be applied and sets Pred_Flag (j) = ON for the subband selected as the quantization target one frame back in time. On the other hand, from the L subbands indicated by the range information m_max entered from the range selection section 102, the prediction encoding execution / not determining section 3804 determines that the prediction coding should not be applied and sets Pred_Flag (j) = OFF for the subband, not selected as the target of quantization one frame back in time. Here, Pred_Flag is a flag indicating the result of determining whether / not to execute predictive coding for each subband, with an ON value meaning that the predictive encoding should be applied to the subband gain value, and an OFF value meaning that the predictive encoding should not be applied to subband gain value. Predictive coding execution / non-execution determination section 3804 outputs a determination result for each subband to quantization section 3805. Then, the prediction encoding execution / non-execution determination section 3804 updates the internal buffer storing the range information using the range information m_max entered from the range selection section 102 in the current frame.

[0248] The gain quantization section 3805 has an internal buffer that stores the quantization gain value obtained in the last frame. The gain quantization section 3805 switches between execution / non-execution of the application of predictive coding when quantizing the gain value of the current frame according to a determination result inputted from the predictive encoding / not determination determination section 3804. For example, if predictive coding is to be performed, the gain quantization section 3805 searches for an internal gain codebook consisting of the number of GQ vectors of the gain code for each of the L subbands, performs distance calculation corresponding to the determination result input from the predictor fulfillment / failure determination section 3804 encoding, and finds the index of the gain code vector, for which the result of Equation (41) below is a minimum. In Equation (41), one or the other distance is calculated in accordance with Pred_Flag (j) for all j 'satisfying j∈Region (m_max), and the index of the gain vector is found for which the total error cost is minimal.

Уравнение 41

Equation 41

[0249] In this equation, GC ⁱ _k indicates the gain code vector constituting the gain codebook, i indicates the index of the gain code vector, and k indicates the index of the gain code vector element. For example, if the number of subbands making up the region is five (L = 5), k has a value from 0 to 4. Here, C ^t _j indicates the gain value of t frames backward in time, so that when t = 1, for example, C ^t _j indicates the gain value of one frame back in time. In addition, α is a fourth-order linear prediction coefficient stored in gain quantization section 3805. Gain quantization section 3805 processes the L subbands within the same region as an L-dimensional vector, and performs vector quantization.

[0250] The gain quantization section 3805 outputs the gain code vector index G_min, for which the result of Equation (41) above is a minimum, to the multiplexing section 106 as encoded gain information.

[0251] The gain quantization section 3805 outputs G_min to the multiplexing section 106 as encoded gain information. Gain quantization section 3805 also updates the internal buffer in accordance with Equation (42) below using the encoded gain information, G_min, and gain quantization value C ^t _j obtained in the current frame. In Equation (42), the value C ¹ _{j ′ is} updated by the index j of the element GC ^{G −min} _j of the gain code vector and j ′ satisfies the condition j′∈ Region (m_max), respectively, associated in ascending order.

Equation 42

[0252] FIG. 28 is a block diagram illustrating a basic configuration of a speech decoding apparatus 1400 according to this embodiment.

[0253] In this drawing, the speech decoding apparatus 1400 is provided with a control section 401, a first level decoding section 402, a upsampling section 403, a frequency domain conversion section 404, a second level decoding section 1405, a time domain conversion section 406, and a switch 407.

[0254] With the exception of the second level decoding section 1405, the configuration items in the speech decoding device 1400 shown in FIG. 28 are identical to the configuration items of the speech decoding device 400 shown in FIG. 8, and therefore, the same reference numbers are assigned to the same configuration items , and their description is omitted here.

[0255] FIG. 29 is a block diagram illustrating a basic configuration of an interior of a second layer decoding section 1405. The second level decoding section 1405 mainly comprises a demultiplexing section 451, a shape dequantization section 202, a predictive decoding execution / non-execution determination section 4503, a gain dequantization section 4504, and a total MDCT coefficient calculation section 452. With the exception of predictive decoding execution / non-execution decoding section 4503 and gain dequantization section 4504, the configuration elements in the second level decoding section 1405 shown in FIG. 29 are identical to the configuration elements of the second level decoding section 405 shown in FIG. 9 and therefore identical the configuration items are assigned the same reference numbers, and their description is omitted here.

[0256] Predictive decoding run / fail determination section 4503 has an internal buffer that stores information of the m_max range entered from demultiplexing section 451 in the last frame. Here, by way of example, a case is described in which the predictive decoding run / fail determination section 4503 has an internal buffer that stores m_max range information for the past three frames. Predictive decoding execution / non-execution determining section 4503 first detects a subband common to the quantization target range of the last frame and the quantization target range of the current frame using the m_max range information input from the last frame demultiplexing section 451 and the m_max range information input from section 451 demultiplexing in the current frame. From the L subbands indicated by the m_max range information inputted from the demultiplexing section 451, the predictive decoding execution / not deciding section 4503 determines that predictive decoding should be applied, and sets Pred_Flag (j) = ON for the subband selected as the quantization target one frame back in time. On the other hand, from the L subbands indicated by the m_max range information inputted from the demultiplexing section 451, the predictive decoding execution / not determining section 4503 determines that predictive decoding should not be applied and sets Pred_Flag (j) = OFF for the subband not selected as a quantization target one frame back in time. Here Pred_Flag is a flag indicating the result of determining whether / not to apply predictive decoding for each subband, with an ON value indicating that predictive decoding should be applied to the subband gain value, and an OFF value indicating that predictive decoding should not be applied to the value subband gain. Then, the predictive decoding run / fail determination section 4503 outputs a determination result for each subband to the dequantization section 4504. Then, the predictive decoding run / fail determination section 4503 updates the internal buffer storing the range information using the range information m_max input from the demultiplexing section 451 in the current frame.

Region(m_max) соответственно ассоциированном в порядке возрастания.[0257] The gain dequantization section 4504 has an internal buffer that stores the gain value obtained in the last frame and switches between performing / not performing predictive decoding when decoding the gain of the current frame according to the determination result input from the execution / not execution determination section 4503 predictive decoding. The gain dequantization section 4504 has the same kind of internal gain codebook as in the gain quantization section 105 of the speech encoding apparatus 100, and when performing predictive decoding, for example, obtains a gain value Gain_q ′ by performing gain dequantization in accordance with Equation (43) below. Here, C ^”t _j indicates the gain value of t frames backward in time, so that when t = 1, for example, C ^{” t} _j indicates the gain value of one frame earlier. In addition, α is a fourth order linear prediction coefficient stored in gain dequantization section 4504. Gain dequantization section 4504 processes the L subbands within the same region as an L-dimensional vector, and dequantizes the vector. In Equation (43), the value Gain_q '(j') is calculated with the index k of the element GC ^G_min _k of the gain code vector and j 'satisfying condition j'

Region (m_max), respectively, associated in ascending order.

Уравнение 43

Equation 43

[0258] Then, the gain dequantization section 4504 calculates the decoded MDCT coefficient in accordance with Equation (44) below using the gain value obtained by dequantizing the current frame and the shape value input from the form dequantization section 202, and updates the internal buffer in accordance with Equation ( 45) below. In Equation (45), the value of C ″ ¹ _j is updated by j dequantized gain value Gain_q ′ (j) and j ′ satisfying j′∈Region (m_max), respectively associated in ascending order. Here, the calculated decoded MDCT coefficient is denoted by X ' _k . In addition, in dequantizing the MDCT coefficient, if k is present in B (j ') - B (j' + 1) -1, the gain value takes on the value Gain_q '(j').

Уравнение 44

Equation 44

Equation 45

[0259] The gain dequantization section 4504 outputs the decoded MDCT coefficient X ' _k calculated in accordance with Equation (44) above to the total MDCT coefficient calculation section 452.

[0260] Thus, according to this embodiment, during the quantization of the gain of the quantization target range selected in each frame, it is detected whether each subband included in the quantization target range was quantized in the last frame. Then, vector quantization is performed with predictive coding applied to the subband quantized in the past frame and with predictive coding not applicable to the subband not quantized in the past frame. By this, the encoding of the frequency domain parameter can be performed more efficiently than with the method by which the switching of application / not application of predictive coding is performed for the entire vector.

[0261] In this embodiment, a method has been described by which switching between applying and not applying predictive coding in a gain quantization section according to the number of subbands common to the quantization target range selected in the current frame and the quantization target range selected one frame back in time, but the present invention is not limited thereto, and many subbands common to the quantization target range selected in the current frame may also be used. , and a quantization target range selected two or more frames backward in time. In this case, even if the number of subbands common for the quantization target range selected in the current frame and the quantization target range selected one frame back in time is less than or equal to a predetermined value, predictive coding can be applied in the gain quantization section according to the number subbands common to the quantization target range selected in the current frame and the quantization target range selected two or more frames backward in time.

[0262] It is also possible to combine the quantization method described in this embodiment with the quantization target range selection method described in Embodiment 6. A case is described below where, for example, a region that is a quantization target range consists of a group of subbands with side of the low range, and a group of subbands on the high side, and the group of subbands on the low side is fixed for all frames, and a vector is quantized in which the group of subbands on the side the low range and the group of subbands on the high range side are made sequential. In this case, in the gain vector of the quantization target range, vector quantization is performed with predictive coding always applied to an element indicating the gain of the subband group from the high range side, and predictive coding is not applied to the element indicating the gain of the subband group from the low range side. By this, quantization of the gain vector can be performed more efficiently than when predictive coding application / non-switching is performed for the entire vector. At this time, in the group of subbands on the low side, a method is also efficient whereby vector quantization is performed with predictive coding applied to a subband quantized in a past frame and with predictive coding not applicable to a subband not quantized in a past frame. In addition, for an element indicating the amplification of the group of subbands from the low range side, quantization is performed by switching between applying and not applying predictive coding using subbands constituting the quantization target range selected in the last frame in time, as described in Embodiment 1. By of this, quantization of the gain vector can be performed more efficiently. It is also possible to apply the present invention to a configuration that combines the above configurations.

[0263] The following concludes the description of embodiments of the present invention.

[0264] In the above embodiments, examples have been described by way of example in which the method of selecting the target quantization range should select the region with the highest energy in all ranges, but the present invention is not limited thereto, and a certain range can also be preselected, after which the target the quantization range is finally selected in a pre-selected range. In such a case, the preselected range may be determined according to the sampling frequency of the input signal, bit rate coding, or the like. For example, one way is to preselect a low range when the sampling rate is low.

[0265] In the above embodiments, the MDCT is used as a transform coding method, and therefore, the “MDCT coefficient” used in the above embodiments essentially means “spectrum”. Therefore, the expression "MDCT coefficient" can be replaced by "spectrum".

[0266] In the above embodiments, examples are shown in which speech decoding devices 200, 200a, 400, 600, 800, 1010, 1200 and 1400 receive and processes encoded data transmitted from devices 100, 100a, 300, 500, 700, 1000, 1100, and 1300 speech coding, respectively, but the encoded data output by an excellent configuration encoding device capable of generating encoded data having a similar configuration can also be input and processed.

[0267] An encoding device, a decoding device, and methods thereof according to the present invention are not limited to the above-described embodiments, and various changes and modifications may be possible without departing from the scope of the present invention. For example, it is possible to combine embodiments that are to be implemented, respectively.

[0268] It is possible to install an encoding device and a decoding device according to the present invention in a communication terminal device and a base station device in a mobile communication system, thereby providing a communication terminal device, a base station device, and a mobile communication system that provide the same kind of operational effects, as described above.

[0269] A case has been described here by way of an example in which the present invention is configured as hardware, but the present invention is also possible to implement as software. For example, the same kind of functions that the encoding device and the decoding device according to the present invention can be implemented by recording the algorithm of the encoding method and the decoding method according to the present invention in a programming language, storing this program in memory, and executing it by the information processing means.

[0270] Functional blocks used in the descriptions of the above embodiments are typically implemented as LSIs, which are integrated circuits. They can be implemented separately as separate microcircuits, or a single microcircuit may include some or all of them.

[0271] Here, the term LSI is used, but the terms IP, system LSI, super-LSI, ultra-LSI, etc. can also be used according to differences in the degree of integration.

[0272] A method for implementing integrated circuitry is not limited to LSIs, and implementation by means of a specialized circuit or universal processor may also be used. An FPGA (User Programmable Gate Array), for which programming is possible after LSI fabrication, or a reconfigurable processor that allows reconfiguration of circuit cell connections and parameters within the LSI, can also be used.

[0273] In the case of the introduction of integrated circuit technology, whereby the LSI is replaced by an excellent more modern technology or derived from semiconductor technology, the integration of function blocks can of course be performed using this technology. The use of biotechnology or the like is also possible.

[0274] Disclosure of Japanese Patent Application No. 2006-336270, filed December 13, 2006, Japanese Patent Application No. 2007-053499, filed March 2, 2007, Japanese Patent Application No. 2007-132078, filed May 17, 2007, and patent application Japan No. 2007-185078, filed July 13, 2007, including a description, drawings and abstracts, are incorporated herein by reference in their entirety.

Industrial applicability

[0275] Encoding device, etc. according to the present invention is suitable for use in communication terminal equipment, a base station device, or the like, in a mobile communication system.

Claims (20)

1. An encoding device comprising:
a conversion section that converts the input speech / audio signal to the frequency domain to obtain a frequency domain parameter;
a selection section that selects at least one subband from the plurality of subbands as the quantization target range, the plurality of subbands obtained by dividing the frequency domain parameter, and generates range information indicating the quantization target range;
a shape quantization section that quantizes a shape of a frequency domain parameter in a target quantization range;
a gain quantization section that encodes a gain of a frequency domain parameter in a target quantization range to obtain encoded gain information; and
a determination section that determines whether predictive coding should be performed based on the number of first subbands of the quantization target range that are common to the second subbands of the quantization target range selected in the past;
wherein the gain quantization section obtains encoded gain information by performing predictive coding with respect to the gain of the frequency domain parameter in the quantization target range using past encoded gain information when the determination section determines that predictive encoding should be performed and obtains encoded information gain by directly quantizing the gain of the parameter h the frequency domain in the target quantization range when the determination section determines that predictive coding should not be performed. 2. The encoding device according to claim 1, further comprising a determination section that determines that predictive encoding should be performed when the number of subbands common for the quantization target range and past quantization range is greater than or equal to a predetermined value, and determines that predictive coding should not be performed when the number of common subbands is less than a predetermined value. 3. The encoding device according to claim 1, in which the gain quantization section obtains encoded gain information by performing quantization of the frequency domain parameter gain vector. 4. The encoding device according to claim 1, wherein the gain quantization section obtains encoded gain information by performing predictive gain quantization using the gain of the frequency domain parameter in the last frame. 5. The encoding device according to claim 1, in which the selection section selects a region for which the energy is the highest among regions composed of a plurality of subbands as the target quantization range. 6. The encoding device according to claim 1, in which the selection section, when there are candidate ranges for which the number of subbands common for the quantization target range selected in the past is greater than or equal to a predetermined value, and the energy is greater than or equal to a predetermined value , selects the range for which the energy is the highest among the candidate ranges, as the target quantization range, and when the candidate ranges do not exist, selects the range for which the energy is the highest Oka in all ranges of the frequency domain as the quantization target band. 7. The encoding device according to claim 1, in which the selection section selects the range closest to the quantization target range selected in the past among the ranges for which the energy is greater than or equal to a predetermined value as the quantization target range. 8. The encoding device according to claim 1, in which the selection section selects the target quantization range after multiplying by a weight coefficient, which is greater, the closer the subrange to the low side of the range. 9. The encoding device according to claim 1, in which the selection section selects a subband fixed on the low side of the range as the target quantization range. 10. The encoding device according to claim 1, in which the selection section selects the target quantization range after multiplying by a weight coefficient, which is the greater, the higher the frequency of selecting a subband in the past. 11. The encoding device according to claim 1, further comprising an interpolation section that interpolates the gain of the frequency domain parameter in a subband not quantized in the past among the subbands indicated by the range information using the encoded past gain information to obtain an interpolation value,
wherein the gain quantization section also uses the interpolation value when performing predictive coding. 12. The encoding device according to claim 1, further comprising a determination section that determines the prediction coefficient in such a way that the weight of the gain value of the past frame is greater, the larger the subband common to the quantization target range of the past frame and the quantization target range of the current frame,
wherein the gain quantization section uses a prediction coefficient when performing predictive coding. 13. The encoding device according to claim 1, wherein the selection section fixedly selects a predetermined subband as part of the quantization target range. 14. The encoding device according to claim 1, in which the selection section selects the target quantization range after multiplying by a weight coefficient, which is greater, the greater the greater the sub-range to the high side of the range in part of the quantization target range. 15. The encoding device according to claim 1, in which the gain quantization section performs predictive coding with respect to the gain of the frequency domain parameter in part of the target quantization range and performs direct quantization with respect to the gain of the frequency domain parameter in the remaining part. 16. The encoding device according to claim 1, wherein the gain quantization section quantizes a gain vector of an inconsistent plurality of subbands. 17. A decoding device comprising:
a receiving section that receives information indicating a quantization target range indicating at least one selected subband from a plurality of subbands, the plurality of subbands obtained by dividing a frequency domain parameter of an input speech / audio signal;
a shape dequantization section that decodes encoded shape information in which a frequency domain parameter shape in a target quantization range is quantized to generate a decoded shape;
a gain dequantization section that decodes encoded gain information in which a gain of a frequency domain parameter in a target quantization range is quantized to generate a decoded gain, and decodes a frequency parameter using a decoded shape and a decoded gain to generate a decoded frequency domain parameter ;
a time-domain conversion section that converts a decoded frequency-domain parameter into a time-domain and receives a time-domain decoded signal; and
a determination section that determines whether predictive decoding should be performed based on the number of first subbands of the quantization target range that are common to the second subbands of the quantization target range selected in the past;
wherein the gain dequantization section performs predictive decoding of the gain of the frequency domain parameter in the target quantization range using the gain obtained in the past decoding of the gain when the determination section determines that predictive decoding should be performed and directly dequantizes the encoded gain information in which the gain of the frequency domain parameter is quantized in the target range of the quant When the determination section determines that predictive decoding should not be performed. 18. The decoding apparatus according to claim 17, further comprising a determination section that determines that predictive decoding should be performed when the number of subbands common for the quantization target range and past quantization range is greater than or equal to a predetermined value, and determines that predictive decoding should not be performed when the number of common subbands is less than a predetermined value. 19. An encoding method comprising the steps of:
converting the input speech / audio signal to the frequency domain to obtain a frequency domain parameter;
selecting at least one subband of the plurality of subbands as the quantization target range, the plurality of subbands being obtained by dividing a frequency domain parameter, and generating band information indicating the quantization target range;
quantizing the shape of the frequency domain parameter in the target quantization range to obtain encoded shape information; and
encoding a gain of a frequency domain parameter in a quantization target range to obtain encoded gain information;
determining whether predictive coding should be performed based on the number of first subbands of the quantization target range that are common to the second subbands of the quantization target range selected in the past; and
gain encoded gain information is obtained by performing predictive encoding with respect to the gain of the frequency domain parameter in the target quantization range using past encoded gain information when the determination section determines that predictive encoding should be performed, and encoded gain information by directly quantizing the gain gain parameter of the frequency domain in the target quantization range, hen determining section determines that predictive encoding is not to be performed. 20. A decoding method comprising the steps of:
receiving information indicative of a quantization target range indicating at least one selected subband from a plurality of subbands, the plurality of subbands obtained by dividing a frequency domain parameter of an input speech / audio signal;
decode encoded shape information, wherein the frequency domain parameter shape in the target quantization range is quantized to generate a decoded shape;
decode the encoded gain information, wherein the gain of the frequency domain parameter in the target quantization range is quantized to generate a decoded gain, and the frequency domain parameter is decoded using the decoded shape and the decoded gain to generate the decoded frequency domain parameter;
converting the decoded parameter of the frequency domain into the time domain to obtain a decoded signal of the time domain;
determining whether predictive decoding should be performed based on the number of first subbands of the quantization target range that are common to the second subbands of the quantization target range selected in the past; and
performing predictive decoding of the gain of the frequency domain parameter in the target quantization range using the gain obtained in the past decoding of the gain when the determination section determines that predictive decoding should be performed, and directly quantizing the encoded gain information in which the gain of the parameter the frequency domain is quantized in the target quantization range when the determination section determines that predictive decoding should not be performed.

Images