JP2000242299A - Weight code book and its generating method, setting method for initial value of ma prediction coefficient in learning in code book designing, encoding method and decoding method for sound signal, computer-readable storage medium storing encoding program and computer- readable storage medium storing decording program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make suppressible deterioration in the sound quality of a relatively-low-sound- pressure part of a decoded sound signal by setting respective elements of a weight code book by using a distance scale determined by taking human sound pressure sense characteristics into account and encoding a sound signal by using the weight code book which is thus generated. SOLUTION: In order to calculate the distortion (d) of a reproduced sound candidate (y) while taking auditory weighting into consideration, a distortion calculation part 1-6 inputs the reproduced sound candidate (d) outputted from a composing filter after passing it through an auditory weighting filter 2-2 and also inputs an input sound (x) to the distance calculation part 2-4 after passing it through an auditory weighting filter 2-3. The auditory weighting filter 2-2 has a linear prediction parameter (a) as a coefficient, converts the reproduced sound candidate (y) into a reproduced sound candidate yw with auditory weight, and outputs it to the distance calculation part 2-4. The auditory weighting filter 2-3 is a filter which uses a fixed coefficient, and converts an input sound (x) into an input sound xw with auditory weight and outputs it to the distance calculation part 2-4.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a weighted codebook, a method for preparing the same, a method for setting initial values of MA prediction coefficients during learning when designing a codebook, a method for encoding an audio signal, a method for decoding the same, and an encoding method. The present invention relates to a computer-readable storage medium in which a program is stored and a computer-readable storage medium in which a decoding program is stored, and more particularly to a technique for encoding or decoding an audio signal.

[0002]

2. Description of the Related Art In digital mobile communication, in order to use a communication line and a storage medium efficiently, for example, a high-efficiency coding system for minimizing distortion of a reproduced audio signal under a predetermined distance scale is adopted. Is done. One of the high efficiency coding methods
First, in the time domain, an audio signal is divided into sections called frames or subframes at regular intervals of about 5 to 50 ms, and one frame is divided into a signal (short-term prediction signal) representing an envelope characteristic of a frequency spectrum and a prediction residual thereof. Has been proposed in which the short-term predicted signal and the driving excitation signal are separately encoded.

[0003] In addition, as a coding method of the above-mentioned driving excitation signal, there is a code driving linearity which separates and codes a frequency component corresponding to a pitch period (basic frequency) of a voice and other components (non-periodic components). Predictive coding method (CELP: Code-
Excited Linear Prediction) is known. This code-driven linear prediction coding method is described in the document "MRSchr
oeder and BSAtal, "Code-Excited Linear Predictio
n (CELP): High QualitySpeech at Very Low Bit Rates ",
IEEE Proc. LCASSP-85, pp937-940, 1985].

In the case of encoding an acoustic signal, for example, vector quantization using MA prediction (MA prediction vector quantization) is used as a code-driven linear prediction encoding method. The prediction vector quantization includes AR prediction vector quantization using AR prediction and MA prediction vector quantization, and both of them can remove correlation between frames using prediction. In general, AR prediction vector quantization is often used as a method of removing predictable components from a correlated signal sequence and whitening it. However, in audio signal encoding, improvement of transmission error resistance is improved. For this reason, MA prediction vector quantization is used.

In the MA prediction vector quantization, a predetermined number of representative vectors are previously set as elements in a codebook,
A signal corresponding to the driving sound source signal is selected by selecting a representative vector for each frame or subframe according to the driving sound source signal and inputting the selected representative vector to an IIR filter having a predetermined filter coefficient (MA prediction coefficient). Generate Then, at the time of decoding, the signal is reproduced using an FIR filter having the same MA prediction coefficient as the IIR filter.

[0006] In such MA prediction vector quantization, it is necessary to appropriately determine the codebook and the MA prediction coefficient. Conventionally, each element of the codebook and the MA prediction coefficient are determined by a recursive learning method as shown in FIG. In this learning method, an initial codebook design process (9
In -1), an initial codebook is designed. The LBG algorithm is applied as a method for designing the initial codebook. This design method uses an initial codebook such that the total sum of distortion when quantizing learning data (sound signal for learning) having a distribution that well represents the distribution of the vector space to be vector-quantized is minimized. It is a design method. LBG
The algorithm determines one initial value of the representative vector,
An operation of obtaining an optimal representative vector and an operation of doubling the number of representative vectors are alternately performed to obtain a desired number of representative vectors.

[0007] This LBG algorithm means an operation of dividing a set of learning vectors into several subsets,
The above subset is called a cluster, and the division process is called clustering. For details of the LBG algorithm, see the document "Y.
Lindle A. Buzo.EMGray, "AnAlgorithm for Vector Qlu
antizer Design ", IEEE Trains.Commun.COM-28 pp84-
95, 1980]. For the prediction vector quantization, see the document "AllenGersho, Robert M. Gray," VECTO QUA.
NTIZATI0N AND SlCNAL C0MPRESSION ", KliwerAcademic P
ublishers, 1992].

Subsequently, MA prediction coefficient determination processing (9-2)
Then, the MA prediction coefficient is provisionally determined using the learning data and the initial codebook so that the total sum of the distortion of the reproduction signal is minimized. Then, when the learning data is encoded in the process (9-3), the processes (9-4) to (9-9) are repeated thereafter, so that the total sum D of the distortion of the reproduction signal is minimized. Finally, the representative vector and the MA prediction coefficient are finally determined.

That is, by recursively updating which representative vector the learning data belongs to, the representative vector is updated so that the total sum of the distortion is minimized in the belonging state (9-4). By repeating the belonging update processing and the update processing of the representative vector alternately until convergence, the representative vector gradually approaches an optimum one. This operation is called Lloyd-Max or k-means algorithm.

The quantization of the MA prediction vector of the weighting codebook is described in the document "ITU-T COM 15-152-E," G.729-Cod.
ing of Speech at 8kbit / s using Conjugate-Structure
Algeraic-Code-Excited / Linear-Prediction (CS-ACEL
P) ", Jul. 1995. Similarly, regarding inter-frame prediction vector quantization using MA prediction, Nakamura Omuro, Takehiro Moriya, Kazunori Mano, Satoshi Miki," LSP using moving average type inter-frame prediction Vector Quantization of Parameters ”, Transactions of the Institute of Electronics, Information and Communication Engineers A Vol.J77-A No.3 pp303-313,1994
And so on.

[0011]

By the way, in the coding of an audio signal, the number of bits in digitally expressing the audio signal is limited. There is also a limit on the number of elements.
The code book determines each element in advance by using a plurality of acoustic signals for learning, and is used for encoding an acoustic signal to be actually transmitted.

However, conventionally, each element of the codebook is learned so that the sum of the waveform distortions of the reproduced audio signal is minimized with respect to the entire set of learning audio signals, that is, the S / N of the reproduced audio signal is maximized. Is determined by When each element of the codebook is determined on a distance scale that minimizes waveform distortion over the entire acoustic signal for learning, a codebook that focuses on a portion where the power of the input audio signal is large is used. Generated.

With respect to the sound pressure perception characteristics of human beings, a small change in power can be sufficiently perceived with a sound having a small power, and the perception sensitivity is rather rough at a portion where the power is large. For this reason, when the audio signal actually transmitted using the codebook obtained by the conventional learning is encoded, the ratio of the codebook element corresponding to the small part of the power of the audio signal is too small,
When the magnitude of the input audio signal is relatively small, sound quality degradation of the reproduced audio signal is noticeably perceived. As a result, the reproduced sound reproduced using such a codebook gives an unstable impression.

On the other hand, in order to obtain a high-quality reproduced audio signal even with a small number of bits, it is necessary to improve the conventional codebook learning method and improve the quantization efficiency of the input audio signal as much as possible. . For the quantization of the codebook elements, MA prediction vector quantization using an FIR filter at the time of decoding has been used to improve transmission error resistance. In order to obtain the prediction performance of the MA prediction, the order of the MA prediction (IIR filter and FI
R order) must be set to a very high order. However, a small order of about four is often used in consideration of a recovery time from a transmission error, and it has been extremely difficult to stably determine an MA prediction coefficient of this low order in an open loop.

In the conventional MA prediction vector quantization, as described above, the MA prediction coefficient cannot be determined unless the initial codebook is determined first, and the initial codebook cannot be determined unless the MA prediction coefficient is determined. There is a constraint condition.
Once the initial codebook is determined, each value asymptotically converges to a local optimal solution by a recursive learning method that alternately learns the initial codebook and MA prediction coefficients. Traditionally,
A method using an initial codebook designed without using MA prediction or a method of empirically determining an initial value of an MA prediction coefficient is employed. Therefore, depending on the initial codebook, a local solution may be obtained or an appropriate value may be given. Since the MA prediction coefficient may not be stable, it is difficult to determine an optimal initial codebook, and a codebook obtained as a result of recursive learning can only obtain a local optimal solution.

The present invention has been made in view of the above problems, and has the following objects. (1) Deterioration of sound quality of a reproduced sound signal when the size of the input sound signal is relatively small. (2) A high-quality reproduced sound signal is obtained even with low bit number coding.

[0017]

In order to achieve the above-mentioned object, according to the present invention, as a first means relating to a weight codebook and a method for producing the same, an input audio signal is converted into an envelope characteristic of a short-term frequency spectrum. Is divided into a short-term prediction signal indicating the prediction result of the above and a driving excitation signal indicating the prediction residual thereof, a normalization vector is selected based on the driving excitation signal, and a weight defining the magnitude of the normalization vector is determined. A driving excitation vector candidate is generated by selecting the weighting amount from each of the weighting codebooks as elements, and distortion of the reproduced audio candidate generated based on the driving excitation vector candidate and the short-term prediction signal with respect to the input audio signal is reduced. Generating a driving excitation signal such that the driving excitation signal is minimized under a predetermined distance measure, and encoding the driving excitation signal thus generated and the short-term prediction signal in a coding method. In creating a weight codebook, to adopt a means of setting each element of the weight codebook using a distance measure that takes into account the human sound pressure sensory properties.

[0018] The second related to the weight codebook and the method of producing the same.
In the first means, a means of using a distance scale in which a human sound pressure perception characteristic is added as a function of the power of an input sound signal is used.

The third related to the weight codebook and the method of producing the same.
In the second means, when the input acoustic signal is x, the function is expressed as (| x | ² ) ^p / | x | ²
The means of giving by means is adopted. Where p is
It is a value in the range of 0 <p <1.

According to the present invention, as a first means relating to a method of setting an initial value of an MA prediction coefficient, a method of setting an initial value of an MA prediction coefficient in MA prediction vector quantization comprises calculating from a plurality of learning data. A means of calculating the initial value of the MA prediction coefficient by approximating the obtained AR prediction coefficient using the MA prediction method is adopted.

As a second means relating to the method of setting the initial value of the MA prediction coefficient, M
In the method of setting the initial value of the A prediction coefficient, a process of calculating an average value of a plurality of learning data and a first AR based on a covariance method using a value obtained by subtracting the average value from each learning data. A step of calculating a prediction coefficient, a step of obtaining an impulse response of a filter using the first AR prediction coefficient as a filter coefficient, and a step of receiving the impulse response and inputting the second impulse response having the same order as the MA prediction coefficient based on the autocorrelation method. The step of calculating the AR prediction coefficient and the inverse filter of the filter using the second AR prediction coefficient as the filter coefficient are used to obtain MA.
Means for calculating the initial value of the prediction coefficient.

Further, in the present invention, as a first means relating to a weighting codebook for quantizing an MA prediction vector and a method for producing the same, the above-mentioned means relating to a method for setting an initial value of an MA prediction coefficient are used. When the initial value of the coefficient is calculated, L
In the process of creating the initial codebook of the weighted codebook based on the BG algorithm, and in the MA prediction vector quantization processing including the initial value of the MA prediction coefficient and the initial codebook, calculation is performed for each learning data by recursive learning. Means for sequentially updating each element of the weighted codebook and the MA prediction coefficient so as to minimize the sum of the distortion of the reproduced signal to be performed, and finally determining each element.

As a second means relating to the weighting codebook for MA prediction vector quantization and a method for producing the same, in the first means, the distortion of the reproduced signal is obtained by using a distance scale taking into account the human sound pressure perception characteristics. A means of calculating by calculation is adopted.

As a third means relating to the weighting codebook for quantizing the MA prediction vector and a method for producing the same, in the first means, the distortion of the reproduced signal may be calculated by using a distance scale in which a function of the power of the input audio signal is added. A means of calculating by using this is adopted.

As a fourth means relating to the weighting codebook for quantizing the MA prediction vector and a method for producing the same, in the above-mentioned first means, the distortion of the reproduced signal is calculated by calculating the power of the input acoustic signal x represented by the following equation (1). Means is calculated using a distance scale that takes into account the function W. Where p is 0 <
It is a value within the range of p <1. W = (| x | ² ) ^p / | x | ² (1)

Further, according to the present invention, as a first means relating to a method of encoding an audio signal, an input audio signal is converted into a short-term prediction signal indicating a prediction result of an envelope characteristic of a short-term frequency spectrum and a prediction residual thereof. And a normalization vector is selected based on the driving excitation signal, and a weight defining the magnitude of the normalization vector is selected from a weight codebook having the weight as an element. To generate a driving sound source vector candidate, such that the distortion of the input sound signal of the reproduced sound candidate generated based on the driving sound source vector candidate and the short-term prediction signal is minimized under a predetermined distance scale. In a sound signal encoding method for generating a driving sound source signal and coding the thus generated driving sound source signal and the short-term prediction signal, a distance considering a sound pressure perception characteristic of a human is added. Adopting means of setting each element of the weight codebook using a scale.

As a second means relating to a method of encoding an audio signal, the first means is a means in which a distance scale is used in which a human sound pressure perception characteristic is added as a function of the power of an input audio signal. adopt.

As a third means relating to an audio signal encoding method, when the input audio signal is x in the second means, the function is expressed by (| x | ² ) ^p / | x | ² The means of giving is adopted. Here, p is 0 <p
<Value within the range of 1.

As a fourth means relating to an audio signal encoding method, in any one of the above first to third means,
In the case of generating a driving excitation vector candidate from a driving excitation signal using prediction vector quantization, the weight codebook and MA prediction obtained by the first means relating to the method for creating a weighting codebook for MA prediction vector quantization are described. A means of using coefficients is employed.

Further, according to the present invention, as the first means relating to the above-described audio signal decoding method, an audio for decoding a code generated by any one of the first to third means relating to the above-described audio signal encoding method is provided. In the signal decoding method, means for decoding the code using a weighting codebook used in the encoding method of these audio signals is employed.

As a second means relating to the audio signal decoding method, a sound signal decoding method for decoding the code generated by the fourth means relating to the audio signal encoding method is provided. Means for decoding the code using a weighted codebook and MA prediction coefficients used in the decoding method.

Further, the present invention employs a means for performing the processing according to each means relating to the above-described audio signal encoding method into an encoded program stored in a computer-readable storage medium.

Further, a means is adopted in which the processing according to each means relating to the method of decoding an acoustic signal is converted into a decoding program stored in a computer-readable storage medium.

[0034]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring now to the drawings, a weighted codebook according to the present invention, a method for producing the codebook, a method for setting an initial value of an MA prediction coefficient at the time of learning at the time of codebook design, and encoding of an audio signal will be described. An embodiment of a method and a decoding method thereof, a computer-readable storage medium storing an encoding program, and a computer-readable storage medium storing a decoding program will be described.

FIG. 1 is a block diagram showing a functional configuration of an encoder according to the present embodiment. As shown in this figure,
The input speech (input audio signal) x is input to the linear prediction analysis unit 1-1.
And the distortion calculator 1-6. The linear prediction analysis unit 1-1 calculates a linear prediction parameter a indicating a frequency spectrum envelope characteristic of the input speech x, and outputs the calculated parameter to the linear prediction parameter encoding unit 1-2 and the distortion calculation unit 1-6.

Here, the input speech x is converted into a frame divided at intervals of 10 ms, for example, as a linear prediction analysis unit 1-1.
The linear prediction analysis unit 1-1
Represents the frequency spectrum envelope characteristic of each frame.

The linear prediction parameter encoding unit 1-2 encodes the linear prediction parameter a for each frame, and outputs the linear prediction parameter code af as the linear prediction parameter decoding unit 1.
-3 and the code transmission section 1-9. The linear prediction parameter decoding unit 1-3 includes a linear prediction parameter encoding unit 1-
2 reproduces the synthesis filter coefficient b from the linear prediction parameter code af input from 2 and sends it to the synthesis filter 1-5.

The synthesis filter 1-5 is a linear filter whose characteristics are defined by the synthesis filter coefficient b.
The synthesis filter coefficient b and the driving sound source vector generation unit 1-
The reproduction sound candidate y is generated and output using the driving sound source vector candidate c input from the step S4. This synthesis filter 1-5
, That is, the order of the linear prediction analysis in the linear prediction analysis unit 1-1 is, for example, about 10 to 16.

The linear prediction analysis section 1-1, linear prediction parameter encoding section 1-2, linear prediction parameter decoding section 1-3 and synthesis filter 1-5 may be replaced with non-linear ones. The details of the linear prediction analysis and the encoding of the linear prediction parameters are well-known technical matters, and are described in detail, for example, in Tadashi Furui, “Digital Speech Processing” (Tokai University Press).

The distortion calculator 1-6 calculates the distortion d of the reproduced speech candidate y with respect to the input speech x based on the input speech x, the linear prediction parameter a, and the reproduced speech candidate y input from the synthesis filter 1-5. It is calculated for each frame and output to the codebook search control unit 1-8.

FIG. 2 is a block diagram showing a detailed configuration of the distortion calculator 1-6. As shown in the figure, the distortion calculation unit 1-6 outputs the reproduced speech candidate y output from the synthesis filter 1-5 in the auditory sense in order to calculate the distortion d of the reproduced speech candidate y in consideration of the auditory weighting. Weight filter 2-2
And input to the distance calculation unit 2-4 through the auditory weight filter 2-3 while inputting to the distance calculation unit 2-4.

The auditory weight filter 2-2 uses the above-mentioned linear prediction parameter a as a coefficient.
Is converted into a perceptually weighted reproduced voice candidate yw and output to the distance calculator 2-4. The auditory weight filter 2-3 is
This is a filter using fixed coefficients, and converts the input voice x into an auditory weighted input voice xw and outputs it to the distance calculation unit 2-4. These auditory weight filters 2-2 and 2-3 are equivalent even if they are inserted as one filter after the distance calculator 2-4, but are independent before the distance calculator 2-4 from the viewpoint of processing amount. It is provided.

On the other hand, the code extension search control unit 1-8 selects the driving excitation code e such that the total sum of the distortions d of the reproduced speech candidate y is minimized for each frame, and selects the driving excitation vector generation unit 1-8. 4 is output. In the present embodiment, FIGS.
, The driving excitation vector generation unit 1-4 generates the driving excitation vector candidate c using the adaptive codebook 3-1, the fixed codebook 2-2, and the weighting codebook 4-1. It is configured to calculate.

In connection with this, the code extension search control unit 1
-8 is a periodic code for the adaptive codebook 3-1, a fixed (noise) code and a weighted code for the fixed codebook 2-2 such that the sum of the distortions d of the reproduced speech candidate y is minimized for each frame. Book 4
The weighting code for −1 is selected and output to the driving excitation vector generating section 1-4 as the driving excitation code e.

The code transmitting section 1-9 outputs the driving excitation code e and the linear prediction parameter code af to a storage device or a transmission path according to a use mode. The driving sound source vector generation unit 1-4 generates a driving sound source vector candidate c having a length of one frame of the input sound x and outputs the generated driving sound source vector c to the synthesis filter 1-5.

FIG. 3 shows the driving sound source vector generating section 1-.
4 is a block diagram showing a detailed configuration of FIG. In this figure, adaptive codebook 3-1 stores driving excitation vector c one frame before (past) stored in a buffer unit (not shown).
Based on (t-1) and the periodic code included in the driving excitation code e, a time-series vector candidate Va corresponding to the periodic component of the input speech x is generated for each frame.

The adaptive codebook 3-1 cuts out the driving excitation vector c (t-1) at a length corresponding to a predetermined period using a periodic code, and repeats the cut-out vector until the length of the frame becomes the length of the frame. Thus, a periodic code vector candidate Va (normalized vector) corresponding to the periodic component of the input speech x is selected and output. The adaptive codebook 3-1 stores a predetermined number of normalized vectors corresponding to the periodic codes as respective elements.

The predetermined period is the distortion d of the reproduced voice candidate y.
Is selected to be small, but generally, the input voice x
This is a cycle corresponding to the pitch cycle of. This embodiment also follows this. The fixed codebook 3-2 selectively outputs a fixed code vector candidate (normalized vector) having a length of one frame corresponding to the aperiodic component of the voice. The fixed code vector candidate is stored in the fixed code book 3-2 as a candidate vector of a predetermined number according to the number of bits for encoding independently of the input speech x.

The periodicization section 3-3 is a time series vector candidate Vf obtained by periodicizing the fixed code vector candidates output from the fixed code book 3-2 at the above-mentioned predetermined period (corresponding to the bitch period) specified by the periodic code. Is output. This periodicization is performed by applying a comb filter having a tap at a specified periodic position, or by repeating a vector cut out at a length corresponding to a specified period from the beginning of a frame similarly to the adaptive codebook 3-1. is there. Such a periodic unit 3-3 is often used from the viewpoint of improving coding efficiency. In addition, when there is no or little pitch component in the input voice x itself, such as in a consonant section, the periodic unit 3-3 does not perform any operation.

The weighting code generation unit 3-7 further divides the weighting code in frame units (10 ms units) included in the driving excitation code e into two subframes every 5 ms, thereby obtaining a normalized vector. Adaptive codebook 3-1
And a weight gf for the periodic code vector candidate Va and the weight gf for the non-periodic time-series vector candidate Vf (normalized vector) output from the periodicization unit 3-3 for each subframe. is there. The details of the weight code generation unit 3-7 will be described in detail below.

The multiplication unit 3-4 multiplies the periodic code vector candidate Va input from the adaptive codebook 3-1 by the weight ga input from the weight code generation unit 3-7 to obtain a vector candidate ca. Output to the adder 3-6. Multiplication unit 3-
5 is a time-series vector candidate Vf input from the periodizing unit 3-3 and a weight amount g input from the weight code generation unit 3-7.
, and outputs the result to the adder 3-6 as a vector candidate cf. The adding unit 3-6 adds the vector candidate ca and the vector candidate cf, and outputs the result to the synthesis filter 1-5 as a candidate c of the driving sound source vector.

FIG. 4 is a diagram showing a detailed configuration of the weight code generation section 3-7, wherein (a) is a block diagram,
(B) is a specific configuration example of the weight codebook 4-1. FIG.
As shown in (a), the weight code generation unit 3-7 applies the MA pre-vector quantization, and
-1 and the MA prediction unit 4-2.

The weighted codebook 4-1 is the adaptive codebook 3-
A predetermined number of two-dimensional vectors (element vectors) each having the weight coefficient ga for the normalized vector registered in 1 and the MA prediction difference xf of the weight for the normalized vector registered in the fixed codebook 3-2 as an element are registered. It was done. For example, when each element vector is 6-bit data, 64 (=) weight codebooks 4-1 as shown in FIG.
It is composed of ²⁶ ) element vectors.

The weighting codebook 4-1 selects an element vector based on the weighting code in the driving excitation code e when the actual input speech x is coded, and weights the selected element vector by a weighting factor. ga is output to the multiplication unit 3-4, and the MA prediction difference xf paired with the weight coefficient ga is calculated by MA.
Output to the prediction unit 4-2.

The MA prediction unit 4-2 delays the MA prediction difference xf by one sub-frame by (m-1) stages of serially connected buffers 4b1, 4b2,... 4b (m-1), A vector multiplication unit that multiplies the MA prediction difference xf output from the book 4-1 and the output of each buffer unit 4b1, 4b2,... 4b (m-1) by MA prediction coefficients a1, a2, a3,. 4
.., 4 km, and vector adders 4 a 1, 4 a 2, 4 a 3,... for sequentially adding the outputs of these vector multipliers 4 k 1, 4 k 2, 4 k 3,.

The MA predicting unit 4-2 thus configured
Is the m successive MA prediction differences xf (n), x
f (n-1), xf (n-2),... xf (nm) and MA prediction coefficient a1,
.., am, and outputs the MA prediction residual to the multiplier 3-5 as a weight coefficient gf for the fixed codebook 3-2.

Each of the element vectors of the weighting codebook 4-1 and the MA prediction coefficients a 1, a 2, a 3,... Am are determined by learning performed prior to actual encoding of the input speech x. In this learning, each element vector of the weight codebook 4-1 and the MA prediction coefficients a1, a2, a3,..., Am are determined so that the distortion d of the reproduced voice candidate y is minimized. The method of setting each element vector of the weight codebook 4-1 and the MA prediction coefficients a1, a2, a3,... Am will be described later.

FIG. 5 is a block diagram showing a functional configuration of a decoder corresponding to the encoder. The code receiving unit 5-1 receives the code received from the transmission path or the storage medium, and outputs the linear prediction parameter code to the linear prediction parameter decoding unit 5-2 and the driving excitation code e to the driving excitation vector generation unit 5- 3 respectively. The linear prediction parameter decoding unit 5-2 decodes the linear prediction parameter code to reproduce the synthesis filter coefficient b, and outputs the synthesis filter coefficient b to the synthesis filter 5-4 and the post-processing unit 5-5.

Driving excitation vector generation section 5-3 generates an excitation vector corresponding to driving excitation code e and outputs the generated excitation vector to synthesis filter 5-4. The configuration of the driving excitation vector generation section 5-3 corresponds to the configuration of the driving excitation vector generation section 1-4 in the encoder described above. The synthesis filter 5-4 reproduces the input sound x based on the driving sound source vector and the synthesis filter coefficient b, and outputs the reproduced sound to the post-processing unit 5-5. The post-processing unit 5-5 performs post-filtering to reduce reproduced audio noise in an auditory sense, and outputs the resultant to the outside as output audio. The post-processing unit 5-
5 may not be provided from the viewpoint of reducing the processing amount.

Next, the weighting codebook 4-1 and the MA prediction coefficient a applied to the encoder and decoder configured as described above.
A method for determining 1, a2, a3,... Am will be described.

First, when determining the weighting codebook 4-1 to be applied to the encoder and the MA prediction coefficients a1, a2, a3,... Am to be applied to the encoder and the decoder, the distance calculation unit 2-4 calculates the distortion d as follows.
That is, the value W given as a function of the power of the input voice x is calculated based on the following equation (1) in consideration of the auditory characteristics related to human sound pressure perception. Where p is in the range (0 <
Although it is a value within p <1), it is preferably about 0.3. W = (| x | ² ) ^p / | x | ² (1)

Also, the distortion d for each subframe is calculated based on the following distortion calculation equation (2) using the value W as a coefficient. Here, H is a lower triangular matrix indicating the characteristics of the synthesis filter 1-5, and the synthesis filter 1 is used as a main diagonal component.
The first-order component, second-order component,... Of the impulse response are arranged as lower-order diagonal components in the zeroth-order component of the impulse response of −5. T is the above auditory weight filter 2-2,
2-3 coefficients (hearing weight coefficients). d = W · T | x-ga · H · Va-gf · H · Vf | ² (2)

Then, by using the encoder in which the distance calculation section 2-4 is set as described above, the processing according to the flowchart shown in FIG. 6 is performed, whereby the weight codebook 4-1 is obtained.
And the MA prediction coefficients a1, a2, a3,..., Am.
Hereinafter, referring to the flowcharts shown in FIGS. 6 and 7,
A method of determining the weight codebook 4-1 and the MA prediction coefficients a1, a2, a3,... Am in the present embodiment will be described in detail.

The difference between this determination method and the conventional determination method (see FIG. 9) is that the MA prediction coefficients a1, a2, a
3,..., After determining the initial value of am (6-1), determining the initial codebook of the weighted codebook 4-1 (6-2). The following processes (6-1) to (6-9) are the same as in the related art.

In this embodiment, the MA prediction coefficients a1,
The initial values of a2, a3,...
Is calculated by open loop processing as shown in FIG. That is, in the DC component removal processing (7-1), the average of the learning data is obtained and divided from the learning data. In the AR prediction coefficient calculation processing 1 (7-2), a first AR prediction coefficient is obtained by a covariance method using data obtained by subtracting the average from the learning data. Impulse response generation processing (7-
In 3), a unit impulse is input to the filter constituted by the first AR prediction coefficient, and an impulse response having a sufficient length is obtained.

Then, AR prediction coefficient calculation processing 2 (7-
In 4), the second AR having the same order as the MA prediction coefficient is input by the autocorrelation method using the above impulse response as an input.
A prediction coefficient (mth order) is obtained. MA prediction coefficient calculation processing (7
In -5), an m-th order MA prediction coefficient a1, a2,
a3,... determine the initial value of am. By such an open loop process, initial values of the MA prediction coefficients a1, a2, a3,... Am having stable filter characteristics are obtained.

Here, if the MA prediction order for the impulse response of the first AR prediction coefficient is cut off at the m-th order as described above, it may be simply considered that the initial value of the MA prediction coefficient can be obtained. Is the AR prediction coefficient
To approximate by A prediction, a very high MA prediction order is required, and an MA prediction order used in audio signal encoding results in an unstable filter due to a truncation error.

Therefore, according to the present embodiment, a better MA prediction coefficient a
The initial values of 1, a2, a3,... Am can be determined. Conventionally, the MA prediction coefficients a1, a2, a3,...
Although the method of empirically determining the initial value of am has been adopted, in the present embodiment, the method of AR prediction and the method of MA prediction are combined by the open loop processing described above.
More optimal initial values of MA prediction coefficients a1, a2, a3,..., Am can be determined.

When the initial values of the MA prediction coefficients a1, a2, a3,... Am are determined as described above, based on the initial values, the weighted codebook 4 is obtained by the LBG algorithm in the same manner as in the prior art.
An initial codebook of -1 is determined. In this case, since the initial values of the more optimal MA prediction coefficients a1, a2, a3,... Am have already been determined as described above, the initial codebook can be efficiently determined.

Then, the MA prediction coefficient a
When the initial values of 1, a2, a3,... Am and the initial codebook of the weighting codebook 4-1 are determined, recursive learning processes (6-1) to (6-9) based on these are performed. MA prediction coefficient a
1, a2, a3,... Am and the weight codebook 4-1 are finally determined. That is, each element of the weighting codebook 4-1 and the MA prediction coefficients a 1, a 2, a 3 are set so that the sum D of the distortions d calculated based on the distortion calculation formula (2) for all the learning data is minimized. ,..., Am are finally determined.

The weight code book 4 finally determined in this way is
1 and the MA prediction coefficients a1, a2, a3,... Am are encoders for actually encoding an input signal x to be transmitted or stored in a storage medium, and a decoder for decoding and reproducing the input signal x (see FIG. 1 to 5). In this case, the distance calculation unit 2-4 minimizes the distortion d for each subframe based on the same distortion calculation formula as the conventional method, that is, the calculation formula with W = 1 in the above-described distortion calculation formula (2). To the input speech x.

An encoder and a decoder to which the weighted codebook 4-1 and the MA prediction coefficients a1, a2, a3,..., Am based on the present embodiment are applied are configured by software on a computer, for example, and subjected to a subjective evaluation test. A comparison with the conventional method was performed. As a result of this experiment, when the weighted codebook 4-1 was designed at the same pit rate, the evaluation result was better in the present embodiment than in the conventional method, and the effectiveness of the present invention was confirmed.

As shown in FIG. 8, by slightly modifying the processing of FIG. 6, the LSP quantizer LSP
It can be applied to codebook determination. After the initial value of the MA prediction coefficient is determined (8-1) as described above, L
When the initial codebook of the SP codebook is determined (8-1) and the learning data is encoded (8-3), the following processing (8-4)
(8-10) is repeated.

That is, in the process (8-4), the process of updating the first stage of the LSP codebook is performed, in the process (8-5), the learning data is encoded, and in the process (8-6), the LSP codebook is updated. Is updated, the MA prediction coefficient is updated in the process (8-7), the learning data is encoded in the process (8-8), and the distortion D (distortion d) is further processed (8-9). ) Can be determined by repeating the process of calculating the sum of the LSP codebooks.

[0075]

As described above, the weighted codebook according to the present invention, the method of creating the same, the method of setting the initial value of the MA prediction coefficient at the time of learning when designing the codebook, the method of encoding the acoustic signal, and the decoding thereof According to the method, the computer-readable storage medium storing the encoding program, and the computer-readable storage medium storing the decoding program, the following effects are obtained.

(1) The input audio signal is separated into a short-term prediction signal indicating the prediction result of the envelope characteristic of the short-term frequency spectrum and a driving sound source signal indicating the prediction residual, and are separated based on the driving sound source signal. A driving excitation vector candidate is generated by selecting a normalization vector and selecting a weight amount that defines the size of the normalization vector from a weighting codebook that uses the weight amount as each element. Generating the driving sound source signal such that distortion of the input sound signal of the reproduced sound candidate generated based on the short-term prediction signal is minimized under a predetermined distance scale, and the driving sound source signal thus generated is generated. And the short-term prediction signal and the encoding method of the weighted codebook in the encoding method, wherein each element of the weighted codebook using a distance scale taking into account human sound pressure perception characteristics. Because a constant, the weight codebook becomes large corresponding element to a relatively low portion of the sound pressure level as compared with the prior art. Therefore, when an audio signal is encoded using the weighted codebook created in this way, it is possible to suppress sound quality deterioration in a portion of the decoded audio signal having a relatively low sound pressure level.

(2) M in MA prediction vector quantization
This is a method of setting an initial value of the A prediction coefficient, in which the initial value of the MA prediction coefficient is calculated by approximating an AR prediction coefficient calculated from a plurality of learning data by using a method of MA prediction. A more accurate initial value of the MA prediction coefficient can be obtained as compared with the method of setting the initial value of the MA prediction coefficient. By using such an initial value of the MA prediction coefficient, it is possible to set the weight codebook and the MA prediction coefficient used for the MA prediction vector quantization to be more optimal.

(3) Based on the present invention, an initial codebook of a weighting codebook in MA prediction vector quantization is created based on the LBG algorithm using the initial value of the MA prediction coefficient, and each learning is performed by recursive learning. The elements of the weighted codebook and the MA prediction coefficient are sequentially updated so that the total sum of the distortions of the reproduction signal calculated for the data for use is minimized, and finally the elements of the weighted codebook and the MA prediction coefficient are determined. Thus, the weight codebook and the MA prediction coefficient can be set to more optimal ones. Therefore, even when the input audio signal is encoded with a low bit number, a high-quality reproduced audio signal can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a functional configuration of an encoder according to an embodiment of the present invention.

FIG. 2 is a block diagram illustrating a functional configuration of a distortion calculator in an encoder according to an embodiment of the present invention.

FIG. 3 is a block diagram showing a functional configuration of a driving excitation vector generation unit in the encoder according to the embodiment of the present invention.

FIG. 4 is a block diagram illustrating a functional configuration of a weight code generation unit in an encoder according to an embodiment of the present invention, and a configuration example of a weight codebook.

FIG. 5 is a block diagram illustrating a functional configuration of a decoder according to an embodiment of the present invention.

FIG. 6 is a flowchart illustrating a method of determining each element of a weight codebook and an MA prediction coefficient according to an embodiment of the present invention.

FIG. 7 is a flowchart illustrating a method of determining an initial value of a MA prediction coefficient according to an embodiment of the present invention.

FIG. 8 is a flowchart illustrating a method of determining an LSP codebook of an LSP quantizer according to the present invention.

FIG. 9 is a flowchart showing a conventional method of determining each element of a codebook and an MA prediction coefficient.

[Explanation of symbols]

 1-1 Linear prediction analysis unit 1-2 Linear prediction parameter encoding unit 1-3 Linear prediction parameter decoding unit 1-4 Drive excitation vector generation unit 1-5 Synthesis filter 1-6 ... Distortion calculator 1-8 Codebook search controller 1-9 Code transmitter 2-2, 2-3 Perceptual weight filter 2-4 Distance calculator 3-1 Adaptive code Book 3-2: Fixed codebook 3-3: Periodizing unit 3-4, 3-5: Multiplying unit 3-6: Addition unit 3-7: Weight code generation unit 4-1: Weight Codebook 4-2 MA prediction unit 4b1, 4b2, 4b (m-1) Buffer unit 4k1, 4k2, 4k3, 4km Vector multiplication unit 4a1, 4a2, 4a3, vector addition Unit 5-1 Code reception unit 5-2 Linear prediction parameter decoding unit 5-3 Driving excitation vector generation unit 5-4 Formed filter 5-5 ...... post-processing unit

Continuing on the front page (72) Inventor Kenichi Furuya 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo F-term in Nippon Telegraph and Telephone Corporation (reference) 5D045 CA01 CC07 DA11 5J064 AA01 BA13 BB01 BB03 BB14 BC12 BC17 BC27 BC29 BD02 BD03 BD04 9A001 CZ05 EE04 FF05 HH15 KZ56 (54) [Title of the Invention] Weighted codebook, method of creating the same, method of setting initial value of MA prediction coefficient during learning at the time of codebook design, and method of encoding acoustic signal and decoding thereof Method and computer-readable storage medium storing encoding program and computer-readable storage medium storing decoding program

Claims (19)

[Claims] 1. An input audio signal is separated into a short-term prediction signal indicating a prediction result of an envelope characteristic of a short-term frequency spectrum and a driving sound source signal indicating a prediction residual thereof. A driving excitation vector candidate is generated by selecting a weighting vector that defines the magnitude of the normalization vector and a weighting codebook that uses the weighting amount as each element, and generates the driving excitation vector candidate and the driving excitation vector candidate and the short-term The drive sound source signal is generated such that the distortion to the input sound signal of the reproduced sound candidate generated based on the prediction signal is minimized under a predetermined distance scale, and the drive sound source signal thus generated and In the method for creating the weight codebook in the encoding method for encoding the short-term prediction signal, each element of the weight codebook is set using a distance scale that takes into account human sound pressure perception characteristics. Weight codebook creation method comprising Rukoto. 2. A method according to claim 1, wherein a distance scale is used in which human sound pressure perception characteristics are added as a function of the power of the input sound signal. How to make. 3. The method according to claim 2, wherein when the input audio signal is x, the function is based on a value p in a range (0 <p <1) (| x | ² ) ^p
/ | X | ^{2, which} is given by: 4. A weighted codebook obtained by the method for creating a weighted codebook according to claim 1. 5. A method for setting an initial value of an MA prediction coefficient in MA prediction vector quantization, wherein an AR prediction coefficient calculated from a plurality of learning data is set to an MA prediction coefficient.
A method for setting an initial value of an MA prediction coefficient, wherein the initial value of the MA prediction coefficient is calculated by approximation using a prediction method. 6. A method for setting an initial value of a MA prediction coefficient in MA prediction vector quantization, comprising: calculating an average value of a plurality of learning data; and subtracting the average value from each learning data. Calculating the first AR prediction coefficient based on the covariance method using: a step of obtaining an impulse response of a filter using the first AR prediction coefficient as a filter coefficient; Calculating the second AR prediction coefficient of the same order as the MA prediction coefficient based on the method, and calculating the initial value of the MA prediction coefficient by obtaining an inverse filter of a filter using the second AR prediction coefficient as a filter coefficient And setting an initial value of the MA prediction coefficient. 7. A method for creating a weighting codebook for quantizing an MA prediction vector using the method for setting an initial value of an MA prediction coefficient according to claim 5, wherein the initial value of the MA prediction coefficient is calculated. Creating an initial codebook of a weighted codebook based on the LBG algorithm, and in the MA prediction vector quantization including the initial value of the MA prediction coefficient and the initial codebook, for each learning data by recursive learning. A step of sequentially updating each element of the weighted codebook and the MA prediction coefficient so as to minimize the sum of the calculated distortions of the reproduced signal, and finally determining each element. How to create a book. 8. The method according to claim 7, wherein the distortion of the reproduced signal is calculated using a distance scale that takes into account human sound pressure perception characteristics. . 9. The method according to claim 7, wherein the distortion of the reproduced signal is calculated using a distance scale taking into account a function of the power of the input audio signal. Method. 10. The method according to claim 7, wherein the distortion of the reproduced signal is calculated based on the power of the input acoustic signal x expressed by the following equation (1) based on a value p within a range (0 <p <1). A weighting codebook, which is calculated using a distance scale taking into account the function W. W = (| x | ² ) ^p / | x | ² (1) 11. A method according to claim 7, wherein the M code is obtained by the method according to any one of claims 7 to 10.
Weight codebook for A prediction vector quantization. 12. An input audio signal is separated into a short-term predicted signal indicating a prediction result of an envelope characteristic of a short-term frequency spectrum and a driving sound source signal indicating a prediction residual thereof, and are normalized based on the driving sound source signal. A driving excitation vector candidate is generated by selecting a weighting vector that defines the magnitude of the normalization vector and a weighting codebook that uses the weighting amount as each element, and generates the driving excitation vector candidate and the driving excitation vector candidate and the short-term The drive sound source signal is generated such that the distortion to the input sound signal of the reproduced sound candidate generated based on the prediction signal is minimized under a predetermined distance scale, and the drive sound source signal thus generated and In the audio signal encoding method for encoding the short-term prediction signal, each element of the weight codebook is set using a distance scale that takes into account human sound pressure perception characteristics. Coding method Hibiki signal. 13. The audio signal encoding method according to claim 12, wherein a distance scale is used in which human sound pressure perception characteristics are added as a function of the power of the input audio signal. Method. 14. The audio signal encoding method according to claim 13, wherein when the input audio signal is x, the function is based on a value p in a range (0 <p <1) (| x | ² ) ^p
/ | X | ^{2, which} is given by: 15. The method according to claim 12, wherein in the encoding method of the acoustic signal according to any one of claims 12 to 14, when a driving excitation vector candidate is generated from the driving excitation signal using MA prediction vector quantization. An audio signal encoding method using a weighted codebook and an MA prediction coefficient obtained by a method for creating a weighted codebook for predictive vector quantization. 16. An audio signal decoding method for decoding a code generated by the audio signal encoding method according to claim 12, wherein a weight used in these audio signal encoding methods is provided. A decoding method of an acoustic signal, comprising decoding the code using a codebook. 17. A decoding method of an audio signal for decoding a code generated by the audio signal encoding method according to claim 15, wherein the weight codebook and the MA prediction coefficient used in the audio signal encoding method are provided. A decoding method for an acoustic signal, comprising decoding the code using 18. A computer-readable storage medium storing an encoding program for processing according to the audio signal encoding method according to claim 12. Description: 19. A computer-readable storage medium storing a decoding program for processing according to the audio signal decoding method according to claim 16. Description: