JP2001013999A - Device and method for voice coding - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a voice coder in which coding distortion is made less noticeable even at a low bit rate. SOLUTION: By windowing first autocorrelation coefficients, that are computed in an autocorrelation computing section 101 from inputted voice signals, using a different shaped autocorrelation window in first and second windowing sections 102 and 103, corrected second and third autocorrelation coefficients are obtained. Using the second autocorrelation coefficients, spectrum parameters, that are to be coded in a coding section 106, are computed in a coding spectrum parameter computing section 104. Moreover, a hearing weighted spectrum parameter computing section 105 computes the spectrum parameters for hearing weighted characteristic setting using the third autocorrelation coefficients. Based on the parameters, a hearing weight setting section 107 sets the hearing weighted characteristic for the coding of residual components conducted by a coding section 108.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding method and apparatus for coding a speech signal with a low bit rate and a high efficiency by expressing the speech signal by spectral parameters and residual components.

[0002]

2. Description of the Related Art CEL is a low bit rate encoding method for storing and transmitting audio signals with a small amount of information.
P (Code Excited Linear Prediction, MRSchroeder
and BSAtal, “Code Excited Linear Prediction (C
ELP): High Quality Speech at Very Low Bit Rate
s ", Proc. ICASSP, pp. 937-940, 1985 (Reference 1). Also," Sound Communication Engineering ", edited by The Acoustical Society of Japan, 1996, Corona Corporation (Reference 2), pp. 33 -42
Also describes the CELP method.

[0003] The CELP system is a coding system based on linear prediction analysis. A speech signal is represented by a spectrum parameter representing a spectrum envelope serving as phoneme information and a residual component representing a pitch of a sound, and both are encoded. Become Although there are various spectral parameters representing the spectral envelope of the audio signal, LPC coefficients (linear prediction coefficients) are most commonly used in the field of audio coding.

In the CELP system, the LPC coefficient is obtained from the autocorrelation coefficient corrected by windowing the autocorrelation coefficient of the audio signal. In order to calculate the LPC coefficient from the autocorrelation coefficient, a method known as a Levinson-Durbin algorithm or a Durbin recursive solution is used. The details of this method are described in, for example, "Digital Speech Processing", Tokai University Press, pp. 75 of Sada Furui (Reference 3). The LPC coefficients obtained in this way are converted into equivalent parameters such as LSP coefficients suitable for encoding (see pp. 89-92 of Document 3). Then, by encoding this, the sign of the spectrum parameter is obtained.

[0005] On the other hand, in coding the residual component, code selection is performed using a distortion measure with an auditory weight so that encoding distortion is hardly heard. In a conventional speech coding technique such as the CELP scheme, L before encoding is used.
A feature is that the PC coefficient is also used for auditory weighting.

In decoding an audio signal, the code of the spectral parameter and the code of the residual component are decoded, and the audio signal is reproduced by giving a spectrum envelope to the decoded residual component according to the decoded spectral parameter. I do.

As described above, in the conventional speech coding technique, the LPC coefficient obtained mainly for the purpose of encoding is also used for setting the auditory weighting characteristic. The auditory weight characteristics cannot be expressed. Therefore, if a conventional speech coding technique is used for low bit rate coding of, for example, about 4 kbit / s or less, deterioration of a residual component having a large influence on coding distortion cannot be completely masked by perceptual weighting, and high-quality Decoded speech cannot be obtained.

[0008]

As described above, in the conventional speech coding technique, the LPC coefficients obtained mainly for the purpose of coding are diverted to the setting of the auditory weighting characteristics. Since sufficient perceptual weight characteristics cannot be expressed, when trying to further reduce the bit rate, the deterioration of the residual component with large coding distortion cannot be completely masked by perceptual weighting, and the quality of decoded speech deteriorates. There was a problem.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech encoding method and apparatus in which encoding distortion is hardly perceived while reducing the bit rate.

[0010]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention represents an input speech signal by a spectrum parameter representing a spectrum envelope and a residual component, and a speech encoding the spectrum parameter and the residual component. At the time of encoding, a spectrum parameter is calculated and encoded from a second autocorrelation coefficient obtained by correcting a first autocorrelation coefficient obtained from an input speech signal, and a first autocorrelation coefficient is calculated. Is corrected under a condition different from the condition under which the second autocorrelation coefficient is obtained, an auditory weighting characteristic is obtained from the third autocorrelation function, and a residual component is obtained by using these spectral parameters and the auditory weighting characteristic. Encoding is a basic feature.

Here, the correction of the autocorrelation coefficient is performed using, for example, an autocorrelation window. By performing windowing on the first autocorrelation coefficient using an autocorrelation window, a corrected second or third autocorrelation coefficient is obtained. In this case, the first autocorrelation window used for obtaining the second autocorrelation coefficient and the second autocorrelation window used for obtaining the third autocorrelation coefficient have different shapes.

More specifically, in the present invention, a first autocorrelation coefficient is calculated for each predetermined time unit from an input audio signal. For the first autocorrelation coefficient, the first window
The second autocorrelation coefficient is obtained by performing windowing using the autocorrelation window of the second autocorrelation window. Similarly, a second autocorrelation window having a shape different from the first autocorrelation window in the second windowing portion is obtained. The third autocorrelation coefficient can be obtained by performing windowing using this.

A first spectral parameter to be encoded is calculated using the second autocorrelation coefficient.
Are encoded. On the other hand, another second spectral parameter is calculated using the third autocorrelation coefficient, an auditory weighting characteristic is set from the second spectral parameter, and the remaining auditory weighting characteristic is set using the first spectral parameter and the auditory weighting characteristic. The difference component is encoded.

According to the present invention, the first autocorrelation window has a shape optimized for obtaining a first spectral parameter (for example, LPC coefficient) to be encoded, and the second autocorrelation window is formed. By adopting a shape optimized for obtaining the second spectral parameter used for setting the hearing weight characteristic, it becomes possible to accurately obtain each of the first spectral parameter to be encoded and the hearing weight characteristic. . Therefore, even at a very low coding bit rate, coding distortion is hardly perceived during decoding,
Audio encoding that can reproduce high-quality decoded audio becomes possible.

[0015]

Embodiments of the present invention will be described below with reference to the drawings.

(First Embodiment) FIG. 1 shows a first embodiment of the present invention.
It is a block diagram showing the composition of the speech coding device concerning an embodiment. This speech coding apparatus includes an autocorrelation calculating unit 10
1, first windowing section 102, second windowing section 103, coding spectrum parameter calculating section 104, perceptual weighting spectrum parameter calculating section 105, spectrum parameter coding section 106, perceptual weight setting section 107, residual component It comprises an encoding unit 108 and a multiplexing unit 109.

The autocorrelation calculating section 101 obtains a first autocorrelation coefficient ri (r0, r1) from an input audio signal sampled and digitized at a predetermined sampling frequency for each predetermined time unit as shown in the following equation. ,..., RN) are calculated.

[0018]

(Equation 1)

Here, {x _n } is the length L of the input audio signal.
, N represents the order of the autocorrelation, and a typical value of N is N = 10 when the sampling frequency of the input audio signal is 8 kHz.

Next, the first autocorrelation coefficient ri obtained by the autocorrelation calculation section 101 is applied to the first windowing section 102
, The first autocorrelation coefficient ri is corrected, and the second autocorrelation coefficient φi (φ0, φ1,..., ΦN) is obtained. An example of the windowing process using the first autocorrelation window is shown in the following equation. φi = ri × wi (i = 0, 1,..., N) (2) where wi represents a first autocorrelation window.

Next, the spectrum parameter to be coded is obtained in the coding spectrum parameter calculating section 104 using the second autocorrelation coefficient φi. Power spectrum, LPC as spectrum parameters
Various types such as cepstrum, mel scale spectral parameter, and subband energy are known. Here, examples of LPC coefficients (linear prediction coefficients) will be described.
The LPC coefficient is calculated by solving the following linear equation. Φα = ψ (3) where Φ is a second autocorrelation coefficient φ as shown in the following equation.
It is an autocorrelation matrix composed of i.

[0022]

(Equation 2)

From the equation (3), the LPC coefficient {αi}
For example, as a method for obtaining
thm and Durbin's recursive method can be used. 75, detailed description is omitted.

The spectrum parameter to be coded (hereinafter referred to as a coding spectrum parameter) (here, LPC coefficient {αi}) thus obtained is coded and quantized by the spectrum parameter coding unit 106. The spectrum parameter and the symbol A representing the spectrum parameter are output.

In coding the spectral parameters, for example, when the spectral parameters are LPC coefficients, the LPC coefficients are converted into equivalent parameters such as LSP (line spectrum pair) coefficients (see pp. 89-92 of Reference 3). ), By encoding this using the vector quantization method, it is possible to encode spectral parameters with less quantization distortion under the same number of bits.

Next, in order to set the auditory weighting characteristic from the autocorrelation function, first, the second windowing section 103 applies a second windowing section 103 to the first autocorrelation coefficient ri obtained by the autocorrelation calculating section 101. By windowing with the autocorrelation window,
The first autocorrelation coefficient ri is corrected, and the third autocorrelation coefficient φ′i (φ′0, φ′1,..., Φ′N) is obtained. An example of the windowing process using the second autocorrelation window is as follows.
It is shown by the following equation. φ′i = ri × vi (i = 0, 1,..., N) (5) Here, vi represents a second autocorrelation window.

The second autocorrelation window vi is a window used for setting the auditory weighting characteristic, and the first autocorrelation window wi is used.
Is different in shape. More specifically, the two autocorrelation windows vi have a relationship such that the degree of correction given to the autocorrelation coefficient is smaller in the second autocorrelation window vi than in the first autocorrelation window wi. It is desirable to set windows wi and vi. The reason is as follows.

First, regarding the first autocorrelation window wi,
The spectral parameters for encoding finally determined by the spectral parameter computing unit for encoding 104 using this are the filter characteristics of a synthesis filter for generating a speech signal after being quantized in the spectral parameter encoding unit 106. Therefore, it is desirable to use a window shape in which the degree of correction to the autocorrelation coefficient is relatively strong so as not to have an excessively strong resonance point in the frequency characteristic.

On the other hand, since the second autocorrelation window vi is used to set an auditory weighting characteristic for reflecting a frequency masking effect corresponding to the shape of the spectrum of the audio signal, an excessively strong resonance point Although it is necessary to prevent the first autocorrelation window wi from being modified, the window shape is not used as a filter characteristic of the synthesis filter. It is desirable to do.

Next, in the auditory weighting spectrum parameter calculating section 105, the third autocorrelation coefficient φ′i obtained by the second windowing section 103 is used to set the spectral parameters necessary for setting the auditory weights (hereinafter, referred to as the following). , A spectral parameter for auditory weight). When the LPC coefficient is used as the auditory weighting spectrum parameter, the LPC coefficient used as the above-described encoding parameter is used.
It goes without saying that the same algorithm as the coefficient calculation method can be used. The linear equation at this time is as follows. Φ′β = ψ ′ (6) Here, Φ ′ is an autocorrelation matrix composed of a third autocorrelation coefficient φ′i as shown in the following equation.

[0031]

(Equation 3)

Since the second auto-correlation window vi is different from the first auto-correlation window vi, the auditory weight calculated using the third auto-correlation function φ′i modified by the second auto-correlation window vi Coefficient ｛β as spectrum parameter for
i｝ has a different spectral characteristic from the LPC coefficient {αi} as the coding spectrum parameter corrected by the first autocorrelation window vi. Therefore, by setting the second autocorrelation window vi appropriately for the auditory weight, there is an effect that a more accurate auditory weight characteristic can be used for encoding the residual component.

The perceptual weight setting unit 107 sets perceptual weight characteristics used for perceptual weighting in the residual component coding unit 108 using the perceptual weight spectral parameters (in this example, LPC coefficients {βi}). Residual component encoding unit 10
In the case where the perceptual weighting is performed in the time domain at 8 to encode the residual component, the perceptual weighting is realized as a filtering process using a weight filter having a characteristic of W (z). A typical example of the auditory weighting filter characteristic W (z) using the LPC coefficient {βi} is represented by the following equation.

[0034]

(Equation 4)

Here, B (z) is given by the following equation.

[0036]

(Equation 5)

Γ1 and γ2 are parameters for setting the auditory weighting characteristics in the residual component encoding unit 108, where 1 ≧ γ
A relationship of 1>γ2> 0 is required. As a typical example, for example, γ1 = 0.94 and γ2 = 0.6 can be used.

The residual component encoding unit 108 receives the input speech signal, the quantized spectral parameters from the spectral parameter encoding unit 106 and the information of the perceptual weight,
The residual component necessary to represent the audio signal is encoded together with the quantized spectral parameters, and the code B of the obtained residual component is output.

The code A of the spectrum parameter obtained by spectrum parameter coding section 106 as described above
And the code B of the residual component obtained by the residual component encoding unit 108 are multiplexed by the multiplexing unit 109 and output as encoded data representing the input audio signal. This encoded data is
It is sent to the storage system or transmission system.

Next, with reference to the flowchart shown in FIG. 2, a description will be given of a processing procedure when the same voice encoding processing as that of the voice encoding apparatus according to the present embodiment is realized by software.

First, a first autocorrelation coefficient ri (r0, r1,..., RN) is obtained for each predetermined time unit from the input audio signal (step S1). Next, this autocorrelation coefficient ri
Is windowed with a first autocorrelation window wi (w0, w1,..., WN), and the corrected second autocorrelation coefficient φi
(Φ0, φ1,..., ΦN) are obtained (step S2).
Next, a coding spectrum parameter to be coded is obtained using the second autocorrelation coefficient φi (step S
3). Next, the encoding spectral parameters are encoded, and the quantized spectral parameters obtained in the encoding process and the sign of the spectral parameters representing the quantized spectral parameters are obtained (step S4).

On the other hand, processing from setting of the first autocorrelation function ri obtained in step S1 to setting of auditory weighting characteristics is performed as follows. That is, the autocorrelation coefficient r
i is windowed with a second autocorrelation window vi (v0, v1,..., vN), and the corrected third autocorrelation coefficient φ ′
i (φ′0, φ′1,..., φ′N) are obtained (step S5). Next, using the third autocorrelation coefficient φ′i, a perceptual weight spectrum parameter required for setting the perceptual weight is determined (step S6). Next, using the perceptual weight spectrum parameters, perceptual weight characteristics used in residual component coding are set (step S7). Next, using the input audio signal, the quantized spectral parameters, and the information on the auditory weighting characteristics, the residual component necessary for representing the audio signal together with the quantized spectral parameters is encoded (step S8). . Then, steps S4 and S8
Are multiplexed with the code of the spectrum parameter and the code of the residual component obtained by the above processing and output as code data of the audio signal (step S9).

When the processing in steps S1 to S9 is completed, one time unit (typically, when the input audio signal is 8k
The encoding process of the audio signal of 20 msec when sampling at Hz is completed. By continuously performing this series of processing for each time unit until it is determined in step S10 that the processing for the next time unit is not to be performed, it is possible to encode a continuously input audio signal.

(Second Embodiment) FIG. 3 shows that the present invention
It is a block diagram which shows the structure of the speech coding apparatus applied to the LP system. In this figure, the residual component encoding unit, which is a feature of the CELP method, is shown in more detail than FIG. Details of the CELP method are described in References 1 and 2 as described above.

This speech coding apparatus includes an autocorrelation calculating section 30.
1, first windowing section 302, second windowing section 303, encoding LPC coefficient calculating section 304, auditory weighting LPC coefficient calculating section 305, LPC coefficient encoding section 306, auditory weight setting section 307, residual component Encoding section 308 and multiplexing section 309
Consists of

Here, autocorrelation calculation section 301, first windowing section 302, second windowing section 303, encoding LPC coefficient calculating section 304, auditory weighting LPC coefficient calculating section 305, L
The PC coefficient encoding unit 306 and the perceptual weight setting unit 307 are based on the autocorrelation calculation unit 10 in the first embodiment.
1, the first windowing unit 102, the second windowing unit 103, the coding spectrum parameter calculation unit 104, the hearing weight spectrum parameter calculation unit 105, the spectrum parameter coding unit 106, and the hearing weight setting unit 107 are the same. Therefore, the description is omitted.

The residual component encoder 308 includes a target signal generator 311, an adaptive excitation encoder 312, and a noise excitation encoder 3
13, a gain encoder 314, a drive signal generator 315, and a weighted synthesis filter 316. Hereinafter, the configuration of each unit of the residual component encoding unit 308 will be described in detail.

The target signal generating section 311 has an auditory weight filter for which an auditory weight characteristic is set by the auditory weight setting section 307. The auditory weighting is performed by filtering the input speech signal using the auditory weight filter. The target signal {fn} which is the target of the residual component encoding is generated by generating the audio signal thus obtained and subtracting the influence of the encoding in the previous time unit from the audio signal having the auditory weight. Generate.

Adaptive excitation coding section 312 has an adaptive codebook well-known in CELP speech coding, and uses target signal {fn} (target vector f) to generate error vector e0 of the following equation. The optimal adaptive code vector c0 to be reduced in size, preferably minimized, is searched in the adaptive codebook. e0 = f−Hwc0 (i) (10) Here, i indicates an index of a code vector that is a candidate for an adaptive code vector. Hw is an impulse response matrix composed of an impulse response of a filter having a spectral envelope characteristic of a hearing-weighted voice (a characteristic of a hearing-weighted synthesis filter) Hw (z).

The perceptually weighted spectral envelope characteristic H
w (z) is represented by the following equation.

[0051]

(Equation 6)

Here, W (z) is the perceptual weight filter characteristic shown in equation (4), and Aq (z) is expressed by the following equation.

[0053]

(Equation 7)

Here, α _qi is a quantized LPC coefficient.

The adaptive code vector index I selected from the adaptive code vector candidates and the adaptive code vector c0 (I) corresponding thereto are output from the adaptive excitation coding section 312.

Next, in the noise excitation coding section 313, a noise excitation book or a pulse excitation source capable of expressing noise in a pseudo manner is constructed by a predetermined method well known in speech coding of the CELP system. The components that cannot be represented by the adaptive excitation coding unit 312 are coded using the above-described method.
The target vector d used at this time is d = fc0 (I)
It can be. Using this target vector d, an optimal noise code vector c1 that makes the magnitude of the error vector e1 of the following equation smaller, preferably minimized, is searched from the noise code vector candidates. e1 = d−Hwc1 (j) (13) Here, j indicates an index of a code vector that is a candidate of a noise code vector.

The noise code vector index J selected from the random code vector candidates and the corresponding noise code vector c1 (J) are output from the noise excitation coding section 311.

Next, gain encoding section 314 performs CELP
The adaptive excitation coding unit 312 has a gain codebook configured by a predetermined method well-known in the speech coding of the system.
And the noise code vector c1 output from the noise excitation coding unit 313.
(J) is encoded with a gain to be multiplied. At the time of encoding, a gain vector candidate g stored in a gain encoding book is used to reduce or preferably minimize the size of an error vector eg represented by the following equation.
Search is performed from 0 (k) and g1 (k) (where k is an index of a gain vector). eg = f−g0 (k) Hwc0 (I) −g1 (k) Hwc1 (J) (14) Thus, gain vector candidates g0 (k) and g1 (k)
Are output from the gain encoding unit 314. The gain index K searched for among the above and the corresponding gain vectors g0 (K) and g1 (K) are output.

Adaptive coded vector c0 (I) output from adaptive excitation coding section 312, noise excitation coding section 313
And a gain vector g0 output from the gain encoding unit 314.
(K) and g1 (K) are input to the drive signal generation unit 315. The drive signal generation unit 315 calculates the adaptive code vector c0 (I) and the noise code vector c1 (J) as shown in the following equation.
Are multiplied by gain vectors g0 (K) and g1 (K), respectively, and then added to obtain a quantized residual vector ex. This residual vector ex is input to adaptive excitation coding section 312 and stored in the adaptive codebook, and is also input as a driving signal to weighted synthesis filter 316.

Ex = g0 (K) c0 (I) + g1 (K) c1 (J) (15) Finally, the residual vector ex and the characteristics W (z) and Aq (z) of the weighted synthesis filter are used. Then, the internal state of the weighted synthesis filter for obtaining the influence on the next time unit encoding of the input audio signal is obtained, and this is supplied to the target signal generation unit 311.

Finally, the index A of the adaptive code vector corresponding to the code A of the spectral parameter (LPC coefficient) obtained as described above, the code B of the residual component in FIG. The multiplexing unit 309 multiplexes the gain vector index K with the gain vector and outputs the coded data representing the input audio signal. This encoded data is sent to a storage system or a transmission system.

Next, the speech decoding apparatus according to this embodiment will be described. FIG. 4 is a block diagram showing a configuration of a speech decoding device corresponding to the speech encoding device shown in FIG. 3 according to the embodiment.

The present invention basically has a feature in a method of extracting a spectrum parameter (for example, LPC coefficient) and a residual component on the encoding side, and codes output from the speech encoding apparatus shown in FIG. The coded data itself is basically the same as that of the conventional CELP system. Therefore, the configuration of the speech decoding device may be the same as that of the conventional CELP system.

The speech decoding apparatus shown in FIG.
0, LPC coefficient decoding section 401, adaptive excitation decoding section 40
2, noise excitation decoding section 403, gain decoding section 404,
The driving signal generator 405 includes a synthesis filter 406 and a post filter 407.

The separation unit 400 converts the coded data input from the speech coding apparatus shown in FIG. 3 through the storage system or the transmission system into a code A of a spectrum parameter (LPC coefficient) and a code of a residual component. The corresponding index I of the adaptive code vector, the index J of the noise code vector, and the index K of the gain vector are separated, and the LPC coefficient decoding unit 401 and the adaptive excitation decoding unit 40, respectively.
2. Noise excitation decoding section 403 and gain decoding section 40
4 is input.

The LPC coefficient decoding section 401 reproduces the quantized LPC coefficient corresponding to the code of the spectrum parameter A in the same manner as in the speech coding apparatus, and supplies this to the synthesis filter 406 and the post filter 407.

Adaptive excitation decoding section 402 has an adaptive codebook like adaptive excitation encoding section 312 in FIG. 3, and determines adaptive code vector c0 (I) corresponding to index I to drive signal generation section 405. To supply. Noise source decoding unit 4
03 has a noise codebook similarly to the noise excitation coding unit 313 of FIG. 3, finds a noise code vector c1 (J) corresponding to the index J, and supplies it to the drive signal generation unit 405. Further, gain decoding section 404 has a gain codebook similarly to gain encoding section 314 in FIG. 3, and gain vectors g0 (K) and g1 (K) corresponding to index K.
Is supplied to the drive signal generation unit 405.

The drive signal generation unit 405, like the drive signal generation unit 315 of FIG. 3, uses the adaptive code vector c0 (I), the noise code vector c1 (J), and the gain vectors g0 (K), g1 according to equation (15). A quantized residual vector ex is obtained from (K). This residual vector ex is
The signal is input to adaptive excitation decoding section 402 and stored in adaptive codebook, and is also input to synthesis filter 406 as a drive signal.

Using the quantized LPC coefficient α _qi obtained by the LPC coefficient decoding section 401, the synthesis filter 406 filters the inverse characteristic 1 / Aq (z) with the driving signal (residual By performing on the vector ex), the decoded audio signal is synthesized. The output signal of the synthesis filter 406 is finally subjected to the spectral shape enhancement by the post filter 407 whose characteristics are set using the quantized LPC coefficient α _qi obtained by the LPC coefficient decoding unit 401, so that the final output signal is obtained. A decoded audio signal is generated.

Although the embodiment of the present invention has been described above, the present invention is not limited to this, and can be implemented with various modifications. For example, in the above embodiment, a method of multiplying an autocorrelation coefficient by an autocorrelation window and performing windowing processing has been described as an example of a method of correcting an autocorrelation coefficient. It is not limited to. In short, the autocorrelation coefficient used for the spectral parameter to be coded and the autocorrelation coefficient used for setting the auditory weighting characteristics are modified under different conditions suitable for each, and the common autocorrelation coefficient starts from the common autocorrelation coefficient. Any method can be used as long as it can be obtained.

The present invention is also applicable to the case where the definition of the autocorrelation coefficient is slightly different from that described in the above embodiment, and the case where the normalized autocorrelation coefficient is used instead of the autocorrelation coefficient. It goes without saying that you can do it.

[0072]

As described above, according to the present invention, the second and third autocorrelation coefficients obtained by correcting the first autocorrelation coefficient obtained from the input speech signal under different conditions are used. By individually calculating the spectral parameter to be encoded and the auditory weighting characteristic used for encoding the residual component, it is possible to accurately determine both the spectral parameter and the auditory weighting characteristic of the encoding target. .

Therefore, according to the present invention, even in a low bit rate encoding of about 4 kbit / s or less,
It is possible to realize speech encoding that can obtain high-quality decoded speech in which encoding distortion is hardly perceived.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech encoding device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a speech encoding processing procedure according to the second embodiment;

FIG. 3 is a block diagram showing a configuration of a speech coding apparatus according to a second embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a speech decoding device corresponding to the speech encoding device according to the second embodiment;

[Explanation of symbols]

101, 301: autocorrelation calculation units 102, 302: first windowing units 103, 303 ... second windowing units 104, 304 ... coding spectrum parameter calculation units 105, 305 ... hearing weight spectrum parameter calculation units 106 ... Spectrum parameter encoder 306 LPC coefficient encoder 107, 307 Perceptual weight setting unit 108, 308 Residual component encoder 109, 309 Multiplexer

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5D045 CA01 CB01 5J064 BA10 BB03 BC02 BC03 BC11 BD02 5K041 AA04 CC01 EE23 HH21 JJ14 JJ21

Claims (4)

[Claims] 1. A speech coding method for representing an input speech signal by a spectrum parameter representing a spectrum envelope and a residual component, and encoding the spectrum parameter and the residual component, wherein a first parameter obtained from the input speech signal is obtained. Calculating and encoding the spectrum parameter from a second autocorrelation coefficient obtained by correcting the autocorrelation coefficient of A sound weighting characteristic is obtained from a third autocorrelation function obtained by correction under a condition different from the above condition, and the residual component is encoded using the spectrum parameter and the hearing weighting characteristic. Encoding method. 2. A speech encoding method for representing an input speech signal by a spectrum parameter representing a spectrum envelope and a residual component, and encoding the spectrum parameter and the residual component, wherein a first parameter obtained from the input speech signal is obtained. The spectrum parameter is calculated and encoded from a second autocorrelation coefficient obtained by correcting the autocorrelation coefficient of the first autocorrelation window using a first autocorrelation window, and the first autocorrelation coefficient is calculated by the first autocorrelation window. A perceptual weighting characteristic is obtained from a third autocorrelation function obtained by using a second autocorrelation window different from the autocorrelation window, and the residual component is encoded using the spectral parameter and the perceptual weighting characteristic. Speech encoding method. 3. A speech encoding method for representing an input speech signal by a spectrum parameter representing a spectrum envelope and a residual component, and encoding the spectrum parameter and the residual component, comprising the steps of: Calculating a first autocorrelation coefficient; and performing a windowing on the first autocorrelation coefficient using a first autocorrelation window to obtain a second autocorrelation coefficient. Calculating a first spectral parameter using the second autocorrelation coefficient; encoding the first spectral parameter; and forming a shape with respect to the first autocorrelation coefficient. Calculating a third autocorrelation coefficient by performing windowing using a second autocorrelation window different from the first autocorrelation window; and using the third autocorrelation coefficient. Calculating a second spectral parameter; setting an auditory weighting characteristic based on the second spectral parameter; encoding the residual component using the first spectral parameter and the auditory weighting characteristic Voice coding method. 4. A speech encoding apparatus for representing an input speech signal by a spectrum parameter representing a spectrum envelope and a residual component, and encoding the spectrum parameter and the residual component. An autocorrelation calculating means for calculating a first autocorrelation coefficient, and windowing the first autocorrelation coefficient using a first autocorrelation window, thereby obtaining a second autocorrelation coefficient. , A first spectral parameter calculating means for calculating a first spectral parameter using the second autocorrelation coefficient, and a first spectral parameter calculating means for calculating the first spectral parameter. A spectrum parameter encoding means for encoding a spectrum parameter; a second parameter having a shape different from the first autocorrelation window with respect to the first autocorrelation coefficient. A second windowing means for obtaining a third autocorrelation coefficient by performing windowing using an autocorrelation window; and a second windowing means for calculating a second spectral parameter using the third autocorrelation coefficient. A second spectral parameter calculating means, an auditory weight characteristic setting means for setting an auditory weight characteristic based on the second spectral parameter, and an auditory weight characteristic set by the first spectral parameter and the auditory weight setting means. And a residual component encoding unit for encoding the residual component using the residual component.