JP3274451B2 - Adaptive postfilter and adaptive postfiltering method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0002] The present invention relates to a speech coding method for compressing a speech signal or the like with high efficiency, and more particularly to an adaptive post-filter and an adaptive post-filtering method used on the decoding side to improve the subjective quality of a synthesized speech signal.

[0003]

2. Description of the Related Art As a conventional method for encoding a speech signal at a transmission rate of about 10 kbit / s or less, a CELP (Code Excit
ed Linear Prediction) system is known.

The CELP method quantizes a prediction residual signal obtained when an input signal is predicted by a prediction filter at a level of a synthesized speech signal weighted by auditory perception, so that noise included in an output synthesized speech signal is reduced. This is a speech coding method characterized in that it can be reduced. A scheme based on this scheme has been widely used in recent years because it can perform better speech coding than other schemes (for example, this content is described in "MR-Schroeder and BSAtal," Code-Excited Line
ar Prediction (CELP): high quality speech at v
ery low bit rates, "Proc. IEEE Int. Conf. Acous
t., Speech, SignalProcessing, pp. 937-940 (1985)
."It is described in. Normally, in the CELP system, a composite synthesized speech signal is finally output via an adaptive post filter. This adaptive post filter has an effect of suppressing the noise included in the synthesized speech signal by enhancing the spectrum envelope and the repetition of the pitch by appropriately setting the parameters, and is used for improving the subjective quality.

[0005] This adaptive postfilter suppresses noise and increases the power of the output audio signal, so that the postfilter has conventionally been used so that the power of the input signal and the output signal of the postfilter become the same in subframe units. Is multiplied by a gain to control the power of the output signal.

However, in the above-described power control, discontinuous points occur in the output signal between blocks with respect to a voice whose power greatly changes, and as a result, the quality of the post-filtered output voice signal deteriorates. There are drawbacks.

FIG. 15 shows "P. Kroon and EF Deprettere,
"A Class of Analysis-by-Synthesis Predictive
Coders for High Quality Speech Coding at Rates Bet
ween4.8 and 16kbits / s ", IEEE SAC-G.February, 198
FIG. 8 shows a block diagram of a commonly used adaptive postfilter as described in 8. pp. 353-363. CE
In the LP scheme, the post filter is represented by a continuous combination of both an LPC synthesis filter representing the decoded spectrum envelope and a pitch synthesis filter representing the fine structure of the spectrum,
It is updated and used in subframe units.

[0008]

As described above, the conventional adaptive postfilter multiplies the gain of the output signal of the postfilter in subframe units to make the power of the input signal and the output signal in the subframe equal. The power control of the output audio signal is performed as described above. For this reason, the output audio signal waveform that is finally multiplied by the gain and output is likely to be discontinuous in subframe units for an audio signal with a large power change. There is a problem that new quality deterioration is caused by the above.

SUMMARY OF THE INVENTION The present invention has been made in view of such a problem, and has an adaptive post-filter capable of obtaining a high-quality output audio signal having a continuous waveform and a small amount of noise without depending on a power change of an input quality. The present invention relates to an adaptive post-filtering method.

[0010]

The present invention provides means for generating a post filter based on LPC coefficients for an audio signal, means for calculating a post filter gain from a signal after post filter processing and a signal before post filter processing, Means for post-filtering the signal multiplied by the gain. The present invention also provides an adaptive post-filter including a filtering unit, a gain calculating unit, and a gain multiplying unit, wherein the filter generates a filter based on an LPC coefficient with respect to the audio signal, and the gain calculated by the gain calculating unit is calculated. Gain multiplying means for multiplying the input signal of the filter;
Means for filtering the signal multiplied by the gain by the filter.

Further, according to the present invention, a post filter is generated based on the LPC coefficient for an audio signal, and a post filter gain is calculated from a signal after the post filter processing and a signal before the post filter processing. An adaptive post-filtering method comprising post-filtering a signal multiplied by a gain. The present invention also provides an adaptive post-filtering method, comprising: generating a filter based on an LPC coefficient for an audio signal; calculating a gain for the filter; and filtering the signal multiplied by the gain with the filter. provide.

[0013]

According to the present invention, the gain of the post-filter (ie, the gain for the signal before the post-filter processing) is calculated for each sub-frame so that the powers of the input signal and the output signal of the post-filter match, and the gain is calculated. Since the post-filter processing is performed on the multiplied synthesized voice signal, the post-filter processing is performed including the noise at the discontinuous portion between the sub-frames due to the gain calculation for each sub-frame, and the output after the post-filter processing is performed. As a signal, it is possible to obtain a continuous waveform in which noise at a discontinuous portion is suppressed.

In addition, since the power of the input signal is completely the same as that of the input signal, the advantages of the noise shaping effect of the post-filter can be fully utilized, and a high-quality voice signal with less noise can be generated.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment based on the principle of the method for resetting the gain of an adaptive post filter according to the present invention will be described below with reference to the block diagram of FIG. 1 and the flowchart of FIG.

As shown in FIG. 1C, the input signal P
x (n) (decoded synthesized speech signal) is input from the input terminal 26 of the post-filter, multiplied by the gain a by the multiplier 27, and then passed through the post-filter 28 to obtain an output signal Py (n). And At this time, the condition that the powers of the input signal and the output signal are equal is expressed by the following equation.

[0017]

(Equation 1)

Here, L represents the section length for which the gain is reset.

An output signal Py (n) of a post filter having a configuration in which a gain is multiplied on the input side as shown in FIG. 1C can be decomposed as in the following equation.

[0020]

(Equation 2)

Here, Py0 (n) represents an output signal (zero input response) output only in the internal state of the post filter when the input signal is zero, as shown in FIG.
As shown in FIG. 1B, y1 (n) is output when Px (n) is directly input to the post-filter without multiplying the gain in a state where the internal state of the post-filter is reset to zero. Represents the output signal (zero state response). Also, a
Represents a reset value of the gain of the post filter. The optimum gain reset value a is determined by solving a quadratic equation of a obtained by substituting equation (2) into equation (1). That is,

[0022]

(Equation 3)

Solving for a gives

[0024]

(Equation 4)

Is obtained.

FIG. 2 is a flowchart showing an example of a method for calculating the output of a post filter based on the above-described principle of the present invention.

In step 11 of FIG. 2, the filter coefficient of the post filter is set, and the internal state of the post filter is temporarily held. Next, a zero input signal is input to the post filter set in the processing 12, and a zero input response Py0 (n) is obtained.
Input signal X (n) input to post filter in process 13
, The internal state of the post-filter is reset to zero, Px (n) is input, and the zero-state response Py1
Find (n).

In a process 16, the zero input response Py0 (n) and the input signal Px obtained in the processes 12, 13, and 14 are respectively obtained.
(N) Using the zero-state response Py1 (n), the optimum post-filter gain a is calculated from the equation (3) and the equation (4). In the process 17, the internal state of the post-filter held in the process 11 is set again, and the input signal Px (n) is multiplied by the optimum gain a as shown in FIG. Find (n).

The post-filter output signal Py (n) generated by the above processing is, as described above, the input signal P
x (n) and the power completely match, and the gain is multiplied on the input side, so that the noise of the discontinuous portion occurring between sections with different gains is also post-filtered and automatically processed. Noise shaping,
Since the output audio signal Py (n) has a continuous waveform, the effect of noise shaping of the post-filter is fully utilized, and the subjective quality of the output audio signal Py (n) is greatly improved.

In this embodiment, the optimum post-filter gain a
Of the input signal Px (n) is passed through a post filter to obtain the output signal Py (n). The zero input response Py0 (n) and the zero state response Py1 (n)
The output signal Py (n) can also be obtained based on the above-described equation (2) using the signal and the optimum gain a.

When the value m2-ln inside the square root in equation (4) is negative, the optimum gain a

[0032]

(Equation 5)

Can be used.

FIGS. 3 and 4 are block diagrams showing a case where the adaptive post-filter according to the second embodiment of the present invention is applied to a CELP-type speech coding apparatus and decoding apparatus.

In FIG. 3, a frame buffer 101 is a circuit for accumulating one frame of an audio signal input to the input terminal 100. Each block in FIG. The following processing is performed.

First, a prediction parameter calculation circuit 102 calculates a short-term prediction parameter for one frame of a speech signal using a known method (normally, 8 to 12 prediction parameters are calculated). The calculation method is described in, for example, (Digital Audio Processing) by Sadateru Furui. The calculated prediction parameters are input to the prediction parameter coding circuit 103. The prediction parameter encoding circuit 103 encodes the prediction parameter based on a predetermined number of quantization bits, outputs the code to the multiplexer 115, and outputs the decoded value P to the prediction filter 104, the perceptual weight filter 105, the influence signal Output to the creation circuit 107, the long-term vector quantization circuit 109, and the short-term vector quantization circuit 111.

The prediction filter 104 calculates a short-term prediction residual signal r from the input speech signal from the frame buffer 101 and the decoded value of the prediction parameter from the encoding circuit 103, and outputs it to the audibility weighting filter 105.

The audibility weighting filter 105 is a filter based on the decoded value P of the prediction parameter, and subtracts the signal x obtained by transforming the spectrum of the short-term prediction residual signal r by a subtraction circuit 106.
Output to This auditory sensation weight filter 105 is for utilizing the masking effect of the auditory sense similarly to the weighting filter in the conventional example, and the details thereof are described in the above-mentioned document, and therefore the description thereof is omitted.

The influence signal generation circuit 107 is provided with the addition circuit 11
2 and the decoded value P of the prediction parameter as inputs, and outputs a past influence signal f. More specifically, a zero input response of an auditory weighting filter, which is the internal state of the past weighted synthesized signal x filter, is calculated, and is output as an influence signal f in preset subframe units. As a typical value in a sub-frame at the time of 8 kHz sampling, one frame (16
0 samples) are divided into four and about 40 samples are used. In the first sub-frame, the influence signal generation circuit 107 receives the composite signal x of the previous frame as an input and the influence signal f
Create The subtraction circuit 106 outputs to the subtraction circuit 108 and the long-term vector quantization circuit 109 a signal u obtained by subtracting the past influence signal f from the perceptually weighted input signal x in subframe units.

The long-term vector quantization circuit 109 includes a difference signal u from the subtraction circuit 106 and a drive signal holding circuit 11 described later.
The drive signal ex from 0 and the prediction parameter P from the encoding circuit 103 are input, and the quantized output signal u of the difference signal u is supplied to the subtraction circuit 108 and the addition circuit 112 on a subframe basis. T is output to the multiplexer 115 and the long-term drive signal t is output to the drive signal holding circuit 110. At this time, t and u
And u = t * h (h is the audibility weighting filter 105
, And * represents convolution).

Vector gain β in subframe units
An example of a detailed method of obtaining (m) and the index T (m) (m is the number of a subframe) will be described below.

A preset index T and a gain β
And a driving signal candidate of the current sub-frame using the past driving signal and inputting the driving signal candidate to a perceptual weighting filter to generate a quantization signal candidate of the difference signal u, and a difference signal u and a quantization signal candidate The optimal index T (m) and the optimal β (m) are determined so as to minimize the error with respect to. Then T
Let t be the drive signal of the current subframe created using (m) and β (m), and let the signal obtained by inputting t into the perceptual weight filter be the quantized output signal u of the difference signal u.

A similar method is described in, for example, “PETER KROO
IEEE February 1988 by Mr. N et al., Vol. SAC-6. pp. 353-
363 "in" A Class of Analysis-by-Synthesi
c Predicative Coders for High Quality Speech Codin
g at Rates Between 4.8 and 1 Gkbits / s "can use a known method similar to the method of obtaining the coefficient of the pitch predictor in a closed loop in the paper.

On the other hand, the subtraction circuit 108 outputs to the short-term vector quantization circuit 111 a difference signal V obtained by subtracting the quantized output signal u from the difference signal u in subframe units. The short-term vector quantization circuit 111 receives the difference signal V and the prediction parameter P as inputs, and outputs the quantization output signal V of the difference signal V to the addition circuit 112 and the short-term drive signal y to the drive signal holding circuit 110 in subframe units. Output. Here, there is a relationship of V = y * h between V and y.

At the same time, the short-term vector quantization circuit 111 outputs the gain G and the index I of the code vector constituting the short-term drive signal to the multiplexer 115.

FIG. 5 shows a specific configuration example of the short-term vector quantization circuit 111. In FIG. 5, a synthetic vector generation circuit 301 creates a pulse train from a prediction parameter P and a code vector C (i) (i is an index of a code vector) in a preset code book 302, and predicts this pulse train. By synthesizing with an audibility weighting filter generated from the parameter P, the synthesized vector V
(I) is generated and output to the inner product calculation circuit 304 and the power calculation circuit 305. The code book 302 stores a pulse amplitude information, and is configured by a memory circuit or a vector generation circuit from which a code vector C (i) predetermined for an index i can be extracted. The inner product calculation circuit 304 obtains an inner product value A (i) of the difference signal V from the subtraction circuit 108 in FIG. 3 and the combined vector V (i), and outputs the result to the index selection circuit 306. Power calculation circuit 30
5 finds the power B (i) of the composite vector V (i),
Output to the index selection circuit 306.

The index selection circuit 306 uses the inner product value A (i) and the power B (i) to calculate the evaluation value

[0048]

(Equation 6)

Is selected from the index candidates i, and the corresponding inner product value A
The combination of (I) and power B (I) is converted to a gain quantization circuit 307.
Output to Further, the index selection circuit 306 further outputs the information of the index I to the codebook 302 and the multiplexer 115 of FIG.

In the gain encoding circuit 307, the inner product value A (I) from the index selecting circuit 306 and the power B
Ratio with (I)

[0051]

(Equation 7)

Is encoded by a predetermined method, and the gain information G is output to the short-term drive signal generation circuit 308 and the multiplexer 115 of FIG.

The above equations (28) and (29) are, for example, IM
International Conference on Acous by Trancoso et al.
tic, speech and Signal Processing paper "EFFICIENT
PROCEDURES FOR FINDING THE OPTIMUM INNOVATI ON IN
STOCHATIC CODERS "may be used.

The short-term drive signal generating circuit 308 receives the gain information G and the code vector C (I) corresponding to the index I, and uses C (I) in the same manner as the method in the composite vector generating circuit 301. A pulse train is created by the method, and the pulse amplitude is multiplied by a value corresponding to the gain information G to generate a short-term drive signal y. This short-term drive signal y
Are output to the audibility weighting filter 309 and the drive signal holding circuit 110 of FIG. The audibility weighting filter 309 is a filter having the same characteristics as the audibility weighting filter 105 of FIG. 3 and is formed based on the prediction parameter P, and receives the short-term drive signal y as an input and outputs the quantized output V of the difference signal V as shown in FIG. To the addition circuit 112.

Returning to FIG. 3, the driving signal holding circuit 1
Reference numeral 10 designates a long-term drive signal t output from the long-term vector quantization circuit 109 and a short-term drive signal y output from the short-term vector quantization circuit 111 as inputs, and the drive signal ex is transmitted to the long-term vector quantization circuit 109 in subframe units. Output. Specifically, for example, a signal obtained by adding t and y for each sample in subframe units may be used as the drive signal ex. The drive signal ex of the current subframe is held in a buffer memory in the drive signal holding circuit 110 so that it can be used in the long-term vector quantization circuit 109 as a past drive signal in the next subframe. Adder circuit 11
2 is the quantization output u (m) and V
A sum signal x of (m) and the past influence signal f generated in the current subframe is obtained, and output to the influence signal generation circuit 107.

The information of the parameters P, β, T, G, and I obtained as described above is multiplexed by the multiplexer 115 and transmitted from the output terminal 116 as a transmission code.

Next, the decoding apparatus shown in FIG. 4 for decoding the code transmitted from the decoding apparatus shown in FIG. 3 will be described.

In FIG. 4, a transmitted code is input to an input terminal 200. The demultiplexer 201 first converts the input code into a prediction parameter, a gain β, a gain G,
The code is separated into the index T and the index I. The decoding circuits 202 to 207 respectively have a gain G,
The codes of the index I, the gain β and the index T are decoded and output to the drive signal generation circuit 209. Another decoding circuit 208 decodes the encoded prediction parameters and outputs the decoded prediction parameters to the synthesis filter 210. The drive signal generation circuit 209 receives the decoded parameters as inputs,
Generate a drive signal.

The drive signal generation circuit 209 is specifically configured, for example, as shown in FIG. 6, a code book 500 has the same function as the code book 302 shown in FIG. 5 in the encoding apparatus, and outputs a code vector C (I) corresponding to the index I to the short-term drive signal generation circuit 501. . Short-term drive signal generation circuit 501
Is a short-term drive signal generation circuit 3 shown in FIG.
08, and has a gain G as an input and outputs a short-term drive signal y to an adder circuit 506. The adder circuit 506 includes a short-term drive signal y and a long-term drive signal generated by a long-term drive signal generation circuit 502. The sum signal with the signal t, that is, the drive signal ex is output to the drive signal buffer 503 and the synthesis filter 210 of FIG.

The driving signal buffer 503 is
A configuration in which the drive signal output from the control signal 6 is held up to a predetermined number of samples from the present to the past, and when an index T is input, the drive signal is output by the number of samples corresponding to the sub-frame length in order from the drive signal T samples past. Has become
The long-term drive signal generation circuit 502 receives a signal output from the drive signal buffer 503 based on the index T, multiplies the input signal by a gain β, and generates a long-term drive signal that repeats at a period of T samples. 5
06 in subframe units.

Returning to FIG. 4, the synthesis filter 210
Is a filter having a frequency characteristic opposite to that of the prediction filter 104 shown in FIG. 3 in the encoding device, and outputs a synthesized signal by using the drive signal and the prediction parameter as inputs.

The post filter 211 has prediction parameters,
Synthesis filter 2 using gain β and index T
The spectrum of the synthesized signal output from 10 is subjectively shaped so as to reduce noise, and output to the buffer 212. FIG. 7 shows a specific configuration example of the post filter 211.

The post filter shown in FIG. 7 is composed of a continuous connection of two post filters. One is FINE SPECT
RAL SHAPING filter, another is GLOBAL SPECT
RALSHAPING filter. The former is a filter for emphasizing the repetition of pitch, and affects the fine structure of the spectrum of the output audio signal. As a transfer function of this filter, for example,

[0064]

(Equation 8)

Is generally used. Where ε is 0.2 ~
It is known that a value of about 0.5 is appropriate.

On the other hand, the latter filter is a filter that emphasizes the shape of the spectral envelope of the audio signal.

[0067]

(Equation 9)

Is generally used. Where 0 <r1 <r
2 <1, and A (Z) represents a prediction filter composed of prediction parameters decoded on the receiving side.

Hereinafter, the post filter will be described in detail with reference to FIG.

In FIG. 7, the synthesized signal Px (n) output from the synthesis filter 210 is input from the post-filter input terminal 33. The input signal Px (n) is output to the gain calculation circuit 37 and the switch terminal 43. The switch 55 connects the terminal 43 and the terminal 34, and the FINE SPECTRAL
The input signal Px (n) is input to the SHAPING filter 35. The output signal of the FINE SPECTRAL SHAPING filter 35 is G
The signal is input to the LOBAL SPECTRAL SHAPING filter 36, and the output signal Py1 (n) is input to the gain calculation circuit 37. Here, the output signal Py1 (n) is a zero-state response,
The filters 35 and 36 calculate and output Py1 (n) based only on the input value of Px (n) without using the internal state of the filter before the current subframe.

On the other hand, the filters 31 and 32 are post-filters having the same characteristics as the above-described filters 35 and 36, respectively. The continuously connected filters receive zero data from the terminal 30 and , A zero input response Py0 (n) is generated using the internal state of the filter, and this is input to the gain calculation circuit 37.

The gain calculation circuit 37 receives the input Px
(N), Py0 (n) and Py1 (n) signals,
The optimum gain a of the post filter is obtained from the above-described equations (3) and (4), and is input to the multiplication circuit 39 and a control signal is input to the switch 55. The switch 55 connects the switch connected to the terminal 34 to the terminal 38 based on the control signal from the gain calculation circuit 37, and outputs the input signal xP
(N) is input to the multiplication circuit 39.

The multiplying circuit 39 inputs a signal whose amplitude is three times the input signal xP (n) to the filter 40. Filter 4
0 is a filter having the same characteristics as the filter 31, and its output is input to the filter 41. The filter 41 has the same characteristics as the filter 32, and its output P
y (n) is output to the post filter output terminal 42. The filters 40 and 41 use a post-filter output signal Py using the internal state of the filter before the current subframe.
(N) is generated.

The description of the post-filter in FIG. 7 has been completed. As another example of the post filter, FIG. 8 shows an example in which the post filter of the configuration of FIG. 7 is shared.

In FIG. 8, when the zero input response Py0 (n) is output to the gain calculation circuit 66, the switch 61 turns off the terminals 61 and 62, and the switch 74 turns on the terminals 70 and 71. What should I do?

When outputting the zero state response Py1 (n) to the gain calculation circuit 66, the switch 75 is turned off,
The switch 74 only has to turn on the terminals 72 and 71. Finally, in order to output the post-filter output signal Py (n), the switch 76 is turned on and the switch 75 is turned on.
Is turned on and the switch 74 is turned off. This is the end of the description of the post filter in FIG.

Referring back to FIG. 4, the buffer 212 combines the input post-filter output signals for each sub-frame and outputs the post-filtered synthesized speech signal to the output terminal 213. FIGS. 9 and 10 are block diagrams of a case where the adaptive post filter according to the third embodiment of the present invention is applied to a CELP-type speech encoding device and decoding device using an adaptive density pulse train as a drive signal.

The adaptive density pulse train indicates a pulse train in which a frame is divided into a plurality of subframes and the pulse interval is variable in subframe units. FIG. 13 shows an example of the adaptive density pulse train.

The adaptive density pulse train adaptively transfers a larger amount of information to an important part of a voice in a frame than a drive signal using a pulse train having a constant density (in other words, a pulse interval) used in the conventional CELP system. This is an improved CELP-type drive signal that can provide high-quality speech even when it is assigned to a start portion of speech or a portion where the power of the speech signal changes, because it can be assigned. This adaptive density pulse train has already been proposed by the present inventors.

FIGS. 9 and 10 are block diagrams of a speech encoding apparatus and a decoding apparatus according to the third embodiment of the present invention.

In FIG. 9, the frame buffer 1101
Is a circuit for accumulating one frame of audio signals input to the input terminal 1100. Each block in FIG. 9 performs the following processing for each frame or subframe using the frame buffer 1101.

First, a prediction parameter calculating circuit 1102 calculates a short-term prediction parameter for one frame of a speech signal using a known method (normally, 8 to 12 prediction parameters are calculated). The calculation method is described in, for example, (Digital Audio Processing) by Sadateru Furui. The calculated prediction parameters are input to the prediction parameter coding circuit 1103. The prediction parameter encoding circuit 1103 encodes the prediction parameter based on a predetermined number of quantization bits, outputs the code to the multiplexer 1115, and outputs the code value P to the prediction filter 1104, the perceptual weight filter 1105, and the influence signal. Creation circuit 1107, long-term vector quantization circuit 110
9 and a short-term vector quantization circuit 1111.

The prediction filter 1104 calculates a short-term prediction residual signal r from the input speech signal from the frame buffer 1101 and the code value of the prediction parameter from the encoding circuit 1103, and outputs it to the auditory weight filter 1105.

The audibility weighting filter 1105 is a filter configured based on the decoded value P of the prediction parameter, and subtracts the signal x obtained by deforming the spectrum of the short-term prediction residual signal r by the subtraction circuit 11.
06. The auditory sensation weight filter 1105 is for utilizing the auditory masking effect in the same manner as the conventional weighting filter.

The influence signal creation circuit 1107 is composed of the addition circuit 1
It receives the past weighted synthesized signal x from 112 and the decoded value P of the prediction parameter, and outputs the past influence signal f. Specifically, the past weighted synthesized signal x
Is calculated as the internal state of the filter, and the zero input response of the perceptual weighting filter is calculated, and is output as an influence signal f in units of subframes set in advance. As a typical value in a subframe at the time of 8 kHz sampling, about 40 samples obtained by dividing one frame (160 samples) into four are used. In the first sub-frame, the influence signal creation circuit 1107 creates an influence signal f by using as input the synthesized signal x of the previous frame created based on the density pattern K determined in the previous frame. The subtraction circuit 1106 outputs, to the subtraction circuit 1108 and the long-term vector quantization circuit 1109, a signal u obtained by subtracting the past influence signal f from the perceptually weighted input signal x in subframe units.

The power calculation circuit 1113 calculates the power (sum of squares) of the short-term prediction residual signal output from the prediction filter 1104 in subframe units, and outputs the power of each subframe to the density pattern selection circuit 1114. .

The density pattern selection circuit 1114 selects one of the preset drive signal density patterns based on the power of the short-term predicted step signal for each subframe output from the power calculation circuit 1115. In particular,
A density pattern is selected so that the density becomes higher in the order of the subframes having higher power. For example, four equal-length subframes, two types of density, and a density pattern shown in FIG.
In the case of setting as follows, the density pattern selection circuit 1115
Compares the power for each subframe, selects the number K of the density pattern in which the subframe with the highest power is dense, and outputs it to the short-term vector quantization circuit 1111 and the multiplexer 1115 as density pattern information.

The long-term vector quantization circuit 1109 includes a difference signal u from the subtraction circuit 1106, a past drive signal ex from a drive signal holding circuit 1110 described later, and the encoding circuit 1
The prediction parameter P from the input signal 103 is input, and the subtraction circuit 108 subtracts the quantized output signal u of the difference signal u in subframe units.
To the adder 1112, the vector gain β and the index T to the multiplexer 1115, and the long-term drive signal t to the drive signal holding circuit 1110.
At this time, there is a relationship between t and u: u = t * h (h represents the impulse response of the auditory weighting filter 1105, and * represents convolution).

Vector gain β in subframe units
An example of a detailed method of obtaining (m) and the index T (m) (m is the number of a subframe) will be described below.

The index T and the gain β are set in advance.
And a driving signal candidate of the current sub-frame using the past driving signal and inputting the driving signal candidate to a perceptual weighting filter to generate a quantization signal candidate of the difference signal u, and a difference signal u and a quantization signal candidate The optimal index T (m) and the optimal β (m) are determined so as to minimize the error with respect to. Then T
Let t be a drive signal of the current subframe created using (m) and the optimal β (m), and let a signal obtained by inputting t into the perceptual weighting filter be a quantized output signal u of the difference signal u. .

A similar method is described in, for example, PETER KROON
IEEE February 1988, Vol. SAC- 6.pp. 353-36
3 "A Class of Analysis-by-Synthesic Pre
dicative Coders for High Quality Speech Coding at
Since a well-known method similar to the method of obtaining the coefficients of the pitch predictor in the closed loop in the paper entitled Rates Between 4.8 and 16 kbits / s "can be used, the description is omitted here.

On the other hand, in the subtraction circuit 1108, the difference signal V obtained by subtracting the quantized output signal u from the difference signal u in subframe units.
Is output to the short-term vector quantization circuit 1111.

The short-term vector quantization circuit 1111 includes a difference signal V, a prediction parameter P, and a density pattern selection circuit 1.
The density pattern number K output from 114 is input,
The quantized output signal V of the difference signal V is added to the adder circuit 1112 on a subframe basis, and the short-term drive signal y is
Output to the respective 110. Here, between V and y,
There is a relationship of V = y * h.

At the same time, the short-term vector quantization circuit 1111 converts the gain G of the drive pulse train, the phase information J, and the index I of the code vector into the multiplexer 11.
15 is output. At this time, parameters G, J, and I output in units of subframes represent the number of pulses N (m) corresponding to the density (pulse interval) of the current subframe (mth subframe) determined by the density pattern number K. Since encoding must be performed within a frame, the number according to a predetermined number of dimensions ND of the code vector (the number of pulses constituting each code vector), that is, N (m) / ND
It is output in the current subframe one by one.

For example, it is assumed that the frame length is 160 samples, the subframe is composed of four equal length 40 samples, and the dimension of the code vector is 20. In this case, assuming that one of the density patterns prepared in advance is the pulse interval 1 of the first subframe and the pulse interval 2 of the second to fourth subframes, the short-term vector quantization circuit 1111 outputs this density pattern. The number of gains, phases, and indices are 40/2 in the first subframe, respectively.
0 = 2 (However, in this case, since the pulse interval is 1,
Phase information is not output), 2 in the second to fourth subframes
0/20 = 1.

FIG. 11 shows a specific configuration example of the short-term vector quantization circuit 1111. In FIG. 11, a combined vector generation circuit 1301 includes a prediction parameter P and a code vector C (i) in a code book 1302 set in advance.
(I is the index of the code vector) and the density pattern information K, zero is set at a predetermined cycle after the first sample of C (i) so as to have a preset pulse interval corresponding to the density pattern information K. By interpolating to create a pulse train with density information and combining this pulse train with an audibility weighting filter generated from the prediction parameter P,
Generate a composite vector V1 (i).

The phase shift circuit 1303 delays the synthesized vector V1 (i) by a predetermined number of samples based on the density pattern information K, and synthesizes the synthesized vectors V1 (i) having different phases.
2 (i), V3 (i),... V1 (i),.
Output to the inner product calculation circuit 1304 and the power calculation circuit 1305. The codebook 1302 is configured by a memory circuit or a vector generation circuit that stores amplitude information of an adaptive density pulse and can extract a predetermined code vector C (i) for an index i. Inner product calculation circuit 1
304 is a difference signal V from the subtraction circuit 1108 of FIG.
An inner product value Aj (i) with the composite vector Vj (i) is obtained,
Output to the index / phase selection circuit 1306. The power calculation circuit 1305 obtains the power Bj (i) of the composite vector Vj (i), and calculates the index / phase selection circuit 13
06.

The index / phase selection circuit 1306 uses the inner product value Aj (i) and the power Bj (i) to calculate the evaluation value

[0099]

(Equation 10)

The phase J and the index I that maximize the maximum value are selected from the phase candidate j and the index candidate i, and the corresponding set of the inner product value AJ (I) and the power BJ (I) is converted to a gain quantization circuit 1307. Output to Further, the index / phase selection circuit 1306 further outputs the information of the phase J to the short-term drive signal generation circuit 1308 and the multiplexer 1115 of FIG. 9, and outputs the information of the index I to the codebook 1302 and the multiplexer 1115 of FIG.

In gain encoding circuit 1307, the ratio between inner product value Aj (I) from index / phase selection circuit 1306 and power BJ (I) is calculated.

[0102]

[Equation 11]

Is encoded by a predetermined method, and the gain information G is output to the short-term drive signal generation circuit 1308 and the multiplexer 1115 in FIG.

The above equations (28) and (29) are, for example, IM
International Conference on Acous by Trancoso et al.
tic, Speech and Signal Processing paper "EFFICIENT
PROCEDURES FOR FINDING THE OPTIMUM INNOVATI ON IN
STOCHATIC CODERS "may be used.

The short-term drive signal generation circuit 1308 receives the density pattern information K, the gain information G, the phase information J, and the code vector C (I) corresponding to the index I, and uses the K and C (I) to perform the above synthesis. A pulse train having density information is created by a method similar to the method of the vector generation circuit 1301, the pulse amplitude is multiplied by a value corresponding to the gain information G, and the pulse train is delayed by a predetermined number of samples based on the phase information J. Generates a short-term drive signal y. This short-term drive signal y is applied to the audibility weighting filter 130
9 and the drive signal holding circuit 1110 of FIG. The perceptual weight filter 1309 is a filter having the same characteristics as the perceptual weight filter 1105 in FIG. 9 and is formed based on the prediction parameter P. Nine adder circuits 111
Output to 2.

Returning to FIG. 9, the driving signal holding circuit 1
110 is a long-term driving signal t output from the long-term vector quantization circuit 1109 and a short-term vector quantization circuit 111
1 is input, and the driving signal ex is input to the long-term vector quantization circuit 11 in subframe units.
09 is output. Specifically, for example, a signal obtained by adding t and y in subframe units for each sample is a drive signal ex.
And it is sufficient. The drive signal ex of the current subframe is held in a buffer memory in the drive signal holding circuit 1110 so that it can be used as a past drive signal in the next subframe in the long-term vector quantization circuit 1109.

The addition circuit 1112 obtains a sum signal x of the quantized outputs u (m) and V (m) in units of subframes and the past influence signal f created in the current subframe. Output to 1107.

The information of the parameters P, β, T, G, I, J, and K obtained as described above is
115 and multiplexed by the output terminal 11 as a transmission code.
16 is transmitted.

Next, the decoding apparatus shown in FIG. 10 for decoding the code transmitted from the encoding apparatus shown in FIG. 9 will be described.

In FIG. 10, a transmitted code is input to an input terminal 1200. The multiplexer 1201 first separates the input code into codes of a prediction parameter, density pattern information K, gain β, gain G, index T, index I, and phase information J. Decoding circuit 1
202 to 1207 decode the codes of the density pattern information K, gain G, phase J, index I, gain β, and index T, respectively, and
Output to Another decoding circuit 1208 decodes the encoded prediction parameter, and
Output to The drive signal generation circuit 1209 receives the decoded parameters as inputs, and generates drive signals having different densities in subframe units based on the density pattern information K.

The drive signal generation circuit 1209 is specifically configured, for example, as shown in FIG. 12, a code book 1500 has the same function as the code book 1302 shown in FIG. 11 in the encoding apparatus, and a code vector C (I) corresponding to the index I is provided.
To the short-term drive signal generation circuit 1501. The short-term drive signal generation circuit 1501 has the same function as the short-term drive signal generation circuit 1308 shown in FIG. 11 in the encoding device, receives the density pattern information K, the phase information J, and the gain G as inputs, and y is output to the addition circuit 1506. The addition circuit 1506 outputs the sum signal of the short-term drive signal y and the long-term drive signal t generated by the long-term drive signal generation circuit 1502, that is, the drive signal ex to the drive signal buffer 1
503 and the synthesis filter 1210 of FIG.

The drive signal buffer 1503 is
A configuration in which the drive signal output from the 506 is held up to the past by a predetermined number of samples from the present, and when the index T is input, the drive signal is output by the number of samples corresponding to the sub-frame length in order from the drive signal T samples past Has become. The long-term drive signal generation circuit 1502 uses the index T
, A signal output from the drive signal buffer 1503 is input, and the input signal is multiplied by a gain β.
A long-term drive signal that repeats at a period of T samples is generated and output to the adder circuit 1506 in subframe units.

Referring back to FIG. 10, the synthesis filter 12
Reference numeral 10 denotes a prediction filter 1104 shown in FIG.
This filter has a frequency characteristic opposite to that of the filter, and receives the drive signal and the prediction parameter as inputs and outputs a synthesized signal.

The post-filter 1211 shapes the spectrum of the synthesized signal output from the synthesis filter 1210 using the prediction parameter, the gain β, and the index T such that the noise is reduced subjectively, and the buffer 1212
Output to The specific configuration of the post filter 1211 is the same as that described in the second embodiment, and a description thereof will not be repeated. The buffer 1212 combines the input signals for each frame and outputs the synthesized audio signal to the output terminal 1
213.

A fourth embodiment of the adaptive post filter according to the present invention will be described with reference to FIG. In FIG.
Px (n) represents an input signal of the post filter, and Py (n) represents an output signal. The signal Px input from the terminal 2060
(N) First, switch 2075 OFF, switch 20
The direct post filters 2064 and 206 in the 74 ON state
5, and the output Py2 (n) of the filter 2065 is input to the gain calculation circuit 2066.

The gain calculation circuit 2066 receives Py2 (n) and Px (n), and sets the gain a of the post filter according to the following equation.

[0117]

(Equation 12)

When the gain a is set by the equation (6), the gain calculation circuit 2066 switches the switch 2075 and the switch 207
4 and a switch 2076 to output a control signal. The switch 2074 is turned off, and then the switch 2075 is turned on.
And switch 2076 is turned on.

By inputting Px (n) from the terminal 2060 in this state, the post-filtered output signal Py (n) is output from the terminal 2067.

In the embodiment described above, the post filter is
FINE SPECTRAL SHAPING filter and GLOBAL SPECTRAL SHA
Although two filters of the ping filter are continuously connected, the configuration of the post filter may be a configuration using only one of the filters, or a configuration using another filter in combination.

Further, an example of a polar filter has been described as a FINE SPECTRAL SHAPING filter.

[0122]

(Equation 13)

Can be used instead. Where 0 <
There is a relationship of κ <ε, and ε needs to satisfy | εβ | <|.

[0124]

According to the present invention, the post-filter gain is calculated for each sub-frame, and then the synthesized voice signal multiplied by the gain is subjected to post-filter processing. Oppress,
It is possible to automatically obtain a noise-shaped output signal.

Further, since the power is completely matched with the input signal, the advantage of the noise shaping effect of the post filter can be fully utilized, and a high-quality audio signal with less noise can be generated.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating the principle of the present invention.

FIG. 3 is a block diagram illustrating a second embodiment of the present invention.

FIG. 4 is a block diagram illustrating a second embodiment of the present invention.

FIG. 5 is a configuration diagram of a short-term vector quantization circuit according to the present invention.

FIG. 6 is a configuration diagram of a drive signal generation circuit according to the present invention.

FIG. 7 is a configuration diagram of a post filter according to the present invention.

FIG. 8 is another configuration diagram of a post filter.

FIG. 9 is a block diagram illustrating a third embodiment of the present invention.

FIG. 10 is a block diagram illustrating a third embodiment of the present invention.

FIG. 11 is a configuration diagram of a short-term vector quantization circuit.

FIG. 12 is a configuration diagram of a drive signal generation circuit.

FIG. 13 is a diagram for explaining an adaptive density pulse train.

FIG. 14 is a block diagram illustrating a fourth embodiment of the present invention.

FIG. 15 is a diagram for explaining a conventional example.

FIG. 16 is a diagram showing a setting example of a density pattern.

[Explanation of symbols]

 33 ... input terminal 31, 35, 40 ... FINE SPECTRAL SHAPING filter 32, 36, 41 ... GLOBAL SPECTRAL SHAPING filter 37 ... gain calculation circuit 39 ... multiplication circuit 42 ... output terminal 55 ... switch

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G10L 19/06 G10L 21/02

Claims (8)

(57) [Claims] 1. A means for generating a post filter based on an LPC coefficient for an audio signal, a means for calculating a post filter gain from a signal after the post filter processing and a signal before the post filter processing, Means for post-filtering the signal multiplied by the gain. 2. The gain calculator according to claim 1, wherein: 2. The adaptive post filter according to claim 1, wherein the gain a is calculated as follows: (the input signal of the post filter is Px (n) and the output signal is Py2 (n)). 3. An adaptive post filter comprising filtering means, gain calculating means, and gain multiplying means,
Means for generating a filter based on the LPC coefficient for the audio signal, gain multiplication means for multiplying the input signal of the filter by the gain calculated by the gain calculation means, and filtering the signal multiplied by the gain by the filter And an adaptive post filter. 4. The adaptive post filter according to claim 3, wherein said gain calculation means calculates a gain from an input signal of a post filter and an output signal of said filter. 5. A post filter is generated based on LPC coefficients for an audio signal, a post filter gain is calculated from a signal after the post filter processing and a signal before the post filter processing, and the calculated gain is multiplied. An adaptive post-filtering method, wherein post-filtering is performed on the obtained signal. 6. The gain calculation is as follows: 6. The adaptive post-filtering method according to claim 5, wherein the gain a is calculated as follows: (the input signal of the post-filter is Px (n) and the output signal is Py2 (n)). 7. An adaptive post-filtering method, wherein a filter is generated based on LPC coefficients for an audio signal, a gain for the filter is calculated, and a signal multiplied by the gain is filtered by the filter. 8. The adaptive post-filtering method according to claim 7, wherein said gain calculation calculates a gain from an input signal and an output signal of a post-filter.