JPH077277B2 - Speech coding method and apparatus thereof - Google Patents

Description

The present invention relates to a low bit rate waveform coding system for speech signals and an apparatus therefor.

(Prior art and its problems) As an effective method for encoding a voice signal with a transmission information amount of, for example, about 16 kbit / sec or less, a drive source signal sequence of a voice signal is reproduced using it. There is known a method of searching every short time on condition that the error power between the signal and the input signal is minimum. These methods are called tree coding (TREE CODING) and vector quantization (VEOTOR QUANTIZATION) depending on the search method. In addition to these methods, a plurality of pulse sequences representing a drive excitation signal sequence are transmitted by the encoder side at a short time every A-b-S ( A
NALYSIS- B Y- S YNTHESIS) technique recently scheme to be obtained a sequential manner with the, has been proposed. The present invention relates to this system. For more information on this method, please see ICS AS (ICAS) by BS ATAL.
SP), 1982 pp. 614-617, "A New Model of LPC Excitement for Producing Natural Sounding Speech at Low Bit"・ Rates "
(“A NEW MODEL OF LPC EXCITATION FOR PRODUCING NA
TURAL-SOUNDING SPEECH AT LOW BIT RATES "), the explanation is given here, so only a brief explanation is given here.

FIG. 1 is a diagram showing the concept of processing on the encoder side in the conventional method described in Document 1 above. In the figure, 10
Reference numeral 0 denotes an encoder input terminal to which the A / D converted audio signal sequence x (n) is input. 110 is a buffer memory circuit, and one frame of the audio signal sequence (for example, in the case of 8KHZ sampling, if the frame length is 10 msec, 80 samples)
Accumulate minutes. The output value of 110 is output to the subtractor 120 and the K parameter calculation circuit 180. However, according to Document 1, it is described as REFLECTION COEFFICIENTS instead of the K parameter, but this is the same parameter as the K parameter. K
The parameter calculation circuit 180 uses the output value of 110, calculates the K parameter Ki representing the speech signal spectrum for each frame for the 16th order (1 ≦ i ≦ 16) according to the covariance method, and outputs these to the synthesis filter 130. . Reference numeral 140 denotes a sound source pulse generating circuit, which generates a predetermined number of pulse sequences in one frame. Here, this pulse sequence is referred to as d (n). FIG. 2 shows an example of a sound source pulse sequence generated by the sound source pulse generation circuit 140. In FIG. 2, the horizontal axis represents discrete time and the vertical axis represents amplitude. Here, the case where eight pulses are generated in one frame is shown. The pulse sequence d (n) generated by the sound source pulse generation circuit 140 is a synthesis filter.
Drive 130. The synthesis filter 130 inputs d (n), obtains a reproduction signal (n) corresponding to the audio signal x (n), and outputs this to the subtractor 120. Here, the synthesis filter 130 inputs the K parameter Ki, converts these into the prediction parameter ai (1 ≦ i ≦ 16), and calculates the reproduction signal (n) using ai. (N) can be expressed by the following equation using d (n) and ai.

In the above equation, P indicates the order of the synthesis filter, and P = 16 in Document 1 above. The subtractor 120 calculates the difference e (n) between the original signal x (n) and the reproduced signal (n), and the weighting circuit
Output to 190. 190 inputs e (n), uses weighting function W (n), and weights error ew (n) according to the following equation.
To calculate.

ew (n) = w (n) * e (n) (2) In the above formula, the symbol “*” represents convolution integral. Also,
The weighting function w (n) is used to perform weighting on the frequency axis. When the Z-transformed value is W (Z), it is represented by the following equation using the prediction parameter ai of the synthesis filter.

In the above equation, r is a constant of 0 ≦ r ≦ 1 and determines the frequency characteristic of W (Z). That is, if r = 1, W (Z) =
The frequency characteristic becomes 1 and the frequency characteristic becomes flat. On the other hand, r = 0
Then, W (Z) has the inverse characteristic of the frequency characteristic of the synthesis filter. Therefore, the characteristic of W (Z) can be changed by the value of r. Also, as shown in the equation (3), W
The reason why (Z) is determined depending on the frequency characteristic of the synthesis filter is that the audible masking effect is used. That is, in a place where the spectrum of the input audio signal is large in power (for example, in the vicinity of the formant), even if the error with the spectrum of the reproduced signal is a little large, the error is hard to hear, which is due to the auditory property. FIG. 3 shows the spectrum of the input audio signal in a certain frame and an example of the frequency characteristic of W (Z). Here, r = 0.8. In the figure, the horizontal axis represents frequency (up to 4 KHz) and the vertical axis represents logarithmic amplitude (up to 60 dB). The upper curve represents the spectrum of the audio signal, and the lower curve represents the frequency characteristic of the weighting function.

Returning to FIG. 1, the weighting error ew (n) is fed back to the error minimization circuit 150. Error minimization circuit 150
Stores the value of ew (n) for one frame and calculates the weighted squared error ε according to the following equation using these values.

Here, N represents the number of samples for calculating the squared error. In the method of Reference 1, this time length is set to 5 msec, which corresponds to N = 40 in the case of 8 KHz sampling. Next, the error minimization circuit 150 calculates 2 using the equation (4).
Pulse position and amplitude information is obtained so as to reduce the multiplication error ε, and these are given to the sound source pulse generation circuit 140. 140
Generates a source pulse sequence based on this information.
The synthesizing filter 130 calculates the reproduction signal (n) using this sound source pulse sequence as a driving source. Next, in the subtractor 120,
A new error e (n) is obtained by subtracting the currently obtained reproduction signal (n) from the previously calculated error e (n) between the original signal and the reproduction signal. The weighting circuit 190 inputs e (n), calculates a weighting error ew (n), and feeds it back to the error minimizing circuit 150. The error minimization circuit 150 calculates the squared error again and adjusts the amplitude and position of the sound source pulse sequence so as to reduce the squared error. In this way, a series of processes from generation of the sound source pulse sequence to adjustment of the sound source pulse sequence by error minimization is repeated until the number of pulses of the sound source pulse sequence reaches a predetermined number, and the sound source pulse sequence is determined.

This is the end of the description of the conventional method.

In the conventional method, the information to be transmitted is the K parameter Ki (1 ≦ i ≦ 16) of the synthesizing filter, the pulse position and the amplitude of the sound source pulse sequence, and an arbitrary transmission rate depending on the number of pulses generated in one frame. Can be realized. Furthermore, in the area where the transmission rate is 16 Kbps or less, good reproduction sound quality can be obtained, which is considered to be one of the effective methods.

However, this conventional method has a drawback that the amount of calculation is very large. This is because the A-B-S process is included in the sound source pulse sequence calculation loop. That is, when calculating the position and amplitude of the sound source pulse sequence, the signal is once reproduced based on the pulse, the error between this reproduced signal and the original signal and the square error are calculated, and these are fed back. This is because the pulse position and amplitude are adjusted so as to reduce the square error. Further, according to this conventional method, at a bit rate of about 16 Kbps or less, in the case of an input signal with a high pitch frequency, for example, when a female voice is input, there is a drawback that the reproduction quality deteriorates. This means that when the pitch frequency is high, many pitch waveforms are included in the frame of the pulse calculation, and in order to reproduce this pitch waveform satisfactorily, it is necessary to compare it with a speaker with a low pitch frequency. , Because of the need for a larger number of source pulses. Therefore, for this reason, it is difficult to significantly reduce the transmission bit rate, that is, the number of pulses in one frame.

(Object of the Invention) It is an object of the present invention to provide a voice encoding system and a device thereof capable of reproducing high quality voice even at a low transmission bit rate.

(Structure of the Invention) According to the present invention, on the transmitting side, a discrete audio signal sequence is input to obtain a plurality of parameter sequences representing short-time spectrum characteristics of the audio signal sequence, and to represent the audio signal sequence. When obtaining the pulse sequence for each frame, the plurality of parameter sequences are switched within the frame to obtain the pulse sequence and the pulse sequence is encoded, or the best parameter is obtained after obtaining the pulse sequence for each parameter. To encode the pulse sequence, and output by combining the code representing the pulse sequence and the code representing the plurality of parameter sequences, and at the receiving side input the combined code to input the pulse sequence. The code that represents the code and the code that represents the plurality of parameter sequences are separated and decoded, and the sound based on the decoded pulse sequence. A voice encoding method is characterized in that, when reproducing a voice signal sequence, the plurality of decoded parameter sequences are switched or the voice signal sequence is reproduced using a selected parameter. .

Further, according to the present invention, a parameter calculation circuit for inputting a discrete audio signal sequence to obtain a plurality of parameter sequences representing short-time spectral characteristics of the audio signal sequence, and a pulse sequence for representing the audio signal sequence are framed. The plurality of parameter sequences are switched within the frame to obtain the pulse sequence when each is obtained, and the pulse sequence is encoded, or the best parameter is selected after obtaining the pulse sequence for each parameter, and A speech coder comprising a pulse calculation circuit for coding a pulse sequence, and a multiplexer circuit for combining and outputting the code sequences of the plurality of parameter sequences and the pulse sequence.

Further, according to the present invention, a code sequence in which a code representing a plurality of parameter sequences representing a short-time spectrum characteristic of a voice signal sequence and a code representing a pulse sequence for representing a voice signal are combined is inputted and each code is inputted. A demultiplexer circuit for separating the input signal, a pulse decoding circuit for inputting and decoding the code representing the pulse sequence obtained by separation, and a code sequence representing the plurality of parameter sequences obtained for separation. A parameter decoding circuit for decoding by using the decoded pulse sequence and the plurality of decoded parameter sequences, or when switching the plurality of decoded parameter sequences when reproducing the audio signal sequence, or And a synthesis filter circuit for reproducing and outputting the audio signal sequence by using a selected parameter. .

(Principle of the Present Invention) The present invention is characterized by a sound source pulse calculation method. The feature is that the parameter representing the short time spectrum characteristic of the original signal is switched in the process of calculating the original sound pulse. The impulse response will be described below as an example of the above parameters, and the excitation pulse calculation method will be described in the case of switching between two types of impulse responses in the excitation pulse calculation process.

Now, let the two types of impulse responses be h ₁ (n) and h ₂ (n). There are several possible ways to select h ₁ (n) and h ₂ (n). Here, as h ₁ (n), consider an impulse response that represents a short-time spectrum of a speech signal including a fine pitch structure, and h ₂ As (n), consider the impulse response that represents the short-time spectral envelope of the audio signal. In this case, h ₁ (n) is obtained from the impulse response of the synthesis filter that is a cascade connection of the pitch prediction filter and the spectrum envelope prediction filter, and h ₂ (n) is obtained from the impulse response of the spectrum envelope prediction filter. Here, the pitch prediction filter may be of a first-order case or a high-order case, but here, for simplification of description, a case of using a first-order pitch prediction filter will be considered. Various methods of calculating the tap coefficient β and the pitch period M of the pitch prediction filter are known, but as a simple method, for example, a method of extracting the peak amplitude of the autocorrelation sequence of the input speech signal and its position is well known. Has been. For more information on this method, see BS ATAL, M.R.
BELL SYSTEM TECHNICAL JOURNAL by Mr. Schroeder (MRSCHROEDER)
JOURNAL magazine, October 1970, 1973-1986, "Adaptive Predictive Coding of Speech Signals"("ADAPTIVE PREDITIVE C
Since it is explained in detail in the paper (Reference 2) entitled "ODING OF SPEECH SIGNALS"), the explanation is omitted here.

If the source pulse sequence in one frame is d (n), the reproduced signal (n) reproduced using the impulse response h ₁ (n) can be written as follows.

(N) = d (n) * h ₁ (n) (5) Here, the symbol “*” represents convolution. Also d (n)
Can be expressed as

Where δ (n, mi) represents the Kronecker delta,
It is 1 when n = mi and 0 when n ≠ mi. gi, mi
Each indicate the amplitude position of the i-th pulse. K ₁ represents the number of pulses obtained using the impulse response h ₁ (n) within one frame.

Next, the weight value of the error power between the input audio signal x (n) and the reproduced signal (n) is multiplied by the following equation.

Where xw (n) = x (n) * w (n) (8) h ₁ w (n) = h ₁ (n) * w (n) and w (n) is the impulse response of the weighting circuit. . For example, w (n) has the same characteristic as that of the conventional weighting circuit shown in FIG.

The sound source pulse sequence that minimizes equation (7) is calculated from the following equation obtained by partially differentiating equation (7) with the amplitude gi of the sound source pulse sequence and setting it to zero.

Here, xh (•) represents the cross-correlation sequence calculated from xw (n) and h ₁ w (n), and hh (•) represents the autocorrelation sequence calculated from h ₁ w (n). In addition, hh (・)
Is often called a covariance function sequence in the field of speech signal processing.

According to the equation (9), the position of the pulse is obtained as a position where the absolute value of the right side of the equation (9) is maximized. The amplitude of the pulse is the value of gi for the obtained position.

Now, ignoring the effects of frame edges, the covariance function hh
It can be said that (ml, mi) is equal to the autocorrelation sequence Rhh (ml-mi |) depending on the time difference (| ml-mi |).

Using this relationship, equation (9) is modified as

Here, Rhh (•) can be expressed as the following equation.

Therefore, when h ₁ (n) is used as the impulse response, K ₁ sound source pulses can be calculated from the equation (10).

Next, after calculating K ₁ pulses, the impulse response is h ₂
A method of calculating K ₂ source pulses by switching to (n) will be shown.

If the amplitude and position of the sound source pulse are gsi and msi respectively,
The weighted error power E'when K ₂ source pulses are applied can be expressed by the following equation.

Here, (n) is obtained from the equation (5) using the sound source pulse train obtained using the equation (10).

The gsi that minimizes the equation (12) is obtained from the following equation obtained by substituting the equation (5) into the equation (12) and partially differentiating the equation (12) with respect to gsi.

Here, sxh represents the number of cross-correlations between the input signal x (n) and the impulse response h ₂ w. Rshh is the impulse response h ₂ w
The autocorrelation number of is shown. Equation (13) is the impulse response h ₁
By subtracting the effect of the pulse train d (n) calculated using (n) from the cross-correlation, the pulse train gsi when the impulse response is switched to h ₂ (n) is the same as the above equation (10). It is shown that it is calculated by the procedure of.

As a method other than the equation (13), the pulse train gsi can be calculated from the equation obtained by partially differentiating the equation (12) with respect to gsi and setting it to 0. However, according to this method, since the signal (n) of the equation (12) must be reproduced at once, the amount of calculation increases as compared with the method of the equation (13).

This completes the description of the sound source pulse calculation method.

(Example) Hereinafter, an example of the present invention will be described in detail with reference to the drawings. FIG. 4 (a) is a block diagram showing an example of the transmitting side of the audio encoding system according to the present invention, and FIG. 4 (b) is a block diagram showing an example of the receiving side. In FIG. 4A, the discrete audio signal sequence x (n) is input from the input terminal 195, divided into a predetermined number of samples, and stored in the buffer memory circuit 340.

Next, the K parameter calculation circuit 280 inputs a sequence of a predetermined length among the audio signal sequences accumulated in the buffer memory circuit 340, and using this, the LPC parameter of the predetermined order P is used. Is calculated according to a known method (for example, a linear predictive analysis method). There are various possible LPC parameters, but in the following, K
The description will proceed assuming that the parameter Ki (1 ≦ i ≦ P) is used. The K parameter is the same parameter as the Percoll coefficient. The K parameter Ki is output to the K parameter encoding circuit 200. The K parameter encoding circuit 200, for example, based on a predetermined number of quantization bits,
Ki is encoded and the code lki is output to the multiplexer 450. Further, the K parameter coding circuit 200 outputs K _i ′ obtained by decoding lki to the impulse response calculation circuit 210, the weighting circuit 410, and the synthesis filter circuit 400.

Next, the pitch analysis circuit 370 inputs the output sequence of the buffer memory circuit 340 and calculates the coefficient of the pitch reproduction filter by a predetermined order Q. Here, the transfer function H _P (Z) of the pitch reproduction filter can be expressed by the following equation using the Z conversion expression.

In the present embodiment, for simplification of the description, the description will proceed assuming that the order of the pitch reproduction filter is the first order. In this case, the calculation of the pitch gain β and the pitch period M is described in detail in, for example, the above-mentioned Document 2. Other publicly known methods can also be used. The pitch gain β and the pitch period M are output to the pitch encoding circuit 380. The pitch encoding circuit 380 encodes the pitch period M and the pitch gain β with a predetermined number of bits, and obtains l _M and l
Output _β to the multiplexer 450. The pitch coding circuit 380, l _M and l M obtained by decrypting the _beta 'and beta' impulse response calculation circuit 210 and the pulse calculating circuit 390
And pulse generator circuit 420.

Next, the impulse response calculation circuit 210 inputs the K parameter decoded value K _i ′ from the K parameter encoding circuit 200,
Further, the decoded values M ′ and β ′ of the pitch period and the pitch gain are input from the pitch encoding circuit 380. The impulse response calculation circuit 210 calculates two types of impulse responses. As one of them, using K _i ′, pitch gain β ′ and pitch period M ′, the weighted impulse response h ₁ w (n) of the synthesis filter composed of the pitch reproduction filter and the spectral envelope synthesis filter is obtained. Calculate only the specified number of samples. Here, the transfer function of this synthesis filter is multiplied by the following equation using the Z-transform expression.

Here, ai represents a prediction coefficient obtained by converting from K _i ′ by a known method. Further, W (z) represents a z-transform expression of the weighting function, and the above equation (3) can be used. As another impulse response, the K parameter decoded value K
Calculate the weighted impulse response sequence h ₂ w (n) of the spectral envelope synthesis filter using _i ′. Here, the transfer function of the weighted spectral envelope synthesis filter can be written as the following equation using the z-transform expression.

Impulse response h ₁ (n), h obtained as described above
₂ (n) is output to the switch circuit 365.

The switch circuit 365 has two types of impulse responses h ₁ (n), h ₂
(N) is input, and these are switched to switch the pulse determination circuit 34.
Output to 0.

Next, the subtractor 285 inputs the audio signal sequence x (n) accumulated in the buffer memory circuit 340, subtracts the output sequence (n) of the synthesis filter circuit 400 from x (n) by one frame sample, The subtraction result is output to weighting circuit 410.
The weighting circuit 410 uses the K parameter encoding circuit 200
The K parameter decoded value K _i ′ is input and the weighting function w
(N) is calculated so that the z-transformed value is, for example, the expression (3). This may be calculated using other frequency weighting methods. Further, the weighting circuit 410 inputs the subtraction result of the subtractor 285, performs a convolution operation with this and the weighting function w (n), and outputs the obtained xw (n) to the pulse determination circuit 340.

The pulse determination circuit 340 inputs xw (n) and two types of impulse responses h ₁ (n) and h ₂ (n), and inputs h ₁ (n) and h ₂ (n).
And the sound source pulse is calculated according to the above equations (10) and (13). The pulse determination circuit 340 includes an autocorrelation coefficient calculation circuit 360, a cross correlation coefficient calculation circuit 350, and a pulse calculation circuit.
Composed of 390. The operation of these circuits will be described below. The autocorrelation coefficient calculation circuit 360 inputs two kinds of impulse responses h ₁ w (n) and h ₂ w (n) in order, and the autocorrelation coefficients R ₁ hh (τ) and R ₂ for each impulse response are input. hh (τ)
Is calculated according to the following formula.

In the above equation, τ is the number of delayed samples, and M is the number of predetermined samples. R ₁ hh (τ) and R ₂ hh (τ) are output to the pulse calculation circuit 390.

The cross-correlation coefficient calculation circuit 350 inputs xw (n) and impulse responses h ₁ (n) and h ₂ (n), and cross-correlation coefficients xh and xw between xw (n) and h ₁ (n). The cross-correlation number sxh between (n) and h ₂ (n) is calculated, and these are output to the pulse calculation circuit 390.

The pulse calculation circuit 390 has two types of cross-correlation xh, sxh and 2
Type the autocorrelation R ₁ hh, R ₂ hh, and enter the impulse response h
₁ Find the sound source pulse by (n) K ₁ according to equation (10). Next, K ₂ pieces of sound source pulses based on the impulse response h ₂ (n) are obtained according to the equation (13). The obtained excitation pulse is output to the encoding circuit 470. This is the end of the description of the pulse determination circuit 340.

Next, the encoding circuit 470 inputs the amplitude and position of the excitation pulse train from the pulse calculation circuit 390, and encodes these using a normalization coefficient described later. The normalization coefficient is also encoded, and the normalization coefficient, the amplitude of the excitation pulse train, and the code indicating the position are output to the multiplexer 450. Also, the decoded values g _i ′, m _i ′ of the amplitude and position of the excitation pulse train are output to the excitation pulse generation circuit 420. The operation of the encoding circuit 470 is described in Japanese Patent Application No. 57-231605.
The detailed description is omitted here.

Next, various methods can be considered for encoding the pulse position. For example, a run length code or the like well known in the field of facsimile signal encoding may be used. This represents the length following the code "0" or "1" using a predetermined code sequence. When the position between pulses is encoded using this code sequence, the following method may be used. Now, let K be the total number of pulses in one frame, K _{1 be} the number of pulses calculated using impulse response h ₁ (n), and K ₂ be the number of pulses calculated using impulse response h ₂ (n). . It suffices to give a code of a predetermined length to the positions of K ₁ pulses and give a predetermined code to the positions of K ₂ pulses. As another method, a discriminant code for one bit is added to the code indicating the position of each pulse, and
It may be possible to determine which _{one of} h ₁ (n) and h ₂ (n) has been used. The former method requires less total position information than the latter method.

Regarding the encoding of the pulse sequence, it is needless to say that the best known method can be used without being limited to the encoding method described here.

Returning to FIG. 4 (a), _{g i} pulse sequence generating circuit 420 which inputs ', _{m i'} with 'amplitude _g i to the position of' _{m i}
Is calculated over one frame length N and is output to the synthesis filter circuit 400 as a drive signal. The method of creating the drive signal will be described below. For the K ₁ pulses calculated using the impulse response h ₁ (n), the pulse train d ₁ (using the amplitude g _i ′, the position m _i ′, and the pitch information β ′, M ′ according to the following equation n) is generated.

d ₁ (n) = d _I (n) + β ′ · d ₁ (n−M ′) (18) On the other hand, for K ₂ pulses calculated using the impulse response h ₂ (n), a pulse train is generated according to the following equation.

The addition sequence of d ₁ (n) and d ₂ (n) is output to the synthesis filter circuit 400 as the drive signal d (n).

The synthesis filter circuit 400 inputs the drive signal d (n) from the pulse generation circuit 420, the K parameter decoded value K _i ′ from the K parameter encoding circuit 200, and inputs two frames (1
A response signal sequence (n) of ≦ n ≦ 2N) is obtained, and the value of (n) (N + 1 ≦ n ≦ 2N) of the second frame is output to the subtractor 285. The operation of the synthesizing filter 400 is the same as the above-mentioned Japanese Patent Application No.
Since the synthesis filter 320 is described in detail in the specification of 7-231605, its description is omitted here.

Next, the multiplexer 450 inputs the output code of the encoding circuit 470, the output code of the K parameter encoding circuit 200, and the output code of the pitch encoding circuit 380, combines them, and outputs them from the transmission side output terminal 480. Output to the communication path. This is the end of the description of the encoder side of the speech encoding system according to the present invention.

Next, the receiving side of the voice encoding system according to the present invention will be described with reference to FIG.

The demultiplexer 500 inputs a code from the reception side input terminal 490. Of the input codes, the demultiplexer 500 separates the code sequence representing the K parameter, the code sequence representing the pitch information, and the code sequence representing the excitation pulse train, and the code sequence lki representing the K parameter to the K parameter decoding circuit 520. A code sequence l _M ,
l _β is output to pitch decoding circuit 510, and a code sequence representing an excitation pulse train is output to excitation pulse decoding circuit 530.
The K parameter decoding circuit 520 decodes the input code sequence and outputs it to the synthesis filter circuit 550. The pitch decoding circuit decodes the input code sequence and outputs it to the pulse generation circuit 540.

Excitation pulse decoding circuit 530 inputs a code sequence representing an excitation pulse train, decodes it, and outputs it to pulse generation circuit 540 as amplitude and position information of the excitation pulse train.

The pulse generation circuit 540 inputs the amplitude, position information and pitch information of the sound source pulse train, and the pulse generation circuit 420 on the transmission side.
The same operation is performed to generate the drive signal d (n). d (n) is output to the synthesis filter circuit 550.

The synthesis filter circuit 550 inputs the drive signal d (n) from the pulse generation circuit 540 and outputs the K parameter decoding circuit 520 to K.
Input the parameter decoded value K _i ′. K _i ′ is converted into a prediction parameter a _i ′ (1 ≦ i ≦ N _P ) by a known method. The synthesis filter circuit 550 uses the a _i ′ and the drive signal d.
The reproduced signal (n) for one frame is calculated according to the following equation using (n), (N) is output through the reception side output terminal 560.

This is the end of the description on the decoder side of the audio encoding system according to the present invention.

In the present embodiment, in the pulse calculation in the pulse calculation circuit 390, the first K ₁ pulses are obtained using the impulse response h ₁ (n) using the pitch information, and the remaining K ₂ pulses are Impulse response without pitch information h ₂ (n)
Was sought by. This may be done as follows. That is, the pitch gain β'may be used to determine whether or not the pitch information is used for pulse calculation. That is, if the pitch gain β'has a predetermined threshold value, the impulse response h ₁ (n) using the pitch information is used for the first K ₁ pulses, and β'is below the threshold value. If so, K pulses may be calculated by the impulse response h ₂ (n) that does not use pitch information. Alternatively, the following may be performed. That is, the pulses calculated using the pitch information (in this case, the pitch information is used for K ₁ pieces, but the pitch information is not used for the remaining K ₂ pieces) and K and all the pulses. The error power for each pulse may be obtained using the pulse calculated without using the pitch information, and the pulse having the smaller error power may be used. The calculation of the error power E using the pulse may be performed according to the following equation, for example.

In the above equation, R (0) is the output sequence xw of the weighting circuit 410.
The power of (n) is shown. Without pitch (21)
The error power may be calculated using the equation as it is. on the other hand,
When the pitch is used, the error power may be calculated according to the following equation that is a modification of equation (21).

In the above equation, gi and mi are the amplitude of the pulse calculated using the pitch,
The position is shown, and gsi and msi show the amplitude and position of the pulse calculated without using the pitch. Further, xh is the same as the cross correlation in the above equation (9), and sxh is the same as the cross correlation in the above equation (13). By using such a method, more accurate determination can be performed as compared with the determination based on the pitch gain β '.

Further, in the present embodiment, when the pulse calculation circuit 390 calculates K ₁ pulses using the pitch and then calculates K ₂ pulses without using the pitch, the equation (13) is used. This means that after removing the effect of K ₁ pulses calculated using pitch in the correlation region, without using pitch
K ₂ pulses were calculated, but as another method, the K ₁ pulses calculated using the pitch are used to regenerate a signal, and the original signal x is calculated as shown in the above equation (12). (N)
It is also possible to calculate K ₂ pulses without using the pitch for the signal after subtracting this signal from.

In the present embodiment, an example was described in which the impulse response of the synthesis filter when the pitch is used in the process of calculating the sound source pulse and the impulse response of the synthesis filter when the pitch is not used are described. In some cases, those parameters may be switched. Also, the parameters representing the short-time spectrum characteristics whose frequencies are selected may be switched.

Further, in the present embodiment, an example in which two types of impulse responses are switched has been described, but this may be configured to switch three or more types of impulse responses or other parameters representing short-time spectrum characteristics of a voice signal. . Further, when switching the impulse response, instead of switching the impulse response calculated and calculated for each frame, the following configuration may be adopted. That is, a sufficient number of impulse responses representing various short-time spectral characteristics of a voice signal are calculated in advance, and a code book in which codes representing all types are predetermined is created. This codebook must have the same code on the sending side and the receiving side. In actual encoding, when using the impulse response for each frame, the impulse response that can best approximate the short-time spectral characteristics of the original signal is selected from this codebook and used. The code corresponding to the impulse response is transmitted to the receiving side. The receiving side uses the impulse response corresponding to the received code.

As a parameter that represents the short-time spectrum characteristics,
Parameters other than the impulse response, such as the above-described prediction coefficient and PARCOR coefficient, may be used as a codebook.

In addition, in the sound source pulse calculation method shown in Eqs. (10) and (13), quasi-optimal pulses were calculated one by one. As a pulse calculation method, a method of re-adjusting the amplitudes of a plurality of pulses obtained in the past when calculating the next pulse can also be used. This method is effective when the independence of each pulse is not established, for example, when each pulse is found very close to each other. In addition to this method, various sound source pulse calculation methods are possible. For example, it is possible to use a method in which the amplitudes of all the pulses are readjusted after all the pulses in one frame are obtained.

Further, in the present embodiment, when calculating the autocorrelation sequence of the impulse response sequence representing the short-time spectral structure, the impulse response calculation circuit 210 uses the K parameter decoded value and the pitch information to calculate the impulse response sequence. After the calculation, the autocorrelation sequence was calculated using this impulse response sequence. As is well known in the field of digital signal processing, the autocorrelation series of impulse response sequences has a correspondence with the power spectrum of short-time spectrum. Therefore, the configuration may be such that the power spectrum of the short-time spectrum is obtained using the K parameter decoded value and the pitch information, and then the autocorrelation sequence is calculated. On the other hand, when calculating the cross-correlation sequence of the voice signal sequence and the impulse response sequence representing the short-time spectrum envelope, in the configuration of the present embodiment, the signal sequence xw (n), which is the output value of the weighting circuit 410, The cross-correlation function calculation circuit 350 calculates the cross-correlation number using the impulse response series obtained by the impulse response calculation circuit 210. As is well known, the cross correlation number corresponds to the cross power spectrum. The cross-correlation sequence may be calculated after obtaining the cross power spectrum using the voice signal sequence, the K parameter decoded value, and the pitch information using this relationship. Regarding the correspondence between the power spectrum and the autocorrelation sequence, and the correspondence between the cross power spectrum and the cross-correlation sequence, "Digital Signal Processing" by AVOPPENHEIM et al. DIGITAL SIGNAL P
ROCESSING ") is described in detail in Chapter 8 of the book (Reference 4), so the description is omitted here.

Further, in the above-mentioned embodiment, the encoding of the excitation pulse sequence in one frame is performed after all the pulse sequences are obtained.
Although the encoding is performed by the encoding circuit 470 of FIG. 4A, the encoding is included in the calculation of the pulse sequence, the encoding is performed every time one pulse is calculated, and the next pulse is calculated. You may make it the structure.

Further, in the embodiment described above, the K parameter is used as the parameter representing the spectrum envelope of the short-time speech signal sequence, but other well-known parameters (such as LSP parameter) may be used. Good. Furthermore, the weighting function w (n) may not be included in the above equation (8).

Further, in the present embodiment, in order to prevent quality deterioration due to discontinuity of the reproduced waveform at the frame boundary, a response signal sequence derived from a sound source pulse one frame before the current frame is calculated, and the input speech of the current frame is calculated. The drive sound source pulse was calculated after this response signal was subtracted from, but as shown in FIG. 5, the data used in the sound source pulse calculation should include the data of the frame that transmits the pulse and the past data. You may make it a different structure. In FIG. 5, N _T indicates a frame for transmitting a pulse, and N indicates a frame for calculating a sound source pulse. With such a configuration, it is not necessary to calculate a response signal sequence derived from a sound source pulse of one frame past, and the circuit configuration is simplified.

(Effect of the Invention) As described in detail above, according to the present invention, when calculating a sound source pulse train of an input audio signal, a parameter representing a short-time spectrum characteristic of the input audio signal is switched and used in a pulse calculation process. Therefore, better parameters can be used for pulse calculation, and even if the transmission bit rate is low, it is possible to obtain a reproduced voice of higher quality than the conventional method.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a conventional system, FIG. 2 is a diagram showing an example of a sound source pulse sequence, and FIG. 3 is a frequency characteristic of an input audio signal sequence and a frequency characteristic of a weighting circuit shown in FIG. FIG. 4 shows an example, FIG. 4 (a) and FIG. 4 (b) are block diagrams showing an embodiment of a voice coding system according to the present invention, and FIG.
The figure is a diagram for explaining the positional relationship between the pulse transmission frame and the sound source pulse calculation frame. In the figure, 110,340 ... buffer memory circuit, 120,285
...... Subtraction circuit, 130,400,550 …… Synthesis filter circuit, 14
0,420,540 ... Pulse generation circuit, 150 ... Error minimization circuit, 180,200 ... K parameter calculation circuit, 190,410 ... Weighting circuit, 200 ... K parameter coding circuit, 210 ...
Impulse response calculation circuit, 350 ... Cross-correlation function calculation circuit, 360 ... Autocorrelation function calculation circuit, 390 ... Pulse calculation circuit, 470 ... Encoding circuit, 450 ... Multiplexer, 50
0 ... Demultiplexer, 520 ... K parameter decoding circuit, 530 ... Pulse decoding circuit, respectively.

Claims (3)

[Claims] 1. A transmitter side receives a discrete audio signal sequence, obtains a plurality of parameter sequences representing short-time spectral characteristics of the audio signal sequence, and obtains a pulse sequence for representing the audio signal sequence for each frame. When obtaining the pulse sequence by switching the plurality of parameter sequences in the frame to obtain a pulse sequence or encoding the pulse sequence, or obtaining the pulse sequence for each parameter, select the best parameter and select the pulse The sequence is encoded, and the code representing the pulse sequence and the code representing the plurality of parameter sequences are combined and output. At the receiving side, the combined code is input and the code representing the pulse sequence and the plurality of And a code representing a parameter sequence of the above are separated and decoded, and the audio signal sequence is reproduced based on the decoded pulse sequence. Speech coding method is characterized in that so as to reproduce the voice signal sequence with or selected parameters to switch the decoded plurality of parameter sequence when. 2. A parameter calculation circuit for inputting a discrete audio signal sequence to obtain a plurality of parameter sequences representing short-time spectral characteristics of the audio signal sequence, and a pulse sequence for representing the audio signal sequence for each frame. When obtaining, the pulse sequence is switched by switching the plurality of parameter sequences within the frame to encode the pulse sequence, or after obtaining the pulse sequence for each parameter, the best parameter is selected to select the pulse sequence. A speech coding apparatus, comprising: a pulse calculation circuit for coding a plurality of parameter sequences; and a multiplexer circuit for combining and outputting the code sequences of the plurality of parameter sequences and the code sequences of the pulse sequences. 3. A code sequence in which a code representing a plurality of parameter sequences representing a short-time spectrum characteristic of a voice signal sequence and a code representing a pulse sequence for representing a voice signal are input, and each code is separated. Demultiplexer circuit, a pulse decoding circuit for inputting and decoding the code representing the pulse sequence obtained separately, and a pulse decoding circuit for inputting and decoding the code sequence representing the plurality of parameter sequences obtained separately A parameter decoding circuit for switching the decoded plurality of parameter sequences when reproducing the audio signal sequence using the decoded pulse sequence and the decoded plurality of parameter sequences. And a synthesis filter circuit that reproduces and outputs the audio signal sequence by using the above parameters.