JP3192051B2 - Audio coding device - Google Patents

Abstract

PURPOSE:To improve sound quality by a small arithmetic quantity even with a low bit rate by calculating the coefficient of a comb filter at every sub-frame and selecting a sound code vector minimizing a distortion while passing a voice signal through the comb filter. CONSTITUTION:An inputted voice signal is divided by a frame dividing circuit 110 and a sub-frame dividing circuit 120 and spectral parameters expressing spectral features of the voice signal are calculated in a spectral parameter calculating circuit 200 and then these parameters are quantized in a spectral parameter quantizing circuit 210. Then, an adaptive code book circuit 300 calculates pitch periods of the voice signal at every sub-frame and a filter circuit 320 calculates the gain of the filter of a prescribed degree having a delay corresponding to pitch periods at every frame or every sub-frame from a voice signal. Moreover, a sound source quantizing circuit 350 quantizes a sound source signal by selecting the best code vector from code vectors stored in a sound source code book 330.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech encoding system for encoding a speech signal at a low bit rate, particularly at a high quality of 4.8 kb / s or less.

[0002]

2. Description of the Related Art M. Schroeder, for example, describes a method of encoding a speech signal at a low bit rate of 4.8 kb / s or less.
"Code-excited linear predicti by BA Atal
on: High quality speech at low bit rates ”(Proc.
ICASSP, pp. 937-940, 1985) (referred to as “Reference 1”) and “Improved speechq” by Kleijn et al.
uality and efficient vector quantization in SELP ”
(Proc. ICASSP, pp. 155-158, 1988), a CELP (Code E) described in a paper (referred to as “Reference 2”), etc.
xcited LPC Coding) is known.

[0003] In this system, the transmitting side performs linear prediction (LP) from a speech signal every frame (for example, 20 ms).
C) Spectral parameters representing the spectral characteristics of the audio signal are extracted using analysis, the frame is further divided into subframes (for example, 5 ms), and a parameter in the adaptive codebook is determined for each subframe based on a past sound source signal. (Delay parameter and gain parameter), and the speech signal of the sub-frame is pitch-predicted by an adaptive codebook, and the residual signal obtained by pitch prediction is a sound source composed of a predetermined type of noise signal. An optimal sound source code vector is selected from a codebook (vector quantization codebook) and an optimal gain is calculated.

[0004] The method of selecting the excitation code vector is such that the error power between the signal synthesized from the selected noise signal and the residual signal is minimized.

[0005] Then, an index and a gain representing the type of the selected excitation code vector, the spectrum parameters and the parameters of the adaptive codebook are transmitted.

[0006] The decoding device on the receiving side synthesizes a speech signal based on the transmission code such as the index of the code vector, the gain, and the spectrum parameter transmitted from the encoding device on the transmitting side. It should be noted that the configuration of the decoding device is not directly related to the subject of the present invention, and a description thereof will be omitted.

[0007]

In the conventional speech coding systems of the above-mentioned references 1 and 2, when the bit rate is reduced, the size of the code book becomes smaller, and the sound quality of female sounds in particular deteriorates rapidly. was there.

In order to solve this problem, a method of performing comb filtering (meaning "filtering by a comb filter") on a sound source signal on a transmission side to enhance the pitch characteristics of the sound source signal to improve sound quality. Has been proposed.

The details of this method are described in, for example, “Improved Excitation for Phonetically-Segmented” by S. Wang et al.
VXC Speech Coding Below 4kb / s ”(Proc. GLOBECOM,
pp. 946-950, 1990) and the like (referred to as “Reference 3”).

[0010] However, when the method of Reference 3 is used, the sound quality can be improved in some cases. However, when searching both the adaptive codebook and the sound source codebook, all code vectors are subjected to comb filtering. However, since the amount of calculation becomes enormous and the gain of the comb filter is set to a constant value, the effect is not always obtained by the comb filtering, and there is a problem that the effect is deteriorated in some cases.

[0011] Accordingly, an object of the present invention is to solve the above-mentioned problems and to achieve a relatively small amount of calculation and a small amount of memory.
An object of the present invention is to provide a speech coding system with good sound quality at 8 kb / s or less.

[0012]

SUMMARY OF THE INVENTION The object of the present invention is to provide a method and apparatus for inputting an audio signal, dividing the input signal into frames of a predetermined time length, and converting the audio signal of the frame into a plurality of frames shorter in time than the frame. Using a past sound source signal for each subframe, a spectrum parameter calculation unit for obtaining a spectrum parameter representing a spectral feature of the audio signal, a spectrum parameter quantization unit for quantizing the spectrum parameter, an adaptive codebook section for obtaining the pitch period of the speech signal Te, the having a delay corresponding to a pitch period, a gain of the order of the filter to a predetermined, the audio signal or said for each of the frame or for each of the sub-frame Adaptive signal from audio signal
A filter unit that calculates a filter using the delay and the gain by calculating from the adaptive codebook residual signal obtained by subtracting the codebook signal, and at least a part of the code vector stored in the sound source codebook to the filter unit. An excitation quantizer for selecting a best code vector for the adaptive codebook residual signal and quantizing an excitation signal, the spectrum parameter quantization unit, the adaptive codebook unit, This is achieved by a speech encoding device (referred to as a “first viewpoint”) including a quantization unit and a multiplexer unit that combines and outputs respective output signals of the filter unit.

In a second aspect of the present invention , in a second aspect, an audio signal is input and divided into frames having a predetermined time length, and the audio signal of the frame is divided into a plurality of frames which are temporally shorter than the frame. And a spectrum parameter calculation unit for obtaining a spectrum parameter representing a spectral characteristic of the audio signal, a spectrum parameter quantization unit for quantizing the spectrum parameter, and a pitch period of the audio signal for each subframe. The adaptive codebook section to be obtained and the gain of a filter of a predetermined order having a delay according to the pitch period are calculated from the audio signal or adaptive codebook residual signal for each frame or each subframe. Filter, and the best code vector from the code vector stored in the sound source code book. A sound source quantization unit that quantizes a sound source signal through the filter unit after selecting a filter, the spectrum parameter quantization unit, the adaptive codebook unit, the sound source quantization unit, the filter unit, And a multiplexer unit that combines and outputs the respective output signals .

Further, the present invention provides, in a third aspect,
An audio signal is input and divided into frames of a predetermined time length, and the audio signal of the frame is divided into a plurality of subframes that are temporally shorter than the frame, and spectral parameters representing spectral characteristics of the audio signal. A spectrum parameter calculation unit for calculating, a spectrum parameter quantization unit for quantizing the spectrum parameter,
An adaptive codebook unit for determining a pitch cycle of a speech signal for each subframe, and a filter of a predetermined order having at least a part of the excitation code vector stored in the excitation codebook, having a delay corresponding to the pitch cycle. And quantizing at least one of the gain of the adaptive codebook unit, the gain of the excitation code vector, and the gain of the filter by the gain code vector selected from the gain code book, and the best excitation code vector And the output signal of the excitation / gain quantization unit for selecting the best gain code vector, the spectrum parameter quantization unit, the adaptive codebook unit, and the excitation / gain quantization unit. Codec having a multiplexer section To provide a device.

According to a fourth aspect of the present invention , in the fourth aspect, an audio signal is input, divided into frames having a predetermined time length, and the audio signal of the frame is divided into a plurality of frames which are temporally shorter than the frame. A spectrum parameter calculation unit that divides into subframes to obtain a spectrum parameter representing a spectral feature of the audio signal; a spectrum parameter quantization unit that quantizes the spectrum parameter; and an adaptation to obtain a pitch period of the audio signal for each subframe. After selecting the best code vector from the code vector portion and the code vector stored in the sound source code book, the filter is passed through a filter of a predetermined order having a delay according to the pitch period, and the gain of the filter is calculated. Source and gain quantization that uses the gain codebook to quantize the gain When, with the spectral parameter quantization unit, the code book portion, the sound source gain quantization section, each of the multiplexer to combine the output signal to the output of, it may have a.

The present invention provides, in first and third aspects,
All of the code vectors stored in the sound source codebook may be passed through the filter unit.

According to the present invention, in the third and fourth aspects, the sound source / gain quantizing unit preferably inputs a delay from the adaptive codebook unit, reads out a gain of the filter from the gain codebook, and A filter circuit that performs a filtering process on a sound source code vector read from a sound source code book based on the delay and the gain, and an output of the filter circuit, and a gain code vector from the gain code book and an adaptive code book unit. It comprises a distortion calculation circuit that quantizes the gain and the gain of the excitation code vector to calculate distortion, and a discrimination circuit that selects an optimal combination of the excitation code vector and the gain code vector that minimizes the distortion.

Further, in the present invention, the filter is constituted by either a non-recursive filter or a recursive filter, and is preferably constituted by a non-recursive filter having an order of 1 to 3.

[0019]

The principle and operation of the speech encoding apparatus according to the present invention will be described in detail below.

The audio signal is divided into frames (for example, 40 ms), and further divided into subframes (for example, 8 ms). Calculate and quantize spectral parameters representing spectral features of speech for each frame.

The adaptive codebook section calculates a delay corresponding to the pitch cycle of the speech for each subframe.

The filter section has the same delay as that of the adaptive code book section, and calculates the gain of a predetermined order M comb filter for each frame or for each subframe.

The comb filter has a recursive type (A
R type; "AR" stands for Auto Regressive) and non-recursive type (MA
Type; "MA" stands for Moving Average, but in the following, a non-recursive type ("MA type comb filter") will be used. The order M of the MA-type comb filter is 1.

In the following, the gain of the comb filter is calculated for each subframe from the residual signal obtained by pitch prediction by the adaptive codebook unit.

The sound source quantization section performs comb filtering on a part or all of the sound source code vectors stored in the sound source code book according to the following equation (1).

C _jz (n) = c _j (n) + η · c _j (n−T) (1)

In the above equation (1), c _j (n) is the j-th sound source code vector, and η is the gain of the MA type comb filter. T is the delay determined in the adaptive codebook.

Using the comb-filtered excitation code vector, one or more of the best excitation code vectors for minimizing the distortion of the following equation (2) are selected.

[0029]

(Equation 1)

In the above equation (2), x _w (n) is the perceptual weighting signal, β is the gain of the adaptive codebook section, v (n) is the adaptive code vector, and γ _j is the sound source code vector c _j (n). The optimal gain, h _w (n), is the impulse response of the auditory weighting synthesis filter. In the above equation (2), the operation symbol * of c _jz (n) * h _w (n) represents the convolution operation (convolution) of c _jz (n) and the impulse response h _w (n).

Here, as a relation equivalent to the above equation (2), the excitation code vector can be selected according to the following equation (3).

[0032]

(Equation 2)

Here, h ′ _w (n) = h _w (n) + η · h _w (n−T) (4)

In this case, the process of passing the sound source code vector through the comb filter is equivalent to modifying the impulse response h _w (n) to h ′ _w (n).

According to the present invention, in the second viewpoint, after selecting the best excitation code vector, the above equation (2) or (3)
Calculates the gain γ _j while comb filtering the sound source code vector.

According to a third aspect of the present invention, when the best excitation code vector is selected while performing comb filtering without selecting the gain of the comb filter in advance, the gain code vector is selected from the gain codebook. The quantization is performed while quantizing the gain, and a combination of a sound source code vector and a gain code vector that minimizes distortion is selected.

In the following, the gain of the adaptive codebook and
An example in which the gain of a sound source code vector and the gain of a comb filter are quantized collectively by a gain codebook will be described. The selection follows the following equation (5).

[0038]

(Equation 3)

Here, (β ′ _k , γ ′ _k , η ′ _k ) is the k-th gain code vector in the three-dimensional gain code book. However, k = 0~2 ^B -1, ^B is the number of bits gain codebook.

The above equation (5) is calculated for the combination of the gain code vector and the excitation code vector, and the combination that minimizes the distortion obtained by the above equation (5) is selected.

The selection may be made using the following equation (6) as an equation having an equivalent relation to the above equation (5).

[0042]

(Equation 4)

Where h ′ _wk (n) = h _w (n) + η ′ _k · h _w (n−T) (7)

According to the present invention, in the fourth aspect, after selecting the best excitation code vector without previously calculating the gain of the comb filter, the equation (2) or (3) is minimized. Then, the gain of the comb filter is calculated, and then the gain is quantized using the gain codebook.

As described above, according to the present invention, the coefficients of the comb filter are calculated for each sub-frame or frame, and distortion is minimized while passing all or a part of the excitation code vector through the comb filter. Select a sound source code vector to perform, select one type of sound source and then select a combination of a gain code vector and a sound source code vector through a comb filter, or select all or part of the sound source code vector through a comb filter to obtain a gain code vector. Or a combination of the sound source code vector and the sound source code vector, the sound quality can be improved with a relatively small amount of calculation even at a low bit rate.

[0046]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings.

[0047]

[Embodiment 1] FIG. 1 is a block diagram showing the configuration of an embodiment of a speech encoding system according to the present invention.

In FIG. 1, an audio signal is input from an input terminal 100, an audio signal is divided by a frame dividing circuit 110 for each frame (for example, 40 ms), and an audio signal of a frame is divided by a sub-frame dividing circuit 120 into frames. It is divided into short subframes (for example, 8 ms).

In the spectrum parameter calculation circuit 200,
The speech signal is cut out by applying a window (for example, 24 ms) longer than the subframe length to the speech signal of at least one subframe, and the spectral parameter is calculated in a predetermined order (for example, P = 10th order).

Since the spectral parameters greatly change with time in a transition section between a consonant and a vowel, it is desirable to analyze the spectral parameters every short time. However, doing so increases the amount of calculation required for the analysis. Therefore, here, the spectral parameters are calculated for any L (L> 1) subframes (eg, L = 3, first, third, and fifth subframes) in the frame.

Then, in the sub-frames not analyzed (here, the second and fourth sub-frames), the spectral parameters of the first and third sub-frames and the third and fifth sub-frames are respectively changed to the LSP (described later). Line spectrum pair)
The above linearly interpolated values are used as spectral parameters.

Here, the calculation of the spectrum parameters includes:
Known LPC analysis (linear predictive coding), Burg analysis, or the like can be used.

In this embodiment, Burg analysis is used. For details of Burg analysis, which is a spectrum estimation method based on the maximum entropy method (MEM), see the book entitled "Signal Analysis and System Identification" by Nakamizo (Corona Publishing Co., 1988), pp. 82-87 (" Reference 4 "
), And the description is omitted.

In the spectrum parameter calculation circuit 200,
Linear prediction coefficient α _i (i = 1) calculated by the Burg method
To 10) are converted into LSP parameters suitable for quantization and interpolation.

Here, the conversion from the linear prediction coefficients to the LSP is performed by a paper entitled "Speech Information Compression by Line Spectrum Pair (LSP) Speech Analysis / Synthesis Method" by Sugamura et al. (Transactions of IEICE, J64-A, pp. 599-606, 1981) (referred to as "Reference 5"). Note that the LSP obtains a spectrum by a line spectrum pair (Line Spectrum Pair) and increases the quantization efficiency on the frequency axis.

The linear prediction coefficients obtained by the Burg method in the first, third, and fifth subframes are converted into LSP parameters, and the LSPs of the second and fourth subframes are obtained by linear interpolation. The LSP is inversely transformed back to the linear prediction coefficients, and the linear prediction coefficients α _il (i = 1 to 10, l = 1 to 5) of the first to fifth subframes are perceptual weighting circuits
Output to 230. Also, the LSPs of the first to fifth subframes
Is output to the spectrum parameter quantization circuit 210.

The spectrum parameter quantization circuit 210 efficiently quantizes LSP parameters of a predetermined subframe.

In the following, it is assumed that vector quantization is used as a quantization method, and that the LSP parameter of the fifth subframe is quantized. A well-known method can be used for the method of vector quantization of LSP parameters.
A specific method is, for example, a series of inventions by the present inventor,
That is, JP-A-4-171500 (Japanese Patent Application No. 2-297600)
(Referred to as “Document 6”), Japanese Patent Application Laid-Open No.
3-261925) (referred to as "Reference 7") and JP-A-5-6199.
Publication (Japanese Patent Application No. 3-155949) (referred to as “Reference 8”) or “LSP Coding Using VQ-SVQ W” by T. Nomura et al.
ith Interpolationin 4.075 kbps M-LCELP Speech Code
r ”(Proc. Mobile Multimedia Communica
tions, pp.B.2.5, 1993) (referred to as “Reference 9”) and the like, and therefore, the description is omitted here.

The spectrum parameter quantization circuit 21
In 0, the LSP parameters of the first to fourth subframes are restored based on the LSP parameters quantized in the fifth subframe.

Here, the quantization LSP parameter of the fifth sub-frame of the current frame and the fifth
The LSPs of the first to fourth subframes are restored by linearly interpolating the quantized LSPs of the subframes.

Here, LSP before quantization and LSP after quantization
After selecting one type of code vector that minimizes the error power with respect to the SP, the LSPs of the first to fourth sub-frames can be restored by linear interpolation.

In order to further improve the performance, after selecting a plurality of code vectors for minimizing the error power, the cumulative distortion is evaluated for each candidate, and the candidate for minimizing the cumulative distortion and the interpolation LSP are interpolated. Can be selected. The details are described, for example, in Japanese Patent Application No. 5-8737 (referred to as “Document 10”) by the present inventor.

The LSPs of the first to fourth sub-frames and the quantized LSP of the fifth sub-frame, which have been reconstructed as described above, are assigned to the linear prediction coefficient α ′ _il (i = 1 to 10, l = 1 to
5), and outputs the result to the impulse response calculation circuit 310.

The index representing the code vector of the quantized LSP of the fifth sub-frame is
Output to 0.

In the above, instead of linear interpolation, LS
The interpolation pattern of P is determined by a predetermined number of bits (for example, 2 bits).
Bits), and a set of a code vector and an interpolation pattern that minimizes cumulative distortion by restoring LSPs of 1 to 4 subframes for each of these patterns may be selected.

In this way, the transmission information increases by the number of bits of the interpolation pattern. However, the temporal change in the LSP frame can be represented more precisely.

Here, the interpolation pattern may be created by learning in advance using LSP data for training, or a predetermined pattern may be stored.
As the predetermined pattern, for example, T. Taniguc
“Improved CELP speech coding at 4kb / sa” by hi et al.
nd below ”(Proc. ICSLP, pp.41-44, 19
92) (referred to as “Document 11”) and the like.

In order to further improve the performance, after selecting the interpolation pattern, an error signal between the true value of the LSP and the interpolation value of the LSP is determined in a predetermined subframe, and the error signal is calculated. Further, it may be represented by an error code book. For details, reference can be made to the aforementioned reference 9.

The perceptual weighting circuit 230 receives the linear prediction coefficient α _il (i = 1 to 10, l = 1 to 5) before quantization for each subframe from the spectrum parameter calculation circuit 200, and Based on the perceptual weighting of the audio signal of the sub-frame, the perceptual weighting signal x _w (n)
Is output.

The response signal calculation circuit 240 receives the linear prediction coefficient α _il for each subframe from the spectrum parameter calculation circuit 200 and
From 0, the linear prediction coefficient α ' _il restored by quantization and interpolation
Is input for each subframe, a response signal with the input signal d (n) = 0 is calculated for one subframe using the stored value of the filter memory, and output to the subtractor 250. Here, the response signal x _z (n) is represented by the following equation (8).

[0071]

(Equation 5)

Here, γ is a weight coefficient for controlling the amount of hearing weighting, and is the same value as the following equation (9).

The subtractor 250 subtracts the response signal by one subframe from the audibility weighting signal according to the following equation (9), and obtains x ′ _w (n)
Is output to the adaptive codebook circuit 300.

X ′ _w (n) = x _w (n) −x _z (n) (9)

The impulse response calculation circuit 310 calculates the impulse response h _w (n) of the weighting filter whose transfer function in the z-transform expression is represented by the following equation (10) by a predetermined number L, and calculates the adaptive codebook. The signal is output to the circuit 300 and the sound source quantization circuit 350.

[0076]

(Equation 6)

The adaptive codebook circuit 300 calculates a pitch parameter. For details, reference can be made to the aforementioned reference 2. Then, an index corresponding to the obtained delay value for each subframe is output to multiplexer 500.

Also, pitch prediction is performed by the adaptive codebook according to the following equation (11), and an adaptive codebook prediction difference signal z (n) is output. Further, it outputs an index representing the delay to multiplexer 500.

Z (n) = x ′ _w (n) −b (n) (11)

Here, b (n) is an adaptive codebook pitch prediction signal, and is given by the following equation (12).

B (n) = β · v (n−T) * h _w (n) (12)

Here, β and T represent the gain and delay of the adaptive codebook circuit 300, respectively. v (n) is an adaptive code vector. The operation symbol * represents a convolution operation (convolution).

The filter circuit 320 first calculates the gain η of the MA comb filter using the adaptive codebook pitch prediction difference signal z (n).

Specifically, for example, the gain η can be calculated so as to minimize the following equation (13). Note that MA
The order of the type comb filter is 1.

[0085]

(Equation 7)

Here, T is a delay obtained by the adaptive codebook circuit.

The sound source quantization circuit 350 optimizes the sound source code vector stored in the sound source code book 330 so as to minimize the above equation (2) or (3) for all or a part of the sound source code vectors. Select the code vector c _j (n).

At this time, one type of the best code vector may be selected, or two or more types of code vectors may be selected, and one type may be permanently selected at the time of gain quantization.
Here, it is assumed that two or more types of code vectors are selected.

When the above equation (2) or (3) is applied to only some of the excitation code vectors, a plurality of excitation code vectors are minimized so as to minimize the following equation (14). Equation (2) or (3) may be preliminarily selected and applied to the preselected excitation code vector.

[0090]

(Equation 8)

The gain quantization circuit 370 reads the gain code vector from the gain code book 340, and applies the excitation code vector and the gain code vector to the selected excitation code vector so as to minimize the expression (5). Select a combination.

The index representing the selected sound source code vector and gain code vector is converted to a multiplexer 500.
Output to Expression (6) can also be used as an expression having a relation equivalent to expression (5).

The weighting signal calculation circuit 360 inputs the output parameters of the spectrum parameter calculation circuit and the respective indexes, reads out the corresponding code vectors from the indexes, and firstly obtains the driving sound source signal v (n ).

V (n) = β ′ _k · v (n−T) + γ ′ _k · c _jz (n) (15)

Next, the spectrum parameter calculation circuit 200
Output parameter, spectrum parameter quantization circuit 21
The weighting signal s _w (n) is calculated for each subframe by the following equation (16) using the output parameter of 0, and is output to the response signal calculation circuit 240.

[0096]

(Equation 9)

[0097]

Embodiment 2 After selecting one best excitation code vector so as to minimize the expression (14) in the first embodiment, the excitation code vector is selected according to the expression (2) or (3). it may be the gain γj re-calculation.

When performing the quantization including the gain quantization by the gain codebook, equations (5) and (6) can be used instead of equations (2) and (3), respectively.

[0099]

Embodiment 3 FIG. 2 is a block diagram showing a configuration of a third embodiment of the present invention. The components having the same reference numerals as those in FIG. 1 perform the same operations as those in FIG.
Only the differences will be described below.

FIG. 3 shows a block diagram of the sound source / gain quantization circuit 400. In FIG. 3, a comb filter circuit 410
Then, the gain is not calculated in advance, the delay T from the adaptive codebook circuit 300 is input, and all the excitation code vectors or a part of the excitation code vectors are read out from the excitation codebook 330.
The gain η ′ _k of the comb filter is read from 340, and the following equation (1
7) Perform comb filtering according to
Output to Here, it is assumed that the gain codebook 340 is three-dimensional and stores the values of (β ′ _k , γ ′ _k , η ′ _k ).

C _jz (n) = c _j (n) + η ′ _k · c _j (n−T) (17)

[0102] In the distortion calculation circuit 420, further, reads the gain code vector from the gain codebook 340, while quantizes the gain of the gain and the sound source code vector of the adaptive codebook circuit 300, in accordance with the equation (5), the distortion D _j Is calculated and output to the determination circuit 430.

The discriminating circuit 430 selects the best combination of the excitation code vector and the gain code vector that minimizes the distortion D _j . As an equation having an equivalent relation to equation (5), equation (6) can be used instead of equation (5).

When performing comb filtering on some sound source code vectors, a plurality of code vectors must be selected in advance from sound source code book 330 so as to minimize equation (14). Can also.

[0105]

Embodiment 4 In the third embodiment, after selecting one of the best excitation code vectors from the code vectors stored in the excitation code book, the gain code book 340 is selected.
Is used to select the best combination of the excitation code vector and the gain code vector while quantizing the adaptive codebook, the excitation code vector, and the gain of the comb filter. Equation (5) or Equation (6) may be used to select the sound source code vector and the gain code vector .
No.

Although the present invention has been described with reference to the above embodiments, various modifications other than the above-described embodiments are possible without impairing the intention of the present invention.

As the spectrum parameters, other well-known parameters other than the LSP can be used.

In the spectrum parameter calculation circuit 200,
When calculating spectral parameters in at least one subframe in a frame, a change in RMS or a change in power between a previous subframe and a current subframe is measured, and a plurality of subframes in which these changes are large are measured. May be calculated. In this way, the spectral parameter is always analyzed at the voice change point, and performance degradation can be prevented even if the number of subframes to be analyzed is reduced.

For quantizing the spectral parameters, known methods such as vector quantization, scalar quantization, and vector-scalar quantization can be used.

Other well-known distance measures can be used for selecting an interpolation pattern in the spectrum parameter quantization circuit.

The delay in the adaptive codebook circuit and the filter circuit may be an integer value or a decimal value.

In the sound source quantization circuit, a case where the code book has one stage has been described, but a two-stage or multi-stage configuration is also possible.

Further, as a distance scale at the time of searching the sound source codebook and learning, or another learning scale, a learning method can be used.

The order of the comb filter circuit can be higher (for example, third order). By doing so, the amount of calculation and the amount of transmission information slightly increase, but the performance is further improved.

Although the non-recursive type (MA type) has been described as the configuration of the comb filter, a recursive type (AR type) can also be used.

Further, the gain codebook learns in advance a codebook having a size several times larger than the number of transmission bits in total, and assigns a partial area of the codebook as a use area for each predetermined mode. In addition, when coding, it is also possible to switch and use the used area according to the mode.

Further, the adaptive codebook section and comb filter can classify only a vowel part of a voice or a voice signal of a frame into a plurality of types of modes and apply the mode to only a predetermined mode.

Then, searching in the adaptive code book circuit,
In the search by the sound source quantization circuit, respectively, equation (2)
~ (6), convolution operation was performed using the impulse response h _w (n) as shown in equations (12) to (14).
The filtering can also be performed by using a weighting filter as represented by (10). By doing so, the amount of calculation increases, but the performance further improves.

[0119]

As described above, according to the present invention, the coefficients of the comb filter are calculated for each subframe or for each frame, and all or some of the excitation code vectors are passed through the comb filter. However, the sound source code vector that minimizes the distortion is selected, and the sound quality can be improved with a relatively small amount of calculation even at a low bit rate.

Further, according to the present invention, by selecting a combination of a gain code vector and a sound source code vector through a comb filter after selecting one kind of sound source,
Further, the amount of calculation is reduced and the sound quality is improved.

Further, according to the present invention, by selecting the best combination of the gain code vector and the excitation code vector for all or a part of the excitation code vector through a comb filter, the amount of operation can be reduced even at a low bit rate. This has the effect of reducing sound quality and improving sound quality.

Further, according to the present invention, by selecting one of the sound sources and then passing through a comb filter to select the best combination of the gain code vector and the sound source code vector, the amount of calculation is further reduced and the sound quality is improved. Is being planned.

In the present invention, when a first-order non-recursive filter is preferably used as a filter circuit, the amount of computation and the amount of transmitted information are significantly reduced, the sound quality is improved, and the order is reduced to the third order. In the case of, the amount of calculation and the amount of transmission information slightly increase, but the sound quality is further improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a third exemplary embodiment of the present invention.

FIG. 3 is a block diagram illustrating a configuration of a sound source / gain quantization circuit according to the present invention.

[Explanation of symbols]

 110 frame dividing circuit 120 subframe dividing circuit 200 spectral parameter calculating circuit 210 spectral parameter quantizing circuit 220 LSP codebook 230 weighting circuit 250 subtracting circuit 240 response signal calculating circuit 300 adaptive codebook circuit 310 … Impulse response calculation circuit 320… Filter circuit 350… Sound source quantization circuit 330… Sound source codebook 340… Gain codebook 360… Weighting signal calculation circuit 370… Gain quantization circuit 400… Sound source / gain quantization circuit 410… Com filter circuit 420: distortion calculation circuit 430: discrimination circuit 500: multiplexer

 ────────────────────────────────────────────────── ─── Continuation of front page (56) References JP-A-5-113799 (JP, A) JP-A-6-75597 (JP, A) JP-A-5-281998 (JP, A) 130977 (JP, A) JP-A-4-73700 (JP, A) JP-A-6-222797 (JP, A)

Claims (6)

(57) [Claims] An audio signal is input and divided into frames of a predetermined time length, and the audio signal of the frame is divided into a plurality of subframes that are temporally shorter than the frame. A spectrum parameter calculation unit for obtaining a spectrum parameter representing a feature, a spectrum parameter quantization unit for quantizing the spectrum parameter, and an adaptive codebook unit for obtaining a pitch period of a speech signal using a past sound source signal for each subframe. An adaptive codebook residual signal obtained by subtracting an adaptive codebook signal from the audio signal for each frame or each subframe, with a gain of a comb filter of a predetermined order having a delay according to the pitch period. And a filter unit that outputs the delay and the gain. Excitation quantization for selecting the best code vector for the adaptive codebook residual signal and quantizing an excitation signal while passing at least a part of the stored code vector through a comb filter having the delay and gain . Unit, the spectrum parameter quantization unit, the adaptive codebook unit, the sound source quantization unit, the filter unit,
And a multiplexer unit that combines and outputs each of the output signals. 2. An audio signal is input and divided into frames of a predetermined time length, and the audio signal of the frame is divided into a plurality of subframes that are temporally shorter than the frame. A spectrum parameter calculation unit for obtaining a spectrum parameter representing a feature; a spectrum parameter quantization unit for quantizing the spectrum parameter; an adaptive codebook unit for obtaining a pitch period of a speech signal for each subframe; and a sound source codebook. At least a part of the sound source code vector having a delay according to the pitch period, and passed through a comb filter of a predetermined order,
The gain of the adaptive codebook unit, the gain of the excitation code vector, and the gain of the comb filter are quantized by a gain code vector selected from a gain code book, and the best excitation code vector and the best gain code vector are selected. A sound source / gain quantization unit; a spectrum unit quantization unit; the adaptive codebook unit; and a multiplexer unit that combines and outputs respective output signals of the sound source / gain quantization unit. Characteristic speech coding device. 3. The speech encoding apparatus according to claim 1, wherein all of the code vectors stored in the excitation codebook are passed through the comb filter . 4. The sound source / gain quantizing unit receives a delay from the adaptive codebook unit, reads a comb filter gain from a gain codebook, and reads a sound source read from a sound source codebook based on the delay and gain. a comb filter for performing comb filtering on code vectors, and inputs the output of the comb filter, the gain from the gain codebook code vectors, the gain of the adaptive codebook section, the distortion of the gain of the sound source code vector by quantizing The speech encoding apparatus according to claim 2, comprising: a distortion calculation circuit for calculating; and a discrimination circuit for selecting an optimal combination of a sound source code vector and a gain code vector that minimizes the distortion. 5. The speech coding apparatus according to claim 1, wherein said comb filter is constituted by one of a non-recursive filter and a recursive filter. 6. The non-recursive filter having an order of 1 to 3 of the comb filter.
Or the speech encoding device according to 2.