JPH0378637B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a low bit rate waveform encoding system for audio signals, and particularly to an encoding apparatus that allows the amount of transmitted information to be 10 kbit/sec or less.

As an effective method for encoding an audio signal with a transmission information amount of approximately 10k bits/second or less, the driving excitation signal sequence of the audio signal is conditioned on the minimum error between the signal reproduced using the sequence and the input signal. A method of searching every short period of time is known. B.S. Attar of Bell Telephone Laboratories, USA
A-b-S (Analysis-by-
The method obtained using the (Synthesis) method is effective.
An explanation of this can be found in the 1982 ICASSP Proceedings 614.
~617 Mitsugu (Reference 1), so the explanation is omitted here. The conventional method disclosed in Reference 1 uses the A-b-S method to obtain the pulse sequence, and therefore has the drawback of requiring a very large amount of calculations. On the other hand, in patent application number 1983-231603 (Document 2),
A method has been proposed to greatly reduce the amount of calculation required to obtain the above pulse sequence. It has been reported that these methods can provide good playback quality at transmission rates of 10k bits/second or less.
The conventional method disclosed in Document 2 (Patent Application No. 1983-231603) will be briefly explained. A driving sound source sequence consisting of K pulse sequences within one frame is expressed as follows.

d(n)= _K 〓 ^k=1 gkδ(n-lk), n=0, 1,...,N-1 -(1) Here, δ(·) is Kronetzker's δ. N
is the frame length, and gk is the amplitude of the pulse standing at position lk. The reproduced signal x〓(n) obtained by inputting d(n) to the synthesis filter is as follows, assuming that the prediction coefficient of the synthesis filter is αi (i=1, . . . , M, M is the order of the synthesis filter). It can be written as follows.

x〓(n)=d(n)+ _M〓 ⁱ⁼¹ αix〓(n−i) −(2) 1 of input audio signal x(n) and playback signal x〓(n)
The weighted squared error within a frame is J= _N-1 〓 ⁱ⁼⁰ ((x(n)-x〓(n))*w(n)) ² -(3). Here, * is a symbol for convolution integral, and w(n) represents a weighted function. Equation (3) is
The Z transformations of x(n), x〓(n), and w(n) are respectively
(z), X〓(z), and W(z), it is expressed as follows.

J=｜X(z)W(z)−X〓(z)W(z)｜ ²
−(4) Here |・| represents the absolute value. Also, from the relationship in equation (2), X〓(z) becomes as follows.

X〓(z)=H(z)D(z)−(5) H(z) is the Z transformation of the synthesis filter, and Z(z) is the Z transformation of the driving sound source. Substituting (5) into (4), J=|X(z)W(z)−H(z)W(z)D
(z) | ² −(6).

Therefore, the inverse Z-transformed signals of X(z)W(z) and N(z)W(z) are respectively xw(n)=x(n)*w
(n) and hw(n)=h(n)*w(n), (6)
becomes as follows.

J= _N-1 〓 ⁿ⁼⁰ (xw(n)= _K 〓 ^k=1 gkhw(n-lk)) ² −(7) Amplitude of the sound source pulse sequence that minimizes equation (7)
To find gk and position mk, the equation (7) is partially differentiated with respect to gk and set to 0, that is, Take advantage of the relationship between

Here, ψxh(・) is the interaction function sequence calculated from xw(n) and hw(n), and ψhh(・) is hw(n)
The autocorrelation sequences of are expressed as follows. Note that ψhh(·) is also called the covariance function.

ψxh (lk) = _N-1 〓 〓 ⁿ⁼⁰ xw (n) hw (n-lk) = ψhx (-lk) 0≦lk≦N
−1(9) ψhh(li, lj)= _N-(li-lj)+1 〓 〓 ⁿ⁼⁰ hw(n-li) hw(n-li) 0≦li, lj≦N-1(10
) The conventional method determines the amplitude and position of the k-th pulse by regarding gk in (8) as a function of only lk. In other words, maximize |gk| in (8)
Let lk be the position of the k-th pulse, and gk at that time
Let be the amplitude of the k-th pulse. In this method, if gk is exactly a function of only lk, the sound source pulse sequence that minimizes equation (7) is calculated, but this is not the case for actual audio signals, and in general
gk is a function such as l ₁ , l ₂ , ..., lk.

FIG. 1 is a block diagram showing an embodiment of the conventional method disclosed in Document 2. FIG. 2 shows the amplitude of the sound source pulse sequence calculated by the sound source pulse sequence calculation circuit 140.
12 is a flowchart showing a processing procedure for determining gk and position lk. In FIG. 1, each component performs processing for each frame. 100 indicates an encoder input terminal,
An A/D converted audio signal sequence x(n) is input. 110 is a buffer memory circuit that stores one frame worth of audio signal series. The K parameter calculation circuit 180 receives the audio signal x(n) stored in the buffer memory circuit 110 and calculates a predetermined number of K parameters Ki (1≦i≦M). This value is the K parameter encoding circuit 19
Output to 0. K parameter encoding circuit 190
encodes Ki based on, for example, a predetermined number of quantization bits, and outputs the code Iki to multiplexer 160. Further, the K parameter encoding circuit 190 decodes Iki and decodes the decoded value.
Ki′ (1≦i≦M) by the impulse response calculation circuit 12
0 and is output to the weighting circuit 200. The weighting circuit 200 inputs the input audio signal x(n) and the K-parameter decoded value Ki′, calculates the above-mentioned xw(n) using a weighting function w(n) that depends on the frequency characteristics of the synthesis filter, The obtained xw(n) is output to the cross-correlation calculation circuit 135. The weighting function w(n) used here is, for example, the Z-transformation W(z) using the prediction parameter αi of the synthesis filter.
By a real constant r satisfying 0≦r≦1, W(z)
= (1- _M 〓 ⁱ⁼¹ αiZ ^-i / (1- _M 〓 ⁱ⁼¹ αir ^j Z _-i ) is adopted. The impulse response circuit 120 is
Input Ki, calculate the above-mentioned hw(n) (convolution integral of the impulse response and the same weighted function as above) for the specified number of samples, and find the hw
(n) is output to the covariance function calculation circuit 130 and the correlation function calculation circuit 135. The covariance function calculation circuit 130 calculates a predetermined number of samples.
Input hw(n) and use ψhh(li,
lj) (0≦li, lj≦N-1) and outputs this to the sound source pulse sequence calculation circuit 140. Next, the sound source pulse sequence calculation circuit will be explained. The sound source pulse sequence calculation circuit 140 includes the mutual function calculation circuit 13
5 to ψxh (lk) (0≦lk≦N−1), and ψhh (li, lj) (0≦li, lj≦) from the covariance function calculation circuit 130.
N-1), and calculate the amplitude of the sound source pulse sequence using the pulse calculation algorithm (8) above.
Calculate gk and position lk. FIG. 2 is a flowchart showing the processing procedure performed by the sound source pulse sequence calculation circuit 140. For the first pulse, in equation (8), K = 1, amplitude g ₁ is a function of position l ₁ , g ₁ =
Expressed as ψxh (l ₁ )/ψhh (l ₁ , l ₁ ). Next, |
Select l ₁ that maximizes g ₁ |, and then set l ₁ and g ₁ to 1
th pulse position and amplitude. For the second pulse, set K=2 in equation (8), select l ₂ that maximizes |g ₂ |, and let l ₂ and g ₂ at that time be the position and amplitude of the second pulse. The third and subsequent pulses are calculated in the same manner until the predetermined number of pulses is reached. In FIG. 2, 1 initializes to 1 a calculation counter that calculates the number of pulses. 2 is a comparison, in which it is determined whether the number of pulses is larger or smaller than a predetermined number, and if it is larger than the predetermined number, the pulse sequence calculation process is finished. 3 calculates equation (8), where l ₁ ,..., l _k-1 and g ₁ ,...,
Assuming that g _k-1 is known, find the lk that maximizes |gk|,
gk at that time is output as the amplitude and position of the k-th pulse. 4 is an adder that increments by one the contents of a calculation counter that calculates the number of pulses. This concludes the explanation of the sound source pulse calculation circuit 140.

Returning to FIG. 1, the encoding circuit 150 receives the amplitude gk and position lk of the pulse sequence output from the excitation pulse calculation circuit 140 and encodes them.
Conventionally well-known methods can be used for encoding the amplitude gk and position lk. amplitude gk
For example, a method can be considered in which the maximum value of the amplitude of a pulse system example within one frame is used as a normalization coefficient, each pulse amplitude is normalized with this value, and then quantized and encoded. As for the position lk, it is conceivable to use run-length encoding, which is well known in the field of facsimile signal encoding, for example. This represents the length of the code "0" using a predetermined code sequence. The multiplexer 160 inputs the output code of the K-parameter encoding circuit 190 and the output code of the encoding circuit 150, combines them, and outputs them from the transmission side output terminal 170 to the communication path.

The method of searching for a driving sound source pulse sequence in the conventional method of Document 2 has been described above.

In the conventional method of Reference 2, in an algorithm for determining the amplitude and position of a sound source pulse sequence, it is assumed that the amplitude of a pulse is a function only of the position at which the pulse stands. However, for actual audio signals, the above assumption does not hold, and as shown in the above equation (8) used to obtain the sound source pulse sequence in the conventional method of Reference 2, gk is generally l ₁ ，
It becomes a function such as l ₂ ,…, lk. Therefore, the sound source pulse sequence determined by the conventional method in Reference 2 is as follows:
There is a more suitable sound source pulse sequence that does not truly reduce J in equation (7). In a method in which the driving sound source signal sequence is represented by multiple pulses, in order to obtain even better audio quality in the region where the transmission rate is 10k bits/second or less, it is necessary to find a more suitable amplitude and position of the sound source pulse sequence. Become. The present invention relates to improvements to this sound source pulse search algorithm.

An object of the present invention is to provide a high-quality speech encoding method that can be applied to transmission rates of 10 kbit/sec or less with a relatively small amount of calculation.

According to the present invention, there is provided a means for inputting a discrete audio signal sequence, dividing the audio signal sequence into short-time intervals to obtain a short-time audio signal sequence, and extracting a parameter representing a spectral envelope from the short-term audio signal sequence. means for calculating an autocorrelation sequence of the impulse response sequence corresponding to the spectral envelope; and a cross-correlation sequence of the impulse response sequence corresponding to the spectral envelope and the short-time speech signal sequence. and means for calculating the autocorrelation sequence and the cross-correlation sequence when successively determining the position and amplitude of the sound source pulse suitable for the driving sound source signal sequence of the short-time audio signal sequence. The position of a new sound source pulse is determined based on the position and amplitude of the sound source pulse that has been found, and the position of the sound source pulse that has been found in the past is determined based on the position of the sound source pulse that was found in the past and the position of the newly determined sound source pulse. driving sound source pulse encoding means for recalculating the amplitude of the new sound source pulse and the new sound source pulse; and means for outputting a combination of the code of the parameter representing the spectral envelope and the code representing the driving sound source signal sequence. There is obtained a speech encoding device characterized by having the following.

Further, according to the present invention, there is provided a means for inputting a discrete audio signal sequence, dividing the audio signal sequence into short-time intervals to obtain a short-time audio signal sequence, and extracting a parameter representing a spectral envelope from the short-term audio signal sequence. means for calculating an autocorrelation sequence of an impulse response sequence having a spectral envelope obtained by adding a predetermined correction to the spectral envelope based on the short-term speech signal sequence; means for calculating a cross-correlation sequence using a target signal sequence obtained by adding a predetermined correction based on the temporal audio signal sequence; When determining the position and amplitude of a pulse sequentially, a new sound source pulse position is determined based on the previously determined sound source pulse position and amplitude, and the previously determined sound source pulse position and the newly determined sound source pulse position are determined. drive sound source pulse encoding means configured to recalculate the amplitude of the sound source pulse obtained in the past and the new sound source pulse based on the position of the sound source pulse obtained in the past; and the code of the parameter representing the spectral envelope. and a code representing the drive excitation signal sequence.

The speech encoding method according to the present invention is characterized by an algorithm for obtaining the above-mentioned sound source pulse sequence. Therefore, when the above equation (7) is given, the amplitude gk of the sound source pulse train that minimizes J in equation (7), k = 1, 2,
...K and the position lk, k=1, 2, . . . , the algorithm of the present invention for determining K will be explained.

First, the amplitude and position are respectively {g ₁ , g ₂ ,...,
g _K-1 }, {l ₁ , l ₂ ..., l _K-1 }, the square error when one more pulse is added to the (K-1) pulse sequence is expressed in equation (7). It is expressed as shown below.

JK _N-1 〓 ⁿ⁼¹ (xw(n)− _K 〓 ^k=1 gkhw(n−lk)) ² −(11) To see the influence of the K-th pulse, use equation (11).
If we partially differentiate with respect to gk and set it to 0, we get the following relationship.

Moreover, JK at this time can be calculated as follows using J _K-1 and gK.

JK＝J _K-1 −g ² _K/ 〓hh(lK,lK) k＞1−(13) However, J _K-1 ＝ _N-1 〓 ⁿ⁼⁰ (xw(n)− _K 〓 ^k=1 gkhw (n-lk)) ² −(14) JK is a function of lK from both equations (12) and (13),
From equation (13), lK at which g ² K in equation (12) is the largest
It can be seen that JK becomes the smallest when the pulse is set at . Next, by partially differentiating equation (11) with respect to gk and setting it to 0, the following relationship is obtained.

ψxn(lk)= _K 〓 ⁱ⁼¹ giψhh(li,lk) , k=1,...K −(15) gk, k=1,..., K that satisfies (15) is the solution of the following simultaneous linear equations: Seek.

The KxK matrix on the left side of equation (16) is a positive definite, symmetric matrix, and gk, k=1, . . . , K can be solved using a high-speed algorithm such as CHOLESKY decomposition. Also, when equation (15) holds, JK becomes minimum and can be calculated as follows.

JK= _N-1 〓 ⁿ⁼⁰ xw(n) ² − _K 〓 ^k=1 gkψxh(lk) −(17) Therefore, in equations (12) and (16), K=1 is the initial value and l ₁ , g ₁ , and then sequentially with respect to K, the position lk is determined by equation (12), and the position lk is determined by equation (16),
Calculate g ₁ , g ₂ , ..., gk until the number of pulses reaches a predetermined value or the calculated g ₁ , g ₂ , ..., gp, l ₁ , l ₂ ..., lp By substituting into equation (17) and repeating it until the value of the squared error obtained becomes smaller than the predetermined value, or the amplitude of the newly rising pulse becomes smaller than the predetermined value. ,(7)
It is possible to search for the amplitude gk and position lk of the driving sound source pulse sequence that reduce the equation. This concludes the explanation regarding the derivation of the algorithm of the present invention.

As described above, the present invention is characterized by an algorithm for determining a sound source pulse sequence, and has the same configuration as that in FIG. The present invention can be realized. Therefore, the sound source pulse sequence calculation circuit 140 according to the present invention will be explained here. FIG. 3 is a flowchart showing the processing procedure performed by the sound source pulse sequence calculation circuit according to the present invention. The position of the first pulse is determined by setting P=0 in equation (12), calculating ψxh(l ₁ )/ψhh(l ₁ , l ₁ ), and calculating (ψxh(l ₁ )/ψhh(l ₁ , l ₁ )) Maximize ² l ₁
Choose. This becomes the first pulse position. The amplitude g ₁ of the first pulse is P = 0 in equation (16)
Then, calculate by substituting l ₁ above. The position of the second pulse is determined by setting P=1 in equation (12) and substituting the above g ₁ and l ₁ , and then calculating the position by {(ψxh(l ₂ )−g ₁ ψhh(l ₁ ,
l ₂ ))/ψhh(l ₂ , l ₂ )} is l ₂ that maximizes ² . g ₁
and
The value of g ₂ is determined by substituting l ₁ and l ₂ into equation (16). The third and subsequent calculations are similar: find the P-th position lp using equation (12), substitute l ₁ ,..., lp into equation (16), and find the values of g ₁ ,..., gp. Repeat the process. In FIG. 3, 5 initializes to 1 a counting counter that counts the number of pulses. 6
is a comparison, and it is determined whether the number of pulses is larger or smaller than a predetermined number, and if it is larger than the predetermined number, the pulse sequence calculation process is finished. 7 calculates the above equation (12), and calculates the position of the pulse. 8 calculates the above equation (16), and calculates the amplitude of the pulse. 9 is an adder that increments by one a counting counter that counts the number of pulses. This completes the explanation of the sound source pulse calculation circuit according to the present invention.

According to the configuration of the present invention, in calculating the sound source pulse sequence, the optimum amplitude is determined sequentially using equation (16) and the optimum position regarding the number of pulses is determined using equation (12). Unlike the conventional method, there is no assumption that the amplitude of a pulse is a function only of the position at which the pulse stands, and a more suitable sound source pulse sequence can be obtained. Therefore, there is an effect that better reproduced sound quality can be obtained. Also, ψxh (lk) (0≦lk≦N-1) and ψhh (l _i , l _j ) (0≦
By calculating the values of l _i , l _j ≦N−1) for each frame in advance, (12)
The operations on the expression are simplified operations of multiplication and subtraction. Furthermore, since equation (16) is a positive symmetric matrix, there is an algorithm that can solve it quickly, as described in Reference 1.
This method has the effect of greatly reducing the amount of calculation compared to the conventional method.

Note that the calculation of the sound source pulse sequence of the present invention described above is as follows:
Although the calculation is performed on a frame-by-frame basis, the frame may be divided into several subframes, and the pulse sequence may be calculated for each subframe.
With this configuration, if the number of frame divisions is d, the amount of calculation can be increased approximately 1/d compared to the configuration shown in FIG. 3.

Furthermore, in the configuration example described above, the frame length is constant, but it may be variable.
The characteristics will improve if it is made variable. Further, although the K parameter is used as the parameter representing the spectral envelope of the short-time audio signal sequence, other well-known parameters (for example, LSP parameters, etc.) may also be used. Furthermore, the weighting function w(n) described above may be omitted.

In addition, the sound source pulse calculation formula (12) according to the present invention,
(16) In both equations, the covariance function for ψhh(・) was calculated according to equation (10), but this may be configured to calculate an autocorrelation sequence as shown in the equation below. .

ψhh(l _i , l _j )= _N-(li-lj)+1 〓 〓 ⁿ⁼⁰ hw(n)hw(n-｜l _i −l _j ｜), 0≦｜l _i −l _j ｜≦ N-
1-(18) By adopting this configuration, ψhh(・)
It is possible to significantly reduce the amount of calculation required for the calculation of , and there is an effect that the amount of calculation as a whole can also be reduced.

Furthermore, in the present invention, when calculating the autocorrelation sequence of the synthesis filter, the impulse response of the single-stage synthesis filter is calculated according to equation (10). It can be obtained by performing an inverse Fourier transform. Furthermore, in the present invention, the cross-correlation sequence between the impulse response of the synthesis filter and the input audio signal is calculated according to equation (9), but the product of the power spectrum of the synthesis filter and the power spectrum of the input audio signal is subjected to inverse Fourier transform. It can be found by

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of a speech encoding method with a conventional configuration, FIG. 2 is a flowchart showing a processing procedure performed by a conventional sound source pulse sequence calculation circuit, and FIG. A flowchart showing the processing procedure performed by the sound source pulse sequence calculation circuit according to the following is shown. In the figure, 110...buffer memory circuit,
120... Impulse response calculation circuit, 130...
Covariance function calculation circuit, 135... Cross-correlation number calculation circuit, 140... Sound source pulse sequence calculation circuit, 1
50... Encoding circuit, 160... Multiplexer, 180... K parameter calculation circuit, 190...
...K parameter encoding circuit, 200...Weighting circuit, 1...Pulse counting counter, 2...Comparator, 3...Pulse calculation circuit, 4...Adder, 5...
Pulse count counter, 6...Comparator, 7...Pulse position calculation circuit, 8...Pulse amplitude calculation circuit,
9 indicates an adder.

Claims (1)

[Scope of Claims] 1. Means for inputting a discrete audio signal sequence, dividing the audio signal sequence into short-time intervals to obtain a short-time audio signal sequence, and determining a parameter representing a spectral envelope from the short-term audio signal sequence. means for extracting and encoding; means for calculating an autocorrelation sequence of the impulse response sequence corresponding to the spectral envelope; and a cross-correlation sequence of the impulse response sequence corresponding to the spectral envelope and the short-time signal sequence. and means for calculating the autocorrelation sequence and the cross-correlation sequence when successively determining the position and amplitude of the sound source pulse suitable for the driving sound source signal sequence of the short-time audio signal sequence. The position of a new sound source pulse is determined based on the position and amplitude of the sound source pulse that has been found, and the position of the sound source pulse that has been found in the past is determined based on the position of the sound source pulse that was found in the past and the position of the newly determined sound source pulse. driving excitation pulse encoding means configured to recalculate the amplitude of the excitation pulse and the new excitation pulse, and outputting a combination of the code of the parameter representing the spectral envelope and the code representing the driving excitation signal sequence. 1. A speech encoding device comprising: means. 2. Means for inputting a discrete audio signal sequence, dividing the audio signal sequence into short-time intervals to obtain a short-time audio signal sequence, and extracting and encoding a parameter representing a spectral envelope from the short-time audio signal sequence. means for calculating an autocorrelation sequence of an impulse response sequence having a spectral envelope obtained by adding a predetermined correction to the spectral envelope based on the short-time audio signal sequence; The position and amplitude of the sound source pulse suitable for the driving sound source signal sequence of the short-time audio signal sequence are calculated using means for calculating a cross-correlation sequence using the target signal sequence to which a predetermined correction has been applied to the original signal sequence. When successively determining drive sound source pulse encoding means configured to recalculate the amplitude of the sound source pulse obtained in the past and the new sound source pulse based on the position; a code of a parameter representing the spectral envelope; and the drive sound source signal. 1. A speech encoding device comprising means for outputting a combination of a code representing a sequence and a code representing a sequence.