JP3103108B2 - Audio coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a speech coding apparatus for compressing a speech signal with high efficiency, and in particular, can reduce a transmission bit rate to, for example, 10 kb / s or less. The present invention relates to a speech coding apparatus based on an adaptive density driving pulse train.

(Prior art) At present, coding techniques capable of transmitting audio signals at a low bit rate of 10 kb / s or less are being actively studied. As a specific method, a driving signal for a synthesis filter is composed of a pulse train arranged at regular intervals. And encoding is performed using this. For more information on this, see IEE by PETER KROON et al.
E October 1986, “Regular Pulse Excitation A Novel Approach to Eff” published in Vol.ASSP-34, pp.1054-1063.
ective and Efficient Multipulse Coding of Speech "
(Reference 1).

This speech coding method will be described with reference to FIGS. 25 and 26. FIG. 25 and FIG. 26 are block diagrams of the encoder and the decoder.

In FIG. 25, an A / D converted audio signal sequence s (n) is input to an input terminal IN. The prediction filter 1 is s
(N) past series and prediction parameters a _i (1 ≦ i ≦ p)
Is used to calculate and output a prediction residual signal r (n) as shown in the following equation.

Here, p is the order of the prediction filter 1, and in Reference 1, p = 12. Transfer function A of prediction filter 1
(Z) is expressed as follows.

The drive signal generation circuit 2 generates drive pulse trains V (n) arranged at predetermined intervals as drive signals. FIG. 27 shows an example of the pattern of the driving pulse train V (n). In this figure, K indicates the phase of the pulse sequence, and indicates the position of the first pulse in the frame. The horizontal axis represents discrete times. Here, a case where the length of one frame is 40 samples (5 ms at a sampling frequency of 8 kHz) and the pulse interval is 4 samples is shown.

The subtractor 3 calculates the prediction residual signal r (n) and the driving signal V
The difference from (n) is calculated and output to the weighting filter 4. This filter 14 is for shaping e (n) in the frequency domain in order to utilize the masking effect of hearing, and its transfer function W (z) is given by the following equation.

The weighting filter and the masking effect are described in, for example, "Digital Speech Processing" by Sadateru Furui, published by Tokai University Press in 1985 (Reference 2), and will not be described here.

Error e ′ weighted by weighting filter 4
(N) is input to the error minimizing circuit 5. The error minimizing circuit 5 determines the amplitude and phase of the drive pulse train so that the square error of e '(n) is minimized. The drive signal generation circuit 2 generates a drive signal based on the information on the amplitude and the phase. The procedure for determining the amplitude and phase of the drive pulse train in the error minimizing circuit 5 will be briefly described below according to the description in Reference 1.

First, the frame length is L samples, the number of drive pulses in one frame is Q, and Q × L representing the position of the drive pulse
Let M _K be a line example of. The element m _{ij of} M _K is expressed as follows. K is the phase of the drive pulse train as described above.

When m _ij = 1; j = i × N; K−1, m _ij = 0; other 0 ≦ 1 ≦ Q−1 0 ≦ j ≦ L−1 (where N = L / Q) (4) Next When a row vector having a non-zero amplitude of a drive signal (drive pulse train) of the phase K as an element is denoted by b ^(K) , a row vector u ^(K) representing the drive signal of the phase K is represented by the following equation. Is represented.

u ^(K) = b ^(K) M _K (5) Let H be the next L × L matrix having the impulse response of the weighting filter 4 as an element.

At this time, an error vector e ^(k) having the weighted error e '(n) as an element is described by the following equation.

e ^(K) = e ⁽⁰⁾ −b ^(K) (7) (K = 1,..., N) where e ⁽⁰⁾ e ₀ + rH (8) H _K = M _K H (9) Vector e ₀ is the output of the weighting filter based on the internal state of the weighting filter in the previous frame, and vector r is the prediction residual signal vector. A vector b ^(K) representing the optimum drive pulse amplitude is obtained by partially differentiating the square error E = e ^(K) e ^{(K) t} (10) of the following equation with b ^(K) and setting it to zero. Is obtained as follows:

^{b (k) = e (0} ) H K t [H K H K t] -1 (11) where, t denotes the transpose.

At this time, the phase K of the drive pulse train is calculated so that the following equation is calculated for each K, and is selected so that E ^(K) is minimized.

^{E (K) = e (0} ) [H K t [H K H K t] -1 H K] · e (0) t (12) amplitude and phase of the thus drive pulse train is determined.

 Next, the decoder will be described.

26, the drive signal generation circuit 7 is the same as the drive signal generation circuit 2 in FIG. 25, and is driven based on the amplitude and phase of the drive pulse train transmitted from the encoder and input to the input terminal 6. Occurs. The synthesis filter 8 receives the drive signal as input, generates a synthesized voice signal s (n), and outputs the synthesized voice signal s (n) to the output terminal 9. This synthesis filter 8 has a relationship of an inverse filter with the prediction filter 1 in FIG. 13, and its transfer function is 1 / A (z).

In the conventional coding method described above, the information to be transmitted is the parameter a _i (1 ≦ i ≦ p) of the synthesis filter 22 and the amplitude and phase of the drive pulse train, and changes the drive pulse train interval N = L / Q. This allows the transmission rate to be set freely. However, according to the experimental results of the conventional method, when the transmission rate is 10 kb / s or less, noise is conspicuous in the synthesized speech, and the quality is deteriorated. In particular, quality degradation is noticeable when an experiment is performed with a female voice having a short pitch cycle. It has been found that this is because the drive pulse train is always represented by pulse trains at a constant interval. That is, since the audio signal is a periodic signal based on the pitch of the voiced sound, the prediction residual signal is also a periodic signal in which the power increases at each pitch cycle. As described above, in the prediction residual signal whose power periodically increases, a portion where the power is large contains important information. In addition, in a portion where the power of the audio signal increases, such as a portion where the correlation of the audio signal changes due to deterioration of the phoneme or the like, or a start portion of utterance, the power of the prediction residual signal also increases in the frame. Also in this case, the part where the power of the residual signal is large is important because it is the part where the property of the audio signal has changed.

However, in the conventional method, although the power of the prediction residual signal changes in the frame, the synthesis filter is driven by a drive pulse train having a constant interval in the frame to obtain the synthesized voice. However, the quality of the synthesized speech is significantly deteriorated.

(Problems to be Solved by the Invention) As described above, in the conventional speech coding method, since the synthesis filter is driven by a driving pulse train having regular constant intervals in a frame, the transmission rate is, for example, 10 kb / s. Below this, there is a problem that the quality of synthesized speech deteriorates.

The present invention has been made in view of such a problem.
It is an object of the present invention to provide a speech encoding device capable of obtaining high-quality synthesized speech even at a low transmission rate such as kb / s or less.

[Constitution of the Invention] (Means for Solving the Problems) The present invention relates to a speech coding apparatus for obtaining a synthesized speech by driving a synthesis filter with a drive signal, wherein a pulse density (pulse interval) is variable in a predetermined section unit. A certain drive pulse train,
Specifically, the frame of the drive signal is divided into a plurality of subframes, and the drive signal is constituted by a drive pulse train in which the pulse density is variable in each subframe or in each frame at equal intervals and in subframe or frame units. I do. The amplitude or the amplitude and the phase of the drive pulse train are determined so that the power of the perceptually weighted error signal between the output signal of the synthesis filter driven by the drive signal and the input audio signal is minimized. (A) a signal obtained by passing a prediction residual signal (short-term prediction residual signal, pitch prediction residual signal, or pitch prediction residual signal obtained by performing pitch prediction on the short-term prediction residual signal) through an auditory weighting filter, b) pitch prediction residual signal obtained by performing pitch prediction on the short-term prediction residual signal;
(C) Each of the drive signals calculated based on the prediction residual signal (short-term prediction residual signal, pitch prediction residual signal, or pitch prediction residual signal obtained by performing pitch prediction on the short-term prediction residual signal) (D) prediction residual signal (short-term prediction residual signal, pitch prediction residual signal, or pitch prediction residual signal obtained by performing pitch prediction on short-term prediction residual signal) The power or the number of zero crossings is determined based on a non-linear weighted function value.

Alternatively, the pulse density may be set in sub-frame units according to a predetermined density pattern, and the drive signal may be configured by a drive pulse train whose density pattern is variable in frame units. In this case, the density pattern of the drive pulse train is selected and determined from a plurality of patterns that are facilitated in advance based on the results obtained in (a) to (d).

(Operation) Based on a signal obtained by passing a prediction residual signal (a short-term prediction residual signal, a pitch prediction residual signal, or a pitch prediction residual signal obtained by performing pitch prediction on a short-term prediction residual signal) through an auditory weighting filter. When the pulse density of the drive pulse train is determined by the above method, a weighted residual signal in which the amplitude of audibly important information is emphasized is obtained from the audibility weighting filter. Can be determined, and the quality of the synthesized speech is further improved.

Further, by determining the pulse density of the driving pulse train based on the pitch prediction residual signal obtained by performing pitch prediction on the short-term prediction residual signal, that is, the residual signal obtained by combining pitch prediction with normal short-term prediction. Compared to the method that determines the pulse density based only on the short-term prediction residual signal, it is possible to determine the pulse density that reduces the distortion of the synthesized speech, resulting in a dramatic improvement in the quality of the synthesized speech. I do.

Each sub-signal of the drive signal calculated based on the prediction residual signal (short-term prediction residual signal, pitch prediction residual signal, or pitch prediction residual signal obtained by performing pitch prediction on the short-term prediction residual signal). When the pulse density of the drive pulse train is determined based on the bit allocation value assigned to the frame,
Since it is possible to determine whether or not the section (subframe) is audibly important depending on the size of the bit allocation value, the pulse density is determined based on this, so that the important section in the synthesized speech can be determined. A pulse density that can be expressed more accurately can be determined, and the quality of synthesized speech is improved.

Further, the power or the number of zero crossings of the prediction residual signal (short-term prediction residual signal, pitch prediction residual signal, or pitch prediction residual signal obtained by performing pitch prediction on the short-term prediction residual signal) is nonlinearly weighted. If the pulse density of the drive pulse train is determined based on the applied function value, fine weight adjustment that cannot be obtained by a normal linear weighting operation can be performed. Therefore, based on the weighted function value, the pulse density can be more accurately determined. As a result, the quality of the synthesized speech is improved as compared with the method of simply determining the pulse density based on only the short-term prediction residual signal.

That is, in the present invention, the density of the drive pulse train is adapted for each sub-frame or frame such that the density of the drive pulse train is dense in a sub-frame or frame containing information that is perceptually significant or a lot of information, and coarse in a sub-frame or frame that is not. As a result, the quality of synthesized speech is improved. In particular, if the density of the drive pulse train is changed in units of subframes, the quality of the synthesized voice is improved in the portion where the characteristics of the voice changes rapidly, and if the density is changed in units of frames, the characteristics of the voice are relatively moderate. The quality of the synthesized speech in the changing part is improved. In particular, if the density of the drive pulse train is changed in units of subframes, the quality of the synthesized voice is improved in the portion where the characteristics of the voice changes rapidly, and if the density is changed in units of frames, the characteristics of the voice are relatively moderate. The quality of the synthesized speech in the changing part is improved.

In addition, the pulse density is set in subframe units according to a predetermined density pattern, a drive signal is formed by a drive pulse train whose density pattern is variable in frame units, and the density pattern of the drive pulse train is adaptively adjusted according to the nature of audio. , The quality of the synthesized speech is improved as a whole in any part where the properties of the speech change.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 and FIG. 2 are block diagrams of an encoding device and a decoding device according to a first embodiment, respectively, showing the basic configuration of the present invention.

In FIG. 1, a frame buffer 11 is a circuit for accumulating one frame of the audio signal input to the input terminal 10. Each block in FIG. 1 uses the frame buffer 11 for each frame or each subframe. Perform processing.

The prediction parameter calculation circuit 12 calculates a prediction parameter using a known method. The prediction filter 14 includes a long-term prediction filter 41 and a short-term prediction filter 42 as shown in FIG.
Are connected in cascade, the prediction parameter calculation circuit 12 calculates the pitch period, the pitch prediction coefficient, and the linear prediction coefficient (α parameter or K parameter) by a known method such as an autocorrelation method or a covariance method. For the calculation method,
For example, Ref. 2 (Sadateru Furui, "Digital Audio Processing")
1985, published by Tokai University Press). The calculated prediction parameters are input to the prediction parameter coding circuit 13. The prediction parameter encoding circuit 13 encodes the prediction parameter based on a predetermined number of quantization bits, and outputs this code to the multiplexer 25.
The decoded value is output to the prediction filter 14, the synthesis filter 15, and the perceptual weight filter 20. The prediction filter 14 receives the input speech signal and the prediction parameter as inputs, calculates a prediction residual signal, and outputs the signal to the density pattern selection circuit 15.

The density pattern selection circuit 15 is a circuit that selects the density pattern or density of the drive pulse train in subframe units or frame units. For example, in a case where selection is performed in units of subframes, for example, a prediction residual signal of one frame is first divided into a plurality of subframes, and the sum of squares of the prediction residual signal of each subframe is calculated. Next, a density (pulse interval) pattern of the drive pulse train signal in each subframe is obtained based on the sum of squares of the prediction residual signal. One example of the specific method is that two types of density patterns having the shortest pulse interval, the number of subframes having a long pulse interval and the number of subframes having a short pulse interval are set in advance, and the density of the prediction residual signal is set to two. This is a method of selecting a density pattern in which the pulse interval becomes shorter in the order of the subframe having a larger sum of squares. When the drive pulse train is changed in frame units, a short pulse interval may be selected when the sum of squares of the prediction residual signal is larger than a threshold, and a long pulse interval may be selected when the sum of squares is smaller than the threshold.

When a density pattern is selected in units of subframes in the density pattern selection circuit 15, the gain calculation circuit 27 receives information of the selected density pattern as input and predicts the gain of the drive signal for all subframes having a short pulse interval, for example. Two types are obtained using the standard deviation of the residual signal and the standard deviation of the predicted residual signal of all subframes having a long pulse interval. When the density is selected in frame units by the density pattern selection circuit 15, information on the selected density is input, and the gain of the drive signal is obtained using the standard deviation of the prediction residual signal. If this standard deviation is σ _e , the gain G is Is calculated by The obtained density pattern or density (pulse interval) and gain are encoded by encoding circuits 16 and 28, respectively, and input to multiplexer 25, and their decoded values are input to drive signal generation circuit 17. The drive signal generation circuit 17 includes a density pattern and a gain input from the encoding circuits 16 and 28, a normalized amplitude of the drive pulse input from the codebook 24, and a drive pulse input from the phase search circuit 22. Based on the phase, a drive signal having a variable density is generated in subframe units or frame units.

FIG. 4 shows an example of the drive signal generated by the drive signal generation circuit 17. G ^(m) is the gain of the drive pulse in the m-th subframe, and the normalized amplitude of the drive pulse is G ^(m) .
Assuming that g _i ^(m) , the number of pulses is Q _m , the pulse interval is D _m , the pulse phase is K _m , and the length of the subframe is L, the drive signal e _X ^(the density of which is selected in subframe units ^{) m)} (n) can be described by the following equation.

Note that the phase K _m is the leading position of the pulse in the subframe, and σ (n) is the Kronecker delta function.

Further, the codebook 24 code vectors stored in _{C i = {C i (1} ), C i (2), ... C i (l)} When is b, the driving signal e _X (n) by the following equation Can be described by

Similarly, the phase K is the leading position of the pulse in the subframe, and σ is the Kronecker delta function.

The drive signal generated by the drive signal generation circuit 17 is input to the synthesis filter 18, and the synthesized signal is output. The synthesis filter 18 has a relationship of an inverse filter with the prediction filter 14. The error between the input speech signal and the synthesized signal, which is the output of the subtraction circuit 19, is input to the square error calculation circuit 21 after its spectrum is transformed by the audibility weighting filter 20. The perceptual weight filter 20 has a transfer function This filter is used to utilize the masking effect of the audibility like the weighting filter in the conventional example, and is described in detail in the literature 2, so the description is omitted.

The square error calculation circuit 21 calculates the sum of squares of the perceptually weighted error signal for each code vector stored in the codebook 24 and for each phase of the drive pulse output from the phase search circuit 22. The result is output to the phase search circuit 22 and the amplitude search circuit 23. The amplitude search circuit 23 is a phase search circuit
For each phase of the drive pulse output from 22, the index of the code word that minimizes the sum of squares of the error signal is searched from the codebook 24, and the minimum value of the sum of squares is output to the phase search circuit 22. In addition, an index of a code word that minimizes the sum of squares is held. The phase search circuit 22 receives the information on the selected density pattern as input, changes the phase K _m of the drive pulse train in the range of 1 ≦ K _m ≦ D _m , gives the value to the drive signal generation circuit 17, and provides D _m The minimum value of the sum of squares of the error signal determined for each of the phases is received from the amplitude search circuit 23, and the phase corresponding to the smallest sum of the squares of the D _m minimum values is output to the multiplexer 25. At the same time, the amplitude search circuit 23 is notified of the phase. In the amplitude search circuit 23,
The index of the codeword corresponding to the phase is output to the multiplexer 25.

Multiplexer 25 contains prediction parameters, density patterns,
The gain, the code of the phase and the amplitude of the drive pulse are multiplexed and output to the transmission line via the output terminal 26. Note that the output of the subtraction circuit 19 may be directly input to the square error calculation circuit 21 without passing through the audibility weighting filter 20.

Next, the decoding device shown in FIG. 2 will be described. In FIG. 2, a demultiplexer 31 separates a code input from an input terminal 30 into a prediction parameter, a density pattern, a gain, and a code of a phase and an amplitude of a driving pulse. Decoding circuits 32 and 37 decode the density pattern of the driving pulse and the sign of the gain of the driving pulse, respectively, and generate a driving signal generation circuit.
Output to 33. The code book 35 is the same as the code book 24 in the encoding apparatus of FIG. 1, and outputs a code word corresponding to the index of the amplitude of the transmitted drive pulse to the drive signal generation circuit 33. Prediction parameter decoding circuit
36 decodes the code of the prediction parameter coded by the prediction parameter coding circuit 13 in FIG. The drive signal generation circuit 33, similar to the drive signal generation circuit 17 in the encoding device, has a drive whose density is variable in subframe units based on the normalized amplitude of the input drive pulse and the phase of the drive pulse. Generate a signal. Synthesis filter
34 is the same as the synthesis filter 18 in the encoding device,
Upon receiving the drive signal and the prediction parameter, the composite signal is output to the buffer 38. The buffer 38 combines the input signals for each frame and outputs the combined signal to the output terminal 39.

FIG. 5 is a block diagram of an encoding device according to a second embodiment of the present invention. This embodiment has the same function as that of the encoding apparatus shown in FIG. 1, but is capable of reducing the amount of calculation required for encoding the pulse train of the drive signal to about half.

Hereinafter, the principle of the calculation amount reduction will be briefly described. The perceptually weighted error signal e _W (n) input to the square error calculating circuit 21 in FIG. 1 is e _W (n) = [s (n) −ex _C (n) * h (n)] * W (N) (15) Here, s (n) is an input voice signal, ex _C (n) is a candidate for a driving signal, h (n) is an impulse response of the synthesis filter 18, and W (n) is an audibility weighting filter 20. , And * represents a time domain convolution operation.

When both sides of the equation (15) are z-transformed, E _W (z) = [S (z) −Ex _C (z) H (z)] W (z) (16)

H (z) and W (z) in the equation (16) are obtained by using the transfer function A (z) of the prediction filter 14, respectively. By substituting equations (17) and (18) into equation (16), the following equation is obtained.

When this is inverse z-transformed, the following equation is obtained.

e _W (n) = x (n) ex _C (n) * h _W (n) (20) where x (n) is an auditory weighting input signal, ex
_C (n) is a candidate of the driving signal, h _W (n) represents the impulse response of the perceptual weighting filter having a transfer function 1 / A (z / γ) .

Comparing Expressions (15) and (20), in Expression (15), the convolution operation of two filters per one drive signal candidate ex _C (n) is used to calculate the perceptual weighted error signal e _W (n). Is required, but it can be seen that the convolution operation of one filter is sufficient in the equation (20). In an actual encoding process, perceptually weighted error signals are calculated for hundreds to thousands of drive signal candidates, and this calculation amount occupies most of the total calculation amount of the encoding device. Therefore, when the configuration of the encoding apparatus is changed to use equation (20) instead of equation (15), the amount of calculation required for encoding is reduced on the order of 1 / Becomes easier.

In FIG. 5, the blocks denoted by the same reference numerals as those in FIG. 1 have the same functions as those in FIG. A first auditory weighting filter 51 having a transfer function of 1 / A (z / γ) receives a prediction parameter as an input,
It receives the prediction residual signal r (n) from 14 and outputs a perceptually weighted input signal x (n). On the other hand, a second perceptual weight filter 52 having the same characteristics as the first perceptual weight filter 51 receives prediction parameters as input, receives a drive signal candidate ex _C (n) from the drive signal generation circuit 17, and receives perceptual weighted synthesis. Signal candidate x
_C (n) is output. The subtraction circuit 53 outputs an error between the perceptually weighted input signal x (n) and the perceptually weighted synthesized signal candidate x _C (n), that is, a perceptually weighted error signal e _W (n) to the square error calculating circuit 21. .

FIG. 6 is a block diagram of an encoding device according to a third embodiment of the present invention. This encoder has the same function as the encoder shown in FIG. 5, but allows the gain of the drive pulse to be optimally determined in a closed-loop manner, and further improves the quality of synthesized speech. is there.

In the encoding apparatus of FIGS. 1 and 5, the gain of the driving pulse is the gain G common to all code vectors extracted from the codebook normalized using the standard deviation of the prediction residual signal of the input signal. , And the phase J and the index I of the codebook are searched.
In this method, the optimum phase J and index I are selected for the determined gain G, but the gain, phase and index are not simultaneously optimized. If gain, phase and index can be optimized simultaneously,
Further, since the drive pulse can be expressed with high accuracy, the quality of the synthesized speech is greatly improved.

Hereinafter, the principle of a method for simultaneously optimizing the gain, phase, and index efficiently will be described.

The above equation (20) can be rewritten as the following equation (21).

e _W (n) = x (n) −G _ij · x _j ⁽ⁱ⁾ (n) (21) where e _W (n) is an error signal weighted by auditory sense, x (n)
Is the input signal with auditory weight, G _ij is the index i, phase j
X _j ⁽ⁱ⁾ (n) is a drive pulse that is not multiplied by the gain of the index i and the phase j.
FIG. 9 shows auditory weighted synthesized signal candidates weighted by an auditory weight filter of a transfer function of 1 / A (z / γ). FIG. Power of perceptually weighted error signal The optimal gain G _ij is determined by setting the value ∂E _W / ∂G _ij obtained by partially differentiating with the optimal gain to zero.

In other words, equation (23) is Can be expressed as By substituting equation (26) into equation (22), the minimum value of the power of the perceptually weighted error signal can be expressed by the following equation.

The index i and the phase j that minimize the power of the perceptually weighted error signal in equation (27) are equal to the index i and the phase j that maximize ｛A _j ⁽ⁱ⁾ ｝ ² / B _j ⁽ⁱ⁾ . Therefore, in order to simultaneously obtain the optimal index I, the phase J, and the gain _GIJ , as an example, first, for the candidate for the index i and the phase j, A ⁽ⁱ⁾ _j , B ⁽ⁱ⁾ _{j are obtained} by the equations (24) and (25)
Then, a set of an index I and a phase J that maximizes ｛A _j ⁽ⁱ⁾ ｝ ² / B _j ⁽ⁱ⁾ is searched, and G is calculated using equation (26).
_What is necessary is _just to obtain _IJ and encode it.

The encoding device of FIG. 6 differs from the encoding device of FIG. 5 only in that it incorporates the above-described method of simultaneously optimizing the index, phase, and gain. The same reference numerals as in FIG. 5 denote the same parts, and a description thereof will be omitted. 6, the phase search circuit 22 receives the density pattern information and the phase update information from the index / phase selection circuit 56, and outputs the phase information j to the normalized drive signal generation circuit 58. The normalized drive signal generating circuit 58 receives the pre-normalized code vector C ⁽ⁱ⁾ (i is an index of the code vector ⁾ stored in the code book 24, the density pattern information and the phase information j, and outputs the density pattern information A normalized drive signal having a constant pulse interval within a subframe is generated by interpolating a predetermined number of zeros after each element of the code vector based on the above-described code vector, and the normalized drive signal is generated based on the input phase information j. A signal obtained by shifting the signal in the positive direction of the time axis is output to the audibility weighting filter 52 as a final output.

The inner product calculation circuit 54 calculates the inner product value A _j ⁽ⁱ⁾ of the perceptually weighted input signal x (n) and the perceived weighted synthesized signal candidate x _j ⁽ⁱ⁾ (n) according to equation (24), and calculates this as an index. • Output to the phase selection circuit 56. The power calculation circuit 55 calculates the power B _j ⁽ⁱ⁾ of the perceptually weighted synthesized signal candidate x _j ⁽ⁱ⁾ (n) by equation (25), and outputs this to the index / phase selection circuit 56. The index / phase selection circuit 56 sequentially searches for an index I and a phase J that maximize the ratio of the square of the input inner product value to the power ｛A _j ⁽ⁱ⁾ ｝ ² / B _j ^(i). And the phase update information to the codebook 24 and the phase search circuit 22. Information on the optimal index I and phase J obtained by this search is output to the multiplexer 25, and A _J ^(I) and B _J ^(I) are temporarily stored. The gain encoding circuit 57 outputs A ^(I) _J and B ^(I) from the index / phase selection circuit 56.
_{With J} as an input, the quantization and encoding of the optimal gains A _J ^(I) / B _J ^(I) are performed, and the information on the gain is output to the multiplexer 25.

FIG. 7 is a block diagram of an encoding device according to a fourth embodiment of the present invention. This encoder has the same function as the encoder of FIG. 6, but is configured to be able to reduce the amount of calculation required for the phase search of the drive signal.

In FIG. 7, a phase shift circuit 59 outputs a perceptually weighted synthesized signal candidate of phase 1 output from the perceptual weight filter 52.
With x ₁ ⁽ⁱ⁾ (n) as input, all possible phase states for index i can be easily created by simply shifting the sample point of x ₁ ⁽ⁱ⁾ (n) in the positive direction of the time axis. it can.

Now, let N _{I be} the number of index candidates in codebook 24,
Assuming that the number of phase candidates is N _J , the number of times the auditory weighting filter 52 in FIG. 6 is used is N _I × N _J per drive signal search.
Whereas the a of the order, the number of uses of the perceptual weighting filter 52 in FIG. 7 become the order of one drive signal search every N _I, can reduce the amount of computation to about 1 × N _J.

Next, fifth and sixth embodiments will be described in which the density pattern selection circuit 15 is shown in more detail, including its preprocessing portion. In the first to fourth embodiments, the prediction filter 14 has a cascade configuration of a long-time prediction filter 41 and a short-time prediction filter 42 as shown in FIG. In the following embodiment, the parameters of the long-term synthesis filter, which is the inverse filter of the long-term prediction filter, are set in a closed loop, that is, such that the root-mean-square error between the input speech signal and the synthesized signal is minimized. The required configuration is adopted. According to this configuration, the parameters are determined so that the error is minimized at the level of the synthesized signal, so that the quality of the synthesized speech is further improved.

FIGS. 8 and 9 are block diagrams of a speech encoding apparatus and a decoding apparatus according to a fifth embodiment of the present invention.

In FIG. 8, a frame buffer 101 has an input terminal 100.
Is a circuit for accumulating one frame of the audio signal inputted to each block. Each block in FIG. 8 performs the following processing for each frame using the frame buffer 101.

First, the prediction parameter calculation circuit 102 calculates a short-term prediction parameter for one frame of a speech signal using a known method (normally, this prediction parameter is 8
~ 12 are calculated). The calculation method is described in, for example, the aforementioned reference 2 (“Digital Speech 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, and
7, and outputs the decoded value P to the prediction filter 104, the perceptual weight filter 105, the influence signal 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 the signal to the auditory weighting filter 105.

The perceptual weight filter 105 calculates the decoded value P of the prediction parameter.
A signal x obtained by transforming the spectrum of the short-term prediction residual signal r with a filter configured based on The audibility weighting filter 105 is for utilizing the auditory masking effect in the same manner as the weighting filter in the conventional example.

The influence signal creation circuit 107 includes the past weighted synthesized signal from the addition circuit 112 and the decoded value P of the prediction parameter.
And outputs a past influence signal f. Specifically, it calculates a zero input response of an auditory weighting filter that uses the past weighted synthesized signal as an internal state of the filter, and outputs it as an influence signal f in preset subframe units. 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. Influence signal creation circuit 10
In the first sub-frame, an influence signal f is created by inputting a synthesized signal of the previous frame created based on the density pattern K determined in the previous frame. Subtraction circuit
A subtraction circuit 108 subtracts a signal u obtained by subtracting the past influence signal f from the perceptually weighted input signal x in subframe units.
And outputs it to the long-term vector quantization circuit 109.

The dividing circuit 113 cuts out the short-term prediction residual signal r, which is the output of the prediction filter 104, in subframe units, and outputs it to the weight filter 114. The weight filter 114 has basically the same configuration as the auditory weight filter 105, but differs from the auditory weight filter 105 in that when the short-term prediction difference signal r is input in subframe units, the internal state is reset. .

The power calculation circuit 115 calculates the power (sum of squares) of the output signal of the weight filter 114 over the frame. Let the short-term prediction residual signal of the m-th subframe be r ^(m) (n), r
^(m) If the output signal of the weight filter 114 when (n) is input is r _w (n), the power P ^(m) of the output signal of the weight filter 114 corresponding to the subframe m is expressed by the following equation. Is calculated by

Here, n is a sample point defined between subframes, and L is a subframe length (sample).

The density pattern selection circuit 116 selects one of the preset drive signal density patterns based on the power of the output signal of the weight filter 114 output from the power calculation circuit 115. More specifically, a density pattern is selected such that the density increases in the order of the subframes having higher power. For example, if there are four equal-length sub-frames, two types of density, and density patterns set as shown in the following table, the density pattern selection circuit 116 compares the powers for each sub-frame, The number K of the density pattern in which the subframe becomes dense is selected, and this is used as the density pattern information as the short-term vector quantization circuit 111 and the multiplexer 1.
Output to 17.

The long-term vector quantization circuit 109 receives the difference signal from the subtraction circuit 106, the past drive signal e _X from the drive signal holding circuit 110 described later, and the prediction parameter P from the encoding circuit 103, and calculates the difference in subframe units. The quantized output signal u of the signal u is output to the subtraction circuit 108 and the addition circuit 112, the vector gain β and the index T are output to the multiplexer 117, and the long-term drive signal t is output to the drive signal holding circuit 110. At this time, there is a relationship between t and t = t * h (h represents the impulse response of the audibility weighting filter 105 and * represents convolution).

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

A drive signal candidate for the current subframe is created using the preset index T, gain β, and past drive signal, and is input to an audibility weighting filter to create a quantized signal candidate for the difference signal u. An optimal index T ^(m) and an optimal β ^(m) are determined so that an error between the difference signal u and the quantized signal candidate is minimized. At this time, t is the drive 8 signal of the current subframe created using T ^(m) and the optimal β ^(m) ,
A signal obtained by inputting t to the perceptual weighting filter is defined as a quantized output signal u of the difference signal.

A similar method is described, for example, by PETER KROON et al.
EEE February 1988, Vol.SAC-6, pp.353-363
class of Analysis-by-Synthesic Predicative Coder
s for High Quality Speech Coding at Rates Between
4.8 and 16 kbits / s "(Reference 3), a known method similar to the method of obtaining the coefficient of the pitch predictor in a closed loop can be used, and the description is omitted here.

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 in subframe units.

The short-term vector quantization circuit 111 receives the difference signal V, the prediction parameter P, and the density pattern number K output from the density pattern selection circuit 116, and sends the difference signal quantized output signal V in subframe units to the addition circuit 112. , And outputs the short-term drive signal y to the drive signal holding circuit 110.
Here, there is a relationship of = y * h between and y.

At the same time, the short-term vector quantization circuit 111 outputs the gain G of the drive pulse train, the phase information J, and the index I of the code vector to the multiplexer 117. At this time, the parameters G, J,
Since the number of pulses N ^(m) corresponding to the density (pulse interval) of the current sub-frame (m-th sub-frame) determined by the density pattern number K must be encoded in the sub-frame, number corresponding to the number of dimensions N _D vector (number of pulses constituting one each codevector), that is, output by N ^(m) / N _D amino each current sub-frame.

For example, the frame length is 160 samples and the subframe is 4
It is assumed that the code vector is composed of 40 samples of equal length and the dimension of the code vector is 20. In this case, one of the density patterns prepared in advance is the pulse interval 1 of the first subframe, the second
Assuming that the pulse interval is 2 to the fourth subframe, the number of gains, phases, and indices output from the short-term vector quantization circuit 111 for this density pattern is 40/20 = 2 in the first subframe ( However, in this case, since the pulse interval is 1, no phase information is output.)
20/20 = 1 in the second to fourth subframes.

The specific configuration example of the short-term vector quantization circuit 111 is
Shown in the figure. In FIG. 10, a synthetic vector generation circuit 301
Represents a prediction parameter P and a preset code book 30
2 from the code vector C ⁽ⁱ⁾ (i is an index of the code vector) and the density pattern information K so that the first sample of C ⁽ⁱ⁾ becomes a preset pulse interval corresponding to the density pattern information K. Thereafter, a pulse train having density information is created by interpolating zero at a predetermined cycle, and this pulse train is synthesized by an audibility weighting filter generated from the prediction parameter P, thereby generating a synthesized vector V ₁ ⁽ⁱ⁾ .

The phase shift circuit 303 delays the combined vector V ₁ ⁽ⁱ⁾ by a predetermined number of samples based on the density pattern information K, and combines the combined vectors V ₂ ⁽ⁱ⁾ , V ₃ ⁽ⁱ⁾ _,. ⁽ⁱ⁾ ,…
Is generated and output to the inner product calculation circuit 304 and the power calculation circuit 305. The codebook 302 stores the amplitude information of the adaptive density pulse, and is configured by a memory circuit or a vector generation circuit from which a predetermined code vector C ⁽ⁱ⁾ can be derived for an index i. Inner product computation circuit 304, and a difference signal V from Figure 8 of the subtracting circuit 108, the resultant vector V _j
obtains an inner product value A _j ⁽ⁱ⁾ and ^(i), and outputs the index phase selection circuit 306. The power calculation circuit 305 calculates the composite vector
Power B obtains the _j ⁽ⁱ⁾ of V _j ^(i), and outputs the index phase selection circuit 306.

The index / phase selection circuit 306 uses the inner product value A _j ⁽ⁱ⁾ and the power B _j ⁽ⁱ⁾ to find the evaluation value ｛A _j ⁽ⁱ⁾ ｝ ² / B _j ⁽ⁱ⁾ (28) A phase J and an index I that become larger are selected from the phase candidate j and the index candidate i, and a set of the corresponding inner product value A _J ^(I) and power B _J ^(I) is obtained by the gain encoding circuit 307.
Output to Further, the index / phase selection circuit 306 further outputs the information of the phase J to the short-term drive signal generation circuit 308 and the multiplexer 117 of FIG.
Is output to the code book 302 and the multiplexer 117 of FIG.

In the gain encoding circuit 307, a ratio A _J ^(I) / B _J ^(I) (29) between the inner product value A _J ^(I) from the index / phase selection circuit 306 and the power B _J ^(I) is determined by a predetermined method. And outputs the gain information G to the short-term drive signal generation circuit 308 and the multiplexer 117 in FIG.

Equations (28) and (29) above are described, for example, by International Conference on Acoustic, Speech and
Signal Processing paper “EFFICIENT PROCEDURES FOR
FINDING THE OPTIMUM INNOVATION IN STOCHATIC CODER
S "(Reference 4) may be used.

The short-term drive signal generation circuit 308 includes density pattern information K,
Gain information G, phase information J, and code vector C ^(I) corresponding to index I are input, and density information is stored using K and C ^(I) in the same manner as in synthetic vector generation circuit 301 described above. A short-term drive signal y is generated by creating a pulse train, multiplying the pulse amplitude by a value corresponding to the gain information G, and delaying the pulse train by a predetermined number of samples based on the phase information J. This short-term drive signal y
Are output to the audibility weighting filter 309 and the drive signal holding circuit 110 in FIG. The perceptual weighting filter 309 is a filter having the same characteristics as the perceptual weighting filter 105 of FIG. 8, is formed based on the prediction parameter P, and receives the short-term drive signal y as input and outputs the quantized output of the difference signal V to the eighth input. It is output to the adder circuit 112 in the figure.

Returning to FIG. 8, the drive signal holding circuit 110 outputs the long-term drive signal t output from the long-term vector quantization circuit 109.
The short-term drive signal y output from the short-term vector quantization circuit 111 is input, and the drive signal e _X is output to the long-term vector quantization circuit 109 in subframe units. Specifically, for example, those obtained by adding the t and y for each sample in the sub-frames may be used as the driving signal e _X. The drive signal e _X of the current sub-frame 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 sub-frame. The addition circuit 112 obtains a sum signal x of the quantized outputs ^(m) and ^{(m) in} units of subframes and the past influence signal f created in the current subframe, and outputs the sum signal x to the influence signal creation circuit 107.

The parameters P, β, T, G,
The information of I, J, and K is multiplexed by the multiplexer 117 and transmitted from the output terminal 118 as a transmission code.

Next, the decoding device shown in FIG. 9 for decoding the code transmitted from the encoding device shown in FIG. 8 will be described.

In FIG. 9, a transmitted code is input to an input terminal 200. The demultiplexer 201 first separates the input code into codes of a prediction parameter, density pattern information K, gain β, gain G, phase J, index T, index I, and phase information J. Decoding circuits 202-207
Decodes the codes of the density pattern information K, the gain G, the index I, the gain β, and the index T, respectively, and outputs the codes to the drive signal generation circuit 209. Another decoding circuit 208 decodes the encoded prediction parameter,
Output to the synthesis filter 210. The drive signal generation circuit 209 is
Drive signals having different densities are generated in subframe units based on the density pattern information K by using the decoded parameters as inputs.

The drive signal generation circuit 209 is specifically configured, for example, as shown in FIG. In FIG. 11, the code book
Reference numeral 500 has the same function as the code book 302 shown in FIG. 10 in the encoder, and outputs a code vector C ^(I) corresponding to the index I to the short-term drive signal generation circuit 501. The short-term drive signal generation circuit 501
It has the same function as the short-term drive signal generation circuit 308 shown in FIG. 0, receives the density pattern information K, the phase information J and the gain G, and outputs the short-term drive signal y to the addition circuit 506. Adding circuit 506 outputs the short drive signal y and the long-term drive signal generating circuit sum signal of the long-term drive signal t generated by 502, that is, the driving signal e synthesis filter 210 of the _X drive signal buffer 503 and Figure 9 .

The drive signal buffer 503 holds the drive signal output from the adder circuit 506 up to a predetermined number of samples from the present to the past, and when an index T is input, the drive signal buffer 503 sequentially increases the sub-frame length from the drive signal T samples past. It is configured to output the corresponding number of samples. 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. Output to 506 in subframe units.

Returning to FIG. 9, the synthesis filter 210 is a filter having a frequency characteristic opposite to that of the prediction filter 104 shown in FIG. 8 in the encoding device, and outputs the synthesized signal by inputting the drive signal and the prediction parameter. I do.

The post filter 211 shapes the spectrum of the synthesized signal output from the synthesis filter 210 using the prediction parameter, the gain β, and the index T so as to subjectively reduce noise, and outputs the resultant to the buffer 212. As a specific configuration method of the post-filter, for example, a method described in the above-mentioned reference 5 may be used. Alternatively, the output of the synthesis filter 210 may be directly supplied to the buffer 212 without using the post filter 211. The buffer 212 combines the input signals on a frame-by-frame basis and outputs the synthesized audio signal to the output terminal 213.

In the above-described embodiment, the density pattern of the drive signal is selected based on the power of the signal obtained by passing the short-term prediction residual signal through the weighting filter. It can also be performed based on the power of the signal obtained by passing the prediction residual signal through a weight filter.

FIG. 12 is a block diagram of an encoding device according to a sixth embodiment of the present invention, in which a density pattern is selected based on the power of a signal obtained by passing a pitch prediction residual signal through a weighting filter. FIG. 12 shows a configuration in which a pitch analysis circuit 119 and a pitch prediction filter 120 are arranged before the division circuit 113 in FIG. The pitch analysis circuit 119 is a circuit that calculates a pitch period and a pitch gain, outputs the calculation result to the pitch prediction filter 120, and outputs the pitch prediction filter 1
20 outputs the pitch prediction residual signal to the dividing circuit 113. The pitch period and the pitch gain can be obtained by a known method, for example, an autocorrelation method or a covariance method.

FIG. 13 is a block diagram of a speech coding apparatus according to a seventh embodiment of the present invention.

In FIG. 13, the frame buffer 101 has an input terminal 100.
Is a circuit for accumulating the audio signal for one frame, which is input to each block. Each block in FIG. 13 performs the following processing for each frame using the frame buffer 101.

First, the prediction parameter calculation circuit 102 calculates a short-term prediction parameter for one frame of a speech signal using a known method (normally, this prediction parameter is 8
~ 12 are calculated). The calculation method is described in, for example, the aforementioned reference 2. The calculated prediction parameters are
It is 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 117, and outputs the decoded value P to the prediction filter 104, the perceptual weight filter 105, the influence signal Creation circuit 10
7. Output to 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 the signal to the auditory weighting filter 105.

The perceptual weight filter 105 calculates the decoded value P of the prediction parameter.
A signal x obtained by transforming the spectrum of the short-term prediction residual signal r with a filter configured based on This auditory weighting filter 105 is for using the masking effect of the auditory sense similarly to the weighting filter in the conventional example.
Description is omitted.

The influence signal creation circuit 107 includes the past weighted synthesized signal from the addition circuit 112 and the decoded value P of the prediction parameter.
And outputs a past influence signal f. More specifically, a quiescent response of a perceptual weighting filter that performs the internal state of the filter on the past weighted synthesized signal is calculated, and is output as an influence signal f in preset subframe units. 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. Influence signal creation circuit 10
In the first sub-frame, an influence signal f is created by inputting a synthesized signal of the previous frame created based on the density pattern K determined in the previous frame. Subtraction circuit
A subtraction circuit 108 subtracts a signal u obtained by subtracting the past influence signal f from the perceptually weighted input signal x in subframe units.
And outputs it to the long-term vector quantization circuit 109.

The power calculation circuit 113 calculates the power (sum of squares) of the short-term prediction residual signal output from the prediction filter 104 in subframe units, and outputs the power of each subframe to the density pattern selection circuit 122.

The density pattern selection circuit 122 selects one of preset drive signal density patterns based on the power of the short-term prediction residual signal for each subframe output from the power calculation circuit 117. More specifically, a density pattern is selected such that the density increases in the order of the subframes having higher power. For example, when there are four equal-length subframes, two types of density, and density patterns set as shown in Table 1, the density pattern selection circuit 117 compares the power for each subframe, and determines the power. The number K of the density pattern in which the maximum subframe is dense is selected, and is output to the short-term vector quantization circuit 111 and the multiplexer 117 as density pattern information.

The long-term vector quantization circuit 109 receives the difference signal u from the subtraction circuit 106, the past drive signal e _X from the drive signal holding circuit 110 described later, and the prediction parameter P from the encoding circuit 103, and receives a sub-frame unit. The quantized output signal of the difference signal u is output to the subtraction circuit 108 and the addition circuit 112, the vector gain β and the index T are output to the multiplexer 117, and the long-term drive signal t is output to the drive signal holding circuit 110. At this time, there is a relationship between t and t = t * h (h represents the impulse response of the audibility weighting filter 105 and * represents convolution).

A detailed method of obtaining the vector gain β ^(m) and the index T ^(m) (m is the number of the subframe ^{) in} subframe units may be the same as in the fifth embodiment, and thus the description is omitted.

On the other hand, in the subtraction circuit 108, the difference signal u is
And outputs the difference signal V obtained by subtracting the quantized output signal from the short-term vector quantization circuit 111.

The short-term vector quantization circuit 111 receives the difference signal V, the prediction parameter P, and the density pattern number K output from the density pattern selection circuit 122, and sends the quantization output signal of the difference signal V to the addition circuit 112 in subframe units. , And outputs the short-term drive signal y to the drive signal holding circuit 110.
Here, there is a relationship of = y * h between and y.

At the same time, the short-term vector quantization circuit 111 outputs the gain G of the drive pulse train, the phase information J, and the index I of the code vector to the multiplexer 117. At this time, the parameters G, J,
Since the number of pulses N ^(m) corresponding to the density (pulse interval) of the current sub-frame (m-th sub-frame) determined by the density pattern number K must be encoded in the sub-frame, number corresponding to the number of dimensions N _D vector (number of pulses constituting one each codevector), that is, output by N ^(m) / N _D amino each current sub-frame.

For example, the frame length is 160 samples and the subframe is 4
It is assumed that the code vector is composed of 40 samples of equal length and the dimension of the code vector is 20. In this case, one of the density patterns prepared in advance is the pulse interval 1 of the first subframe, the second
Assuming that the pulse interval is 2 to the fourth subframe, the number of gains, phases, and indices output from the short-term vector quantization circuit 111 for this density pattern is 40/20 = 2 in the first subframe ( However, in this case, since the pulse interval is 1, no phase information is output.)
20/20 = 1 in the second to fourth subframes.

The specific configuration example of the short-term vector quantization circuit 111 is
Shown in the figure. In FIG. 10, a synthetic vector generation circuit 301
Represents a prediction parameter P and a preset code book 30
2 from the code vector C ⁽ⁱ⁾ (i is an index of the code vector) and the density pattern information K so that the first sample of C ⁽ⁱ⁾ becomes a preset pulse interval corresponding to the density pattern information K. Thereafter, a pulse train having density information is created by interpolating zero at a predetermined cycle, and this pulse train is synthesized by an audibility weighting filter generated from the prediction parameter P, thereby generating a synthesized vector V ₁ ⁽ⁱ⁾ .

The phase shift circuit 303 delays the combined vector V ₁ ⁽ⁱ⁾ by a predetermined number of samples based on the density pattern information K, and combines the combined vectors V ₂ ⁽ⁱ⁾ , V ₃ ⁽ⁱ⁾ _,. ⁽ⁱ⁾ ,…
Is generated and output to the inner product calculation circuit 304 and the power calculation circuit 305. The codebook 302 stores the amplitude information of the adaptive density pulse, and is configured by a memory circuit or a vector generation circuit from which a predetermined code vector C ⁽ⁱ⁾ can be derived for an index i. The inner product calculation circuit 304
And a difference signal V from the three view subtraction circuit 108, the resultant vector V _j
obtains an inner product value A _j ⁽ⁱ⁾ and ^(i), and outputs the index phase selection circuit 306. The power calculation circuit 305 calculates the composite vector
Power B obtains the _j ⁽ⁱ⁾ of V _j ^(i), and outputs the index phase selection circuit 306.

The index / phase selection circuit 306 uses the inner product value A _j ⁽ⁱ⁾ and the power B _j ⁽ⁱ⁾ to determine the phase J and index I for which the evaluation value shown in the above equation (28) is the largest, as a phase candidate. j and an index candidate i, and outputs a corresponding set of the inner product value A _J ^(I) and the power B _J ^(I) to the gain quantization circuit 307. The index / phase selection circuit 306 further outputs the information on the phase J to the short-term drive signal generation circuit 308 and the first
Output to the multiplexer 117 shown in FIG. 3 and the information of the index I are stored in the codebook 302 and the multiplexer 117 shown in FIG.
Output to

In the gain encoding circuit 307, the ratio between the inner product value A _J ^(I) and the power B _J ^(I) from the index / phase selection circuit 306 shown in the above equation (29) is encoded by a predetermined method. The gain information G is output to the short-term drive signal generation circuit 308 and the multiplexer 117 in FIG.

The short-term drive signal generation circuit 308 includes density pattern information K,
Gain information G, phase information J, and code vector C ^(I) corresponding to index I are input, and density information is stored using K and C ^(I) in the same manner as in synthetic vector generation circuit 301 described above. A short-term drive signal y is generated by creating a pulse train, multiplying the pulse amplitude by a value corresponding to the gain information G, and delaying the pulse train by a predetermined number of samples based on the phase information J. This short-term drive signal y
Are output to the audibility weighting filter 309 and the drive signal holding circuit 110 in FIG. The perceptual weight filter 309 is a filter having the same characteristics as the perceptual weight filter 105 of FIG. 13, is formed based on the prediction parameter P, and receives the short-term drive signal y as an input and converts the quantized output of the difference signal V into the thirteenth. It is output to the adder circuit 112 in the figure.

Returning to FIG. 13, the drive signal holding circuit 110 outputs the long-term drive signal t output from the long-term vector quantization circuit 109.
The short-term drive signal y output from the short-term vector quantization circuit 111 is input, and the drive signal e _X is output to the long-term vector quantization circuit 109 in subframe units. Specifically, for example, those obtained by adding the t and y for each sample in the sub-frames may be used as the driving signal e _X. The drive signal e _X of the current sub-frame 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 sub-frame.

The adder circuit 112 outputs a quantized output u ^{(m) in} subframe units.
And a sum signal of ^(m) and the past influence signal f created in the current subframe, and outputs the sum signal to the influence signal creation circuit 107.

The parameters P, β, T, G,
The information of I, J, and K is multiplexed by the multiplexer 117 and transmitted from the output terminal 118 as a transmission code.

In the above-described seventh embodiment, the density pattern of the drive signal is selected based on the power of the short-term prediction residual signal, but may be determined based on the number of zero crossings of the short-term prediction residual signal. FIG. 14 is a block diagram of an encoding apparatus according to the eighth embodiment based on this method.

In FIG. 14, the zero-crossing number calculation circuit 123 counts the number of times that the short-term prediction residual signal r crosses r = 0 in subframe units, and outputs the value to the density pattern selection circuit 122. In this case, the density pattern selection circuit 122 selects one of the preset density patterns based on the number of zero crossings for each subframe.

Further, the density pattern can be selected based on the power or the number of zero crossings of the pitch prediction residual signal obtained by applying the pitch prediction to the short-term prediction residual signal. In Figure 15,
FIG. 16 shows an encoding apparatus according to a ninth embodiment in which a density pattern is selected based on the power of a pitch prediction residual signal. FIG. 16 shows a density pattern based on the number of zero crossings of the pitch prediction residual signal. 15 shows an encoding apparatus according to a tenth embodiment for performing selection. FIGS. 15 and 16 show a configuration in which a pitch analysis circuit 124 and a pitch prediction filter 125 are arranged before the power calculation circuit 113 and the zero-crossing number calculation circuit 123 in FIGS. 13 and 14, respectively. The pitch analysis circuit 124 is a circuit that calculates the pitch period and the pitch gain, and outputs the calculation result to the pitch prediction filter 125. The pitch prediction filter 125 outputs the pitch prediction residual signal to the power calculation circuit 113 or the number of zero-crossings calculation circuit. Output to 123. The pitch period and the pitch gain can be obtained by a known method, for example, an autocorrelation method or a covariance method.

FIG. 17 is a block diagram of a speech coding apparatus according to an eleventh embodiment of the present invention.

In FIG. 17, a frame buffer 101 has an input terminal 100.
Is a circuit for accumulating one frame of the audio signal input to each block. Each block in FIG. 17 performs the following processing for each frame using the frame buffer 101.

First, the prediction parameter calculation circuit 102 calculates a short-term prediction parameter for one frame of a speech signal using a known method (normally, this prediction parameter is 8
~ 12 are calculated). The calculation method is described in, for example, the aforementioned reference 2. The calculated prediction parameters are
It is 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 117, and outputs the decoded value P to the prediction filter 104, the perceptual weight filter 105, the influence signal Creation circuit 10
7. Output to 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 the signal to the auditory weighting filter 105.

The perceptual weight filter 105 calculates the decoded value P of the prediction parameter.
A signal x obtained by transforming the spectrum of the short-term prediction residual signal r with a filter configured based on The audibility weighting filter 105 is for utilizing the auditory masking effect in the same manner as the weighting filter in the conventional example.

The influence signal creation circuit 107 includes the past weighted synthesized signal from the addition circuit 112 and the decoded value P of the prediction parameter.
And outputs a past influence signal f. Specifically, it calculates a zero input response of an auditory weighting filter that uses the past weighted synthesized signal as an internal state of the filter, and outputs it as an influence signal f in preset subframe units. 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. Influence signal creation circuit 10
7 inputs the synthesized signal of the previous frame generated based on the density pattern K determined in the previous frame in the first sub-frame to generate the influence signal f. Subtraction circuit 10
6 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 power calculation circuit 131 calculates the mean square value of the short-term prediction residual signal output from the prediction filter 104 in subframe units, and outputs the value to the bit allocation calculation circuit 132.

The bit allocation calculation circuit 132 calculates the bit allocation of the drive signal allocated to each subframe based on the mean square value of the short-term prediction residual signal. The calculation of the bit allocation can be performed using an optimum bit allocation formula that minimizes the root mean square value of the quantization error under the condition that the total number of allocated bits is constant. For optimal bit allocation, NS
Tayant and P. Nall: “DIGITAL CODING OF WAVEFORMS”, P
RENTICE-HALL, 1984 (Reference 4)
Here, the description is omitted. If the total number of allocated bits is R bits, the number of subframes is M = 4, and the mean square value of the short-term prediction residual signal of each subframe is σi ² , i
The bit distribution bi of the subframe is calculated by the following equation.

The bit allocation table 134 stores the bit allocation value of the subframe corresponding to the density pattern of the drive signal set in advance, and outputs the pattern number and the pit allocation value to the density pattern selection circuit 133. . Next Table 2
Shows examples of density patterns and bit allocation values. Here, R = 50 and M = 4.

The density pattern selection circuit 133 includes a bit allocation calculation circuit 132
A vector B = (b ₁ , b ₂ ,... B _M ) whose elements are the output b _i of
A vector B _K (K = 1, 2 ,,...) Having the bit allocation value of each subframe stored in the bit allocation table 134 as an element
, M, and K are the density pattern numbers), and outputs the number K of the density pattern having the minimum distance to the short-term vector quantization circuit 111 and the multiplexer 117 as density pattern information.

The long-term vector quantization circuit 109 receives the difference signal u from the subtraction circuit 106, the past drive signal e _X from the drive signal holding circuit 110 described later, and the prediction parameter P from the encoding circuit 103, and receives a sub-frame unit. The quantized output signal of the difference signal u is output to the subtraction circuit 108 and the addition circuit 112, the vector gain β and the index T are output to the multiplexer 117, and the long-term drive signal t is output to the drive signal holding circuit 110. At this time, there is a relationship between t and t = t * h (h represents the impulse response of the audibility weighting filter 105 and * represents convolution). The detailed method of obtaining the vector gain β ^(m) and index T ^(m) (m is the number of the subframe ^{) in} subframe units may be the same as in the fifth embodiment.

On the other hand, in the subtraction circuit 108, the difference signal u is
And outputs a difference signal V obtained by subtracting the quantized output signal u from the short-term vector quantization circuit 111.

The short-term vector quantization circuit 111 receives as input the difference signal V, the prediction parameter P, and the density pattern number K output from the density pattern selection circuit 133, and sends the quantization output signal of the difference signal V to the addition circuit 112 in subframe units. , And outputs the short-term drive signal y to the drive signal holding circuit 110.
Here, there is a relationship of = y * h between and y.

At the same time, the short-term vector quantization circuit 111 outputs the gain G of the drive pulse train, the phase information J, and the index I of the code vector to the multiplexer 117. At this time, the parameters G, J,
Since the number of pulses N ^(m) corresponding to the density (pulse interval) of the current sub-frame (m-th sub-frame) determined by the density pattern number K must be encoded in the sub-frame, number corresponding to the number of dimensions N _D vector (number of pulses constituting one each codevector), that is, output N ^(m) / N _D pieces by.

For example, the frame length is 160 samples and the subframe is 4
It is assumed that the code vector is composed of 40 samples of equal length and the dimension of the code vector is 20. In this case, one of the density patterns prepared in advance is the pulse interval 1 of the first subframe, the second
Assuming that the pulse interval is 2 to the fourth subframe, the number of gains, phases, and indices output from the short-term vector quantization circuit 111 for this density pattern is 40/20 = 2 in the first subframe ( However, in this case, since the pulse interval is 1, no phase information is output.)
20/20 = 1 in the second to fourth subframes.

The specific configuration example of the short-term vector quantization circuit 111 is
Shown in the figure. In FIG. 10, a synthetic vector generation circuit 301
Represents a prediction parameter P and a preset code book 30
2 from the code vector C ⁽ⁱ⁾ (i is an index of the code vector) and the density pattern information K so that the first sample of C ⁽ⁱ⁾ becomes a preset pulse interval corresponding to the density pattern information K. Thereafter, a pulse train having density information is created by interpolating zero at a predetermined cycle, and this pulse train is synthesized by an audibility weighting filter generated from the prediction parameter P, thereby generating a synthesized vector V ₁ ⁽ⁱ⁾ .

The phase shift circuit 303 delays the combined vector V ₁ ⁽ⁱ⁾ by a predetermined number of samples based on the density pattern information K, and combines the combined vectors V ₂ ⁽ⁱ⁾ , V ₃ ⁽ⁱ⁾ _,. ⁽ⁱ⁾ ,…
Is generated and output to the inner product calculation circuit 304 and the power calculation circuit 305. The codebook 302 stores the amplitude information of the adaptive density pulse, and is configured by a memory circuit or a vector generation circuit from which a predetermined code vector C ⁽ⁱ⁾ can be derived for an index i. The inner product calculation circuit 304
And a difference signal V from the 7 Figure subtraction circuit 108, the resultant vector V _j
obtains an inner product value A _j ⁽ⁱ⁾ and ^(i), and outputs the index phase selection circuit 306. The power calculation circuit 305 calculates the composite vector
Power B obtains the _j ⁽ⁱ⁾ of V _j ^(i), and outputs the index phase selection circuit 306.

The index / phase selection circuit 306 uses the inner product value A _j ⁽ⁱ⁾ and the power B _j ⁽ⁱ⁾ to determine the phase J and index I for which the evaluation value shown in Expression (28) is the largest, as the phase candidate j
And index candidate i, and the corresponding inner product value
A set of A _J ^(I) and power B _J ^(I) is output to gain encoding circuit 307. The index / phase selection circuit 306 further outputs the information of the phase J to the short-term drive signal generation circuit 308 and the multiplexer 117 of FIG. 17, and outputs the information of the index I to the codebook 302 and the multiplexer 117 of FIG. .

In the gain encoding circuit 307, the inner product value A _J ^(I) and the power B _J from the index / phase selection circuit 306 shown in Expression (29) are obtained.
The ratio to ^(I) 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 117 in FIG.

The short-term drive signal generation circuit 308 includes density pattern information K,
A gain quantization value G, phase information J and a code vector C ^(I) corresponding to an index I are input, and K and C ^(I)
To generate a pulse train having density information in the same manner as in the synthetic vector generation circuit 301, multiply the pulse amplitude by a value G corresponding to the gain information, and obtain a predetermined number of samples based on the phase information J. The short-term drive signal y is generated by delaying the pulse train. This short-term drive signal y
Are output to the audibility weighting filter 309 and the drive signal holding circuit 110 in FIG. The perceptual weighting filter 309 is a filter having the same characteristics as the perceptual weighting filter 105 of FIG. 17 and is formed based on the prediction parameter P, and receives the short-term drive signal y as input and outputs the quantized output V of the difference signal V. The signal is output to the adding circuit 112 shown in FIG.

Referring back to FIG. 17, the drive signal holding circuit 110 outputs the long-term drive signal t output from the long-term vector quantization circuit 109.
The short-term drive signal y output from the short-term vector quantization circuit 111 is input, and the drive signal e _X is output to the long-term vector quantization circuit 109 in subframe units. Specifically, for example, those obtained by adding the t and y for each sample in the sub-frames may be used as the driving signal e _X. The drive signal e _X of the current sub-frame 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 sub-frame.

The adder circuit 112 outputs a quantized output u ^{(m) in} subframe units.
And a sum signal of ^(m) and the past influence signal f created in the current subframe, and outputs the sum signal to the influence signal creation circuit 107.

The parameters P, β, T, G,
The information of I, J, and K is multiplexed by the multiplexer 117 and transmitted from the output terminal 118 as a transmission code.

In the above-described eleventh embodiment, the selection of the density pattern of the drive signal is performed based on the power of the short-term prediction residual signal. However, the selection may be performed based on the number of zero crossings of the short-term prediction residual signal. FIG. 18 shows a block diagram of an encoding device according to a twelfth embodiment based on this method.

In FIG. 18, a zero-crossing number calculation circuit 135 counts the number of times that the short-term prediction residual signal r crosses r = 0 for each subframe, and outputs the value via a bit allocation calculation circuit 132 to a density pattern selection circuit 133. Output to In this case, the density pattern selection circuit 133 selects one of the preset density patterns based on the number of zero crossings for each subframe.

Further, the density pattern can be selected based on the power or the number of zero crossings of the pitch prediction residual signal obtained by applying the pitch prediction to the short-term prediction residual signal. In FIG. 19,
FIG. 20 shows an encoding apparatus according to a thirteenth embodiment in which a density pattern is selected based on the power of a pitch prediction residual signal, and FIG. 20 shows a density pattern based on the number of zero crossings of the pitch prediction residual signal. An example of the encoding device according to the fourteenth example in which selection is performed is shown. FIGS. 19 and 20 show a pitch analysis circuit before the power calculation circuit 131 and the zero-crossing number calculation circuit 135 in FIGS. 17 and 18, respectively.
136 and a pitch prediction filter 137 are arranged. The pitch analysis circuit 136 is a circuit that calculates the pitch period and the pitch gain, and outputs the calculation result to the pitch prediction filter 137. The pitch prediction filter 137 outputs the pitch prediction residual signal to the power calculation circuit 131 or the number of zero crossings. 135
Output to The pitch period and the pitch gain can be obtained by a known method, for example, an autocorrelation method or a covariance method.

FIG. 21 is a block diagram of a speech coding apparatus according to a fifteenth embodiment of the present invention.

In FIG. 21, a frame buffer 101 has an input terminal 100.
21 is a circuit for accumulating one frame of audio signal input thereto. Each block in FIG. 21 performs the following processing for each frame using the frame buffer 101.

First, the prediction parameter calculation circuit 102 calculates a short-term prediction parameter for one frame of a speech signal using a known method (normally, this prediction parameter is 8
~ 12 are calculated). The calculation method is described in, for example, the aforementioned reference 2. The calculated prediction parameters are
It is 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 117, and outputs the decoded value P to the prediction filter 104, the perceptual weight filter 105, the influence signal Creation circuit 10
7. Output to 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 the signal to the auditory weighting filter 105.

The perceptual weight filter 105 calculates the decoded value P of the prediction parameter.
A signal x obtained by transforming the spectrum of the short-term prediction residual signal r with a filter configured based on The audibility weighting filter 105 is for utilizing the auditory masking effect in the same manner as the weighting filter in the conventional example.

The influence signal creation circuit 107 includes the past weighted synthesized signal from the addition circuit 112 and the decoded value P of the prediction parameter.
And outputs a past influence signal f. Specifically, it calculates a zero input response of an auditory weighting filter that uses the past weighted synthesized signal as an internal state of the filter, and outputs it as an influence signal f in preset subframe units. 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. Influence signal creation circuit 10
In the first sub-frame, an influence signal f is created by inputting a synthesized signal of the previous frame created based on the density pattern K determined in the previous frame. Subtraction circuit
A subtraction circuit 108 subtracts a signal u obtained by subtracting the past influence signal f from the perceptually weighted input signal x in subframe units.
And outputs it to the long-term vector quantization circuit 109.

The power calculation circuit 141 calculates the power (sum of squares) of the short-term prediction residual signal output from the prediction filter 104 for each subframe, and weights the power of each subframe.
Output to 2. The weighting circuit 142 outputs a function value obtained by performing non-linear weighting on the power of each subframe. When the number of subframes is 4, for example, the weighting function W _i
(I = 1,2,3,4) is set as follows.

_{W 1 = 0.4 W 2 = 0.25} W 3 = 0.20 W 4 = 0.15 The weighting is a process for the importance of each sub-frame in the sub-frame is not uniform. The output signal of the weighting circuit 142 is output to the density pattern selection circuit 143.

The density pattern selection circuit 143 selects one of preset drive signal density patterns based on a weighting function value of the power of the short-term prediction step signal for each sub-frame output from the weighting circuit 142. Specifically, a density pattern is selected such that the density increases in the order of subframes having the largest weighting function value. For example, when there are four equal-length subframes, two types of density, and density patterns set as shown in Table 1, the density pattern selection circuit 143
Compares the power weighting function value for each sub-frame, selects the number K of the density pattern in which the sub-frame with the maximum value is dense, and uses it as the density pattern information as the short-term vector quantization circuit 111 and the multiplexer. Output to 117.

The long-term vector quantization circuit 109 receives the difference signal u from the subtraction circuit 106, the past drive signal e _X from the drive signal holding circuit 110 described later, and the prediction parameter P from the encoding circuit 103, and receives a sub-frame unit. The quantized output signal of the difference signal u is output to the subtraction circuit 108 and the addition circuit 112, the vector gain β and the index T are output to the multiplexer 117, and the long-term drive signal t is output to the drive signal holding circuit 110. At this time, there is a relationship of u = t * h (h is the impulse response of the auditory weighting filter 105 and * is convoluted) with t.

The detailed method of obtaining the vector gain β ^(m) and index T ^(m) (m is the number of the subframe ^{) in} subframe units may be the same as in the fifth embodiment.

On the other hand, in the subtraction circuit 108, the difference signal u is
And outputs the difference signal V obtained by subtracting the quantization output signal from the short-term vector quantization circuit 111.

The short-term vector quantization circuit 111 receives as input the difference signal V, the prediction parameter P, and the density pattern number K output from the density pattern selection circuit 143, and sends the quantized output signal of the difference signal V to the addition circuit 112 in subframe units. , And outputs the short-term drive signal y to the drive signal holding circuit 110.
Here, there is a relationship of = y * h between and y.

At the same time, the short-term vector quantization circuit 111 outputs the gain G of the drive pulse train, the phase information J, and the index I of the code vector to the multiplexer 117. At this time, the parameters G, J,
Since the number of pulses N ^(m) corresponding to the density (pulse interval) of the current sub-frame (m-th sub-frame) determined by the density pattern number K must be encoded in the sub-frame, number corresponding to the number of dimensions N _D vector (number of pulses constituting one each codevector), that is, output by N ^(m) / N _D amino each current sub-frame.

For example, the frame length is 160 samples and the subframe is 4
It is assumed that the code vector is composed of 40 samples of equal length and the dimension of the code vector is 20. In this case, one of the density patterns prepared in advance is the pulse interval 1 of the first subframe, the second
Assuming that the pulse interval is 2 to the fourth subframe, the number of gains, phases, and indices output from the short-term vector quantization circuit 111 for this density pattern is 40/20 = 2 in the first subframe ( However, in this case, since the pulse interval is 1, no phase information is output.)
20/20 = 1 in the second to fourth subframes.

The specific configuration example of the short-term vector quantization circuit 111 is
Shown in the figure. In FIG. 10, a synthetic vector generation circuit 301
Represents a prediction parameter P and a preset code book 30
2 from the code vector C ⁽ⁱ⁾ (i is an index of the code vector) and the density pattern information K so that the first sample of C ⁽ⁱ⁾ becomes a preset pulse interval corresponding to the density pattern information K. Thereafter, a pulse train having density information is created by interpolating zero at a predetermined cycle, and this pulse train is synthesized by an audibility weighting filter generated from the prediction parameter P, thereby generating a synthesized vector V ₁ ⁽ⁱ⁾ .

The phase shift circuit 303 delays the combined vector V ₁ ⁽ⁱ⁾ by a predetermined number of samples based on the density pattern information K, and combines the combined vectors V ₂ ⁽ⁱ⁾ , V ₃ ⁽ⁱ⁾ _,. ⁽ⁱ⁾ ,…
Is generated and output to the inner product calculation circuit 304 and the power calculation circuit 305. The codebook 302 stores the amplitude information of the adaptive density pulse, and is configured by a memory circuit or a vector generation circuit from which a predetermined code vector C ⁽ⁱ⁾ can be derived for an index i. The inner product calculation circuit 304
And a difference signal V from the 1 Figure subtraction circuit 108, the resultant vector V _j
obtains an inner product value A _j ⁽ⁱ⁾ and ^(i), and outputs the index phase selection circuit 306. The power calculation circuit 305 calculates the composite vector
Power B obtains the _j ⁽ⁱ⁾ of V _j ^(i), and outputs the index phase selection circuit 306.

The index / phase selection circuit 306 uses the inner product value A _j ⁽ⁱ⁾ and the power B _j ⁽ⁱ⁾ to determine the phase J and index I for which the evaluation value shown in the above equation (28) is the largest, as a phase candidate. j and an index candidate i, and outputs a corresponding set of the inner product value A _J ^(I) and the power B _J ^(I) to the gain encoding circuit 307. The index / phase selection circuit 306 further outputs the information of the phase J to the short-term drive signal generation circuit 308 and the second
The information of the index I is output to the multiplexer 117 shown in FIG.
Output to

In the gain encoding circuit 307, the inner product value A _J ^(I) and the power B _J from the index / phase selection circuit 306 shown in Expression (29) are obtained.
The ratio with ^(I) 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 117 in FIG.

The short-term drive signal generation circuit 308 includes density pattern information K,
Gain information G, phase information J, and code vector C ^(I) corresponding to index I are input, and density information is stored using K and C ^(I) in the same manner as in synthetic vector generation circuit 301 described above. A short-term drive signal y is generated by creating a pulse train, multiplying the pulse amplitude by a value corresponding to the gain information G, and delaying the pulse train by a predetermined number of samples based on the phase information J. This short-term drive signal y
Are output to the audibility weighting filter 309 and the drive signal holding circuit 110 in FIG. The perceptual weight filter 309 is a filter having the same characteristics as the perceptual weight filter 105 of FIG. 21 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 of the difference signal V to the 21st. It is output to the adder circuit 112 in the figure.

Returning to FIG. 21, the drive signal holding circuit 110 outputs the long-term drive signal t output from the long-term vector quantization circuit 109.
The short-term drive signal y output from the short-term vector quantization circuit 111 is input, and the drive signal e _X is output to the long-term vector quantization circuit 109 in subframe units. Specifically, for example, those obtained by adding the t and y for each sample in the sub-frames may be used as the driving signal e _X. The drive signal e _X of the current sub-frame 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 sub-frame.

The adder circuit 112 outputs a quantized output u ^{(m) in} subframe units.
And a sum signal of V ^(m) and the past influence signal f created in the current subframe, and outputs the sum signal to the influence signal creation circuit 107.

The parameters P, β, T, G,
The information of I, J, and K is multiplexed by the multiplexer 117 and transmitted from the output terminal 118 as a transmission code.

In the fifteenth embodiment described above, the density pattern of the drive signal is selected based on the function value obtained by applying nonlinear weighting to the power of the short-term prediction residual signal. It can also be performed based on a function value subjected to nonlinear weighting. You may use what was done. FIG. 22 is a block diagram of a speech coding apparatus according to a sixteenth embodiment based on this method.

In FIG. 22, the zero-crossing number calculation circuit 144 counts the number of times that the short-term prediction residual signal r crosses r = 0 in subframe units, and outputs the value to the weighting circuit 142. In this case, the density pattern selection circuit 143 selects one of the preset density patterns based on the weighting function value of the number of zero crossings for each subframe.

Further, the density pattern can be selected based on a function value obtained by weighting the power or the number of zero crossings of the pitch prediction residual signal obtained by applying pitch prediction to the short-term prediction residual signal. FIG. 23 shows an encoding device according to a seventeenth embodiment in which density pattern selection is performed based on a function value obtained by performing nonlinear weighting on the power of the pitch prediction residual signal, and FIG. An encoding apparatus according to an eighteenth embodiment is configured to select a density pattern based on a function value obtained by weighting the number of zero crossings of a pitch prediction residual signal nonlinearly.

FIGS. 23 and 24 show the power calculation circuit 141 and the zero-crossing number calculation circuit 144 in FIGS. 21 and 22, respectively.
, A pitch analysis circuit 145 and a pitch prediction filter 146 are arranged. The pitch analysis circuit 145 is a circuit that calculates a pitch period and a pitch gain, and outputs the calculation result to the pitch prediction filter 146. The pitch prediction filter 146 outputs the pitch prediction residual signal to the power calculation circuit 141 or the zero-crossing number calculation circuit. Output to 144. The pitch period and the pitch gain can be obtained by a known method, for example, an autocorrelation method or a covariance method.

In the above-described fifth to eighteenth embodiments, the density of the drive pulse train is selected in units of subframes, but may be selected in units of frames. A pattern may be used. In that case, Table 1,
The subframe number in Table 2 is read as a frame number. However, although the number of density patterns is four in the first and second, it goes without saying that the number of density patterns can be arbitrarily selected.

Further, regarding the selection of the density of the drive pulse train, the first to eighteenth
It is also possible to combine the selection of the sub-frame unit described in the embodiment with the selection of the frame unit. That is, the pulse density is set for each sub-frame in accordance with a predetermined density pattern, and the drive signal is constituted by a drive pulse train whose density pattern is variable for each frame.

More specifically, the density pattern selection circuits 15, 116, 122, 133, and 143 described in each embodiment are provided with tables storing a plurality of (predetermined in this example, six) density patterns, for example, as shown in Table 3 below. However, this Table 3
Is a case where the number of sub-frames is four, as in the previous case. In addition to the four patterns shown in Table 1, a pattern in which the drive pulse trains of all the sub-frames in one frame are all sparse and a pattern in which all are dense , A total of six patterns are set.

Then, from these density patterns, one density pattern is selected by a method similar to the method of selecting the density pattern in each embodiment. For example, taking the case where the same method as in the first embodiment is used as an example, a code vector and a phase of the optimum pulse amplitude are searched for each density pattern, and the minimum value of the sum of squares of the error at that time is transmitted over the frame. Ask. As a result, a density pattern that gives the smallest minimum value among the obtained minimum values is selected.

According to this embodiment, the advantage of changing the density of the driving pulse train on a subframe basis is the improvement of the quality of the synthesized speech in the portion where the nature of the speech changes abruptly, and the advantage of changing on the frame basis There is an advantage that the effect of improving the quality of the synthesized voice is obtained both in the portion where the nature of the voice changes relatively slowly, and the quality of the synthesized voice is uniformly improved in the portion where the voice changes. .

In the above-described embodiment, the density (pulse interval) of the drive pulse train is set to two types, ie, sparse and dense. However, the density may be changed in three or more stages.

[Effects of the Invention] According to the present invention, the pulse density of a drive pulse constituting a drive signal for driving a synthesis filter is densely set in a sub-frame or a frame including important information or a large amount of information, and not so. By changing each sub-frame or frame such that it is coarse in a sub-frame or frame, or adaptively changing each sub-frame or frame, even at a low bit rate such as 10 kb / s or less, the quality can be improved. High synthesized speech can be reproduced.

[Brief description of the drawings]

1 and 2 are block diagrams respectively showing the configuration of an encoding device and a decoding device according to a first embodiment of the present invention, and FIG. 3 is a block diagram showing an example of the configuration of a prediction filter in FIG. FIG. 4 is a diagram showing an example of a drive signal generated in the embodiment, and FIGS. 5, 6, and 7 are diagrams according to the second, third, and fourth embodiments of the present invention, respectively. 8 and 9 are block diagrams showing the configurations of an encoding device and a decoding device, respectively, according to a fifth embodiment of the present invention. FIG. 10 is a block diagram showing the configuration of the encoding device. FIG. 11 is a block diagram showing a configuration example of a short-term vector quantization circuit in FIG. 11, FIG. 11 is a block diagram showing a configuration example of a drive signal generation circuit in FIG. 9, and FIGS. 25 is a block diagram illustrating a configuration of a coding apparatus according to the thirteenth to eighteenth embodiments. Block diagram showing the configuration of an encoder according to the prior art, Figure 26 is a block diagram similarly showing the configuration of a decoder, Figure 27 is a diagram showing an example of a driving signal according to a conventional method. 14,104 prediction filter 15,116,122,133,143 density pattern selection circuit 17 drive signal generation circuit 20,105 hearing weight filter 21 square error calculation circuit 55,115,121,141 power calculation circuit 114,142 weight filter 119,124,136,145 pitch analysis Circuits 120, 125, 137, 146 Pitch prediction circuits 123, 135, 144… Zero crossing number calculation circuit 132… Bit allocation calculation circuit 134… Bit allocation table

──────────────────────────────────────────────────続 き Continued on the front page (31) Priority claim number Japanese Patent Application No. 2-25840 (32) Priority date February 5, 1990 (1990.2.5) (33) Priority claim country Japan (JP) (56) References 1989 IEICE Autumn National Convention Lecture Papers, Supplement 1, “A-3 CELP Coding Scheme Based on Adaptive Density Pulse Train Model”, p. 1-3, (August 15, 1989) Proceedings of the IEICE Autumn Meeting, 1988, Supplement A-1, “Study on 8 kbps speech coding using A-6 ARMA model” , P. A-1-6, (published August 15, 1988) Proceedings of the 1989 IEICE Spring Conference, Volume 1, "A-5 Speech Coding Based on ARM A Model and Adaptive Density Pulse Train Model" Method ", p. 1-5, (issued March 15, 1989) Proceedings of the 1990 IEICE Spring Conference, Volume 1, “Code Error Sensitivity of Parameters in A-226 ADP-CELP System,” p. 1-226, (published March 5, 1990) Proceedings of the 1990 IEICE Spring Conference, Volume 1, “Analysis of DSP Implementation of SA-5-7 ADP-CELP Coding Scheme,” p. . 1-433, (published March 5, 1990) Proceedings of the 1990 IEICE National Conference, Volume 3, “Development of 4.8 kbps ADP-CELP Speech Codec”, p. 3-285, (published September 15, 1990) Proceedings of IE 1989 International Conference on Acoustics, Speech and Signal Processing, Vol. 1, "S4.8 ARMA Model Based Speed Coding at 8 kb / s" p. 148-151 Proceedings of IEEE 1990 International Conference on Acoustics, Speech and Signal Processing, Vol. 1, "S1.8 CELP Coding with an Adaptive Density Pulse Excitation Model" p. 29-32 Conference Record of IEEE Global Telecommunications Conference, GLOBECOM '91, Vol. 1 of 3, "Improvement of ADP-CE LP Speech Coding at 4 Kbits / s", p. 53.2.1-53.2.5, Phoenix, Arizona, December 2-5, 1991 (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19 / 00-21/06 INSPEC (DIALOG) JICST File (JOIS) IEEE / IEEE Electronic Library Online

Claims (12)

(57) [Claims] 1. A drive signal generating means for generating a drive signal comprising a drive pulse train whose pulse density is variable in a predetermined section unit, a synthesis filter driven by the drive signal, an output signal of the synthesis filter and an input Means for determining the amplitude or the amplitude and phase of the drive pulse train so that the power of the perceptual weighted error signal with the audio signal is minimized, and passing the short-term prediction residual signal to the input audio signal through a perceptual weight filter Density determining means for determining a pulse density of the driving pulse train based on the obtained signal. 2. A drive signal generating means for generating a drive signal composed of a drive pulse train whose pulse density is variable in predetermined intervals, a synthesis filter driven by the drive signal, an output signal of the synthesis filter and an input Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptual weighted error signal with the voice signal is minimized, and passing the pitch prediction residual signal for the input voice signal through a perceptual weight filter Density determining means for determining a pulse density of the driving pulse train based on the obtained signal. 3. A drive signal generating means for generating a drive signal comprising a drive pulse train whose pulse density is variable in a predetermined section unit; a synthesis filter driven by the drive signal; an output signal of the synthesis filter; Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptually weighted error signal with the audio signal is minimized, and performing pitch prediction on the short-term prediction residual signal for the input audio signal. And a density determining means for determining a pulse density of the drive pulse train based on a signal obtained by passing a pitch prediction residual signal obtained through an auditory weighting filter. 4. A drive signal generating means for generating a drive signal consisting of a drive pulse train whose pulse density is variable in predetermined intervals, a synthesis filter driven by the drive signal, an output signal of the synthesis filter and an input. Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptually weighted error signal with the audio signal is minimized, and performing pitch prediction on the short-term prediction residual signal for the input audio signal. And a density determining means for determining a pulse density of the drive pulse train based on a pitch prediction residual signal obtained. 5. The pulse density is equally spaced within a predetermined section,
A drive signal generating means for generating a drive signal composed of a drive pulse train that is variable in units of the predetermined section; a synthesis filter driven by the drive signal; and an audible weight of an output signal of the synthesis filter and an input audio signal. Means for determining the amplitude or the amplitude and phase of the drive pulse train so that the power of the error signal is minimized; and assigned to each subframe of the drive signal based on a short-term prediction residual signal for the input audio signal. A speech coding apparatus comprising: means for calculating a bit allocation value; and density determining means for determining a pulse density of the drive pulse train based on the bit allocation value calculated by the means. 6. A drive signal generating means for generating a drive signal comprising a drive pulse train whose pulse density is variable in a predetermined section unit; a synthesis filter driven by the drive signal; an output signal of the synthesis filter; Means for determining the amplitude or the amplitude and phase of the drive pulse train so that the power of the perceptual weighted error signal with the audio signal is minimized; and the drive signal based on a pitch prediction residual signal for the input audio signal. And a density determining means for determining a pulse density of the driving pulse train based on the bit allocation value calculated by the means. Device. 7. A drive signal generating means for generating a drive signal composed of a drive pulse train whose pulse density is variable in a predetermined section unit; a synthesis filter driven by the drive signal; an output signal of the synthesis filter; Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptually weighted error signal with the audio signal is minimized, and performing pitch prediction on the short-term prediction residual signal for the input audio signal. Means for calculating a bit allocation value assigned to each subframe of the drive signal based on the obtained pitch prediction residual signal; and determining a pulse density of the drive pulse train based on the bit allocation value calculated by the means. A speech coding apparatus comprising: a density determining unit. 8. A drive signal generating means for generating a drive signal composed of a drive pulse train whose pulse density is variable in predetermined intervals, a synthesis filter driven by the drive signal, an output signal of the synthesis filter and an input. Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptual weighted error signal with the audio signal is minimized, and the power or the number of zero crossings of the short-term prediction residual signal for the input audio signal. And a density determining means for determining a pulse density of the drive pulse train based on a function value subjected to nonlinear weighting. 9. A drive signal generating means for generating a drive signal composed of a drive pulse train whose pulse density is variable in predetermined intervals, a synthesis filter driven by the drive signal, an output signal of the synthesis filter and an input. Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptual weighted error signal with the audio signal is minimized, and the power or the number of zero crossings of the pitch prediction residual signal for the input audio signal And a density determining means for determining a pulse density of the drive pulse train based on a function value subjected to nonlinear weighting. 10. A drive signal generating means for generating a drive signal comprising a drive pulse train whose pulse density is variable in predetermined intervals, a synthesis filter driven by the drive signal, an output signal of the synthesis filter and an input. Means for determining the amplitude or amplitude and phase of the drive pulse train so that the power of the perceptually weighted error signal with the audio signal is minimized, and performing pitch prediction on the short-term prediction residual signal for the input audio signal. And a density determining means for determining a pulse density of the driving pulse train based on a function value obtained by nonlinearly weighting the power or the number of zero crossings of the pitch prediction residual signal to be obtained. 11. The driving signal generating means according to claim 1, wherein said driving signal generating means comprises: a driving pulse train in which a frame is divided into a plurality of sub-frames, and a pulse density is variable in sub-frame units or frame units at equal intervals in each sub-frame or each frame. 3. A driving signal comprising:
11. The speech encoding device according to any one of 10. 12. The drive signal generating means, wherein the frame is divided into a plurality of subframes, the pulse density is set in subframe units according to a predetermined density pattern, and the density pattern is variable in frame units. 11. The speech encoding apparatus according to claim 1, wherein a drive signal including a pulse train is generated, and the density determination unit determines a density pattern of the drive pulse train.