JPH0441838B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a pseudo-formant vocoder, and particularly to a pseudo-formant vocoder that has two different types of synthesis means on the synthesis side to adapt to the operating environment.

CSM (Composite Sinusoidal
Modeling: Composite sine wave model) Extract pseudo-formant information from the audio signal through audio analysis, transmit this to the synthesis side, and reproduce speech from this pseudo-formant information (CSM parameters) using a unique CSM speech synthesis method. The pseudo-formant type vocoder that allows
This is clarified by Vol.J64-ANo.2.

As will be explained later, this pseudo-formant vocoder does not include a band filter whose band characteristics are specified by parameters transmitted to the synthesis side, and therefore becomes unstable even if an error occurs in the transmitted information. Because it does not cause errors, it has particularly strong resistance to deterioration in the error rate of the transmission path. However, if the operating environment is such that the error rate is not a problem, the
It has the disadvantage that the sound quality is slightly inferior to the sound reproduced by the LPC speech synthesis method.

(Objective of the Invention) The object of the present invention is to provide an LPC speech synthesis method that can obtain higher speech quality when the error rate of the transmission path is good, and a CSM speech synthesis method that has higher resistance to error rate degradation. The object of the present invention is to provide a pseudo-formant type vocoder that combines the features of both synthesis methods.

(Structure of the Invention) The vocoder of the present invention includes a transmitter that transmits CSM parameter information extracted by CSM analysis of an input audio signal, and a transmitter that transmits CSM parameter information that is transmitted via the transmission path. a plurality of variable frequency oscillators with a phase reset function that are set to different frequencies; a plurality of variable gain amplifiers that correspondingly set the outputs of the oscillators to respective strengths specified by the transmitted CSM parameter information; and the plurality of amplifiers. The oscillator is equipped with a variable length window function generator and a random number generator that operate on the synthesized output of the oscillator.When synthesizing voiced sounds, the phase of each oscillator is reset in accordance with the pitch period.When synthesizing unvoiced sounds, the phase of each oscillator is reset based on the output of the random number generator. CSM synthesis means for performing CSM speech synthesis by resetting the phase of each of the oscillators in accordance with the period of the oscillator and synchronizing the operation of the window function generator with the operation of the phase reset; LPC synthesis means converts parameter information into LPC parameter information and performs LPC speech synthesis using this; and the output of the CSM synthesis means and the output of the LPC synthesis means are varied according to the evaluation of the transmission quality of the transmission path. and a receiving side including variable weight combining means adapted to perform weight combining.

(Principle) Next, there is an important relationship in the establishment of this vocoder.
The principles of CSM speech analysis and synthesis will be explained.

CSM expresses an audio signal as the sum of a specific number of sine waves whose amplitude and frequency are freely selectable parameters. A predetermined number of 4 to 6 sine waves is used at most.

Therefore, when performing CSM speech synthesis, it is first necessary to express the speech signal as a sum of a predetermined number of sine waves by CSM speech analysis.
CSM voice analysis will be explained in detail later.
Only the main points will be explained here.

In CSM analysis, as in LPC analysis,
An autocorrelation coefficient is used as an intermediate parameter for the purpose of ignoring phase information, averaging the influence of sound sources, and avoiding instability due to noise components.

In other words, in CSM analysis, N low-order taps of sample autocorrelation coefficients directly calculated from the speech waveform to be expressed for each analysis frame are combined with N low-order taps of autocorrelation coefficients of a synthesized wave. The purpose is to determine the frequency and intensity (power amplitude) of each sine wave to be synthesized so that they match the N sine waves.

Now, if the number of sine waves to be synthesized is n, the angular frequency of each sine wave is ω _i (i=1, 2,...n), and the intensity of each sine wave is m _i , then the CSM composite wave y _t is y _t = _o 〓 ⁱ⁼¹ √ _i sin(ω _it +φ _i ), but the autocorrelation coefficient γ _l of this tap l is ω _i ,
It is easily expressed using m _i , and γ _l = _o 〓 ⁱ⁼¹ m _i cosl ω _i .

On the other hand, the sample of the audio waveform to be expressed is
If X _t , the sample autocorrelation coefficient υ _l of tap l in a certain frame is given as υ _l =1/M _M-1 〓 ^t=l X _t X _tl .

However, M is the number of samples in one analysis frame.

Now, in the CSM analysis, the value of each m _i and ω _i is determined so that the above-mentioned γ _l is equal to the given υ _l for N low-order values.

That is, γ _l =υ _l : However, the values of m _i and ω _i are determined so that l=0, 1, 2, . . . N holds true.

This specific method will be explained in detail later, but here, m _i and ω _i of the n sine waves mentioned above are shall be obtained.

FIG. 1 shows an example of a speech feature vector pattern based on the CSM parameters m _i and ω _i obtained in this way.

In addition, the 9th order (N = 9) CSM (number of sine waves n =
5) Figure 2 shows an example of the correspondence between the line spectrum and the 9th-order LPC spectrum envelope (frequency transmission characteristic of the LPC synthesis filter) obtained from the same audio sample.

Note that there is a relationship of N=2n-1 between the above-mentioned order N and the number n of sine waves, as described later.

From these figures, it can be seen that the CSM contains information that extracts the features of the original speech that should be expressed.

However, using the n values of m _{i and} ω _i obtained as a result of CSM analysis, the intensity specified by these m _i and ω _i (the actual amplitude is √ _i as mentioned above)
If you create n sine waves with angular frequencies and simply add and synthesize them, the human ear will hear:
The sound can simply be heard as a synthesized sine wave,
The purpose of reproducing the original sound is hardly achieved.

This means that even if you simply add sine waves, the spectrum of the generated signal is just a discretized n line spectrum, whereas the spectrum of the audio signal has a continuous spectral envelope, and In addition, voiced sounds have a pitch structure, while unvoiced sounds have a fine spectral structure that is represented by a stochastic process. This is because the spectral structure of a simply added CSM and a speech signal is completely different. Conceivable.

Therefore, to synthesize speech using CSM,
It is necessary to use some method to spread the line spectrum into a continuous spectrum. In other words
CSM speech synthesis can be thought of as generating a speech spectrum pattern from a speech feature vector pattern expressed by a line spectrum as shown in FIGS. 1 and 2.

In the present invention, the following method is used to perform the above-described spread spectrum in CSM speech synthesis.

That is, since voiced sounds have a clear pitch structure, the phase of each of the n sine waves specified as described above is reset every pitch period. This makes it possible to easily generate a spectral envelope and a fine pitch spectral structure.

Furthermore, by applying special time window processing to the above-mentioned phase reset waveform as detailed in the explanation of the embodiment, discontinuity in the synthesized waveform at the time of phase reset can be removed and continuity of the audio waveform can be ensured. ing.

Through the above implementation, the CSM line spectrum shown in Figure 2 is diffused as shown in Figure 3A, and changed to a spectrum having a spectral envelope and pitch fine structure, which is sufficiently practical for auditory purposes. Experimental results have shown that durable sound quality can be obtained.

For reference, FIG. 3B shows a CSM spectrum obtained by simple addition without performing the above-mentioned processing. As mentioned above, a waveform with such a spectrum can only be heard audibly as a synthesized sound of sine waves, and the purpose of reproducing speech is hardly achieved.

The above is for voiced sounds, but in the case of unvoiced sounds, it is performed as follows. That is, in the case of voiced sounds mentioned above, the phase reset and special time window processing performed at every pitch period are performed, whereas in the case of unvoiced sounds, randomly generated pulses are used as a stochastic process instead of pitch periods. , the above-mentioned processing is performed every time this pulse is generated.

By using the above method, it is possible to perform CSM speech synthesis that is auditorily sufficient for practical use.
Note that since the above CSM speech synthesis is a synthesis method that does not use a filter, there is no need to consider stability on the synthesis side. For this reason, when transmitting the information of m _i and ω _i to the synthesis side and using it as a communication means to reproduce the voice by the above-mentioned CSM speech synthesis on the synthesis side, the quality of the transmission path is relatively poor and the transmission is interrupted. Even when an error occurs, it is highly resistant to errors and maintains relatively good sound quality.

However, on the other hand, if there are few errors,
The disadvantage is that the sound quality is slightly inferior to speech synthesis using LPC parameters using the same amount of information.

Therefore, in this vocoder, speech synthesis is performed using LPC parameters on the synthesis side.
The LPC speech synthesis means, the above-mentioned CSM speech synthesis means, and the above-mentioned CSM speech synthesis means are separately provided, and the outputs of both are subjected to variable weight synthesis according to the quality evaluation of the transmission path. Let's do it. As a result, when the quality of the transmission path is very good, the output of the LPC speech synthesis means is used, and when the quality of the transmission path is very poor, the output of the CSM speech synthesis means is used. In the transition region of
This eliminates the unnaturalness caused by the sudden switch from LPC synthesized sound to CSM synthesized sound.

Furthermore, as described later, the ratio between the output of the CSM speech synthesis means and the output of the LPC speech synthesis means is used to evaluate the line quality. is almost unaffected, whereas the latter makes use of the characteristic that the LPC synthesis filter becomes unstable, which increases the probability that the output power will increase abnormally.

(Example) Next, the present invention will be explained in detail using examples.

FIG. 4 is a block diagram showing one embodiment of the present invention.

As shown in FIG. 4, this embodiment consists of a transmission side 1 including a CSM analysis section 2 and a coder section 3, a decoder section 5, a CSM synthesis section 6, a CSM/LPC parameter conversion section 7, an LPC synthesis section 8 and The receiving side 4 includes a variable multiplexing unit 9.

The input audio signal is input to the transmitting side 1 via the line 1000, transmitted from the transmitting side 1 to the receiving side 4 via the transmission line 3000, and from the receiving side 4 to the line 9.
000 as an output audio signal.

As shown in FIG. 5, the CSM analysis section 2 further includes an A/D converter 201 and a Hamming window processor 20.
2. Autocorrelation coefficient measuring device 203, CSM analyzer 2
04 and a pitch V/UV analyzer 205.

Further, as shown in FIG. 6, the CSM synthesis section 6 includes n variable frequency oscillators 601-1 to 601-n with a phase reset function, n variable gain amplifiers 602-1 to 602-n, and additive synthesis. device 603, multiplier 604, multiplier 605, V/UV switch 60
6. Variable length window function generator 607, period calculator 60
8 and a random number generator 609.

Furthermore, the CSM/LPC parameter converter 7 includes a CSM parameter/autocorrelation coefficient converter 701 and an autocorrelation coefficient/LPC parameter converter 702, as shown in FIG. 8 is LPC as shown in Figure 8.
Synthesis filter 801, pulse oscillator 802, V/
It includes a UV switch 803, a power controller 804, and a noise generator 805.

Finally, the variable load synthesis section 9 includes temporary memories 901 and 904, a power calculator 9, and
02 and 905, multipliers 903 and 906, a weight calculator 907, and an addition synthesizer 908.

Now, the operation of this embodiment is as follows. The audio waveform to be transmitted is supplied to the A/D converter 201 via the input line 1000 (the fifth
), where the amplitude and time axes are converted into quantized digital data, the outputs of which are sent to a Hamming window processor 202 and a pitch V/V, respectively.
It is supplied to a UV analyzer 205. The digital data supplied to the Hamming window processor 202 is subjected to weight multiplication by a known Hamming window function for each predetermined frame, and is supplied to the autocorrelation coefficient measuring device 203 for each frame of data. .

The autocorrelation coefficient measuring device 203 calculates the lowest N autocorrelation coefficients υ _l (where l=
1, 2, ..., N).

In other words, data for one frame is X _t (where t
=0, 1,...M-1), each of N υ _l is obtained by performing the calculation process υ _l =1/M _M-1 〓 ^t=l X _t X _tl .

The thus obtained set of υ _l for each frame is supplied to the next CSM analyzer 204, and υ _p (that is, υ _p = 1/M _M-1 〓 ^t=0 X ² _t ) is It is supplied to the coder section 3 as power information in the frame.

Now, the autocorrelation coefficient υ _l for each frame mentioned above
The CSM analyzer 204, supplied with the set m _i , ω _i ( However, i=1, 2,...n)
and supplies it to the coder section 3.

In addition, the pitch V/UV receiving the digital data of the original audio signal from the A/D converter 201
The analyzer 205 extracts information regarding the pitch period and voiced sounds (t)/unvoiced sounds (UV) and supplies the extracted information to the coder section 3.

Now, the coder section 3 receives the above signals.
These signals can be easily separated on the receiving side, and are coded and synthesized with a coarseness commensurate with the required sound quality and the transmission capacity of the transmission line, and then sent to the transmission line 30.
00 to the receiving side 4.

Now, on the receiving side 4, the thus transmitted signal is received by the decoder section 5 as shown in FIG. 4, and various signals supplied to the input side of the coder section 3 of the transmitting side 1 are separated. These signals are supplied to a CSM synthesis section 6 and a CSM/LPC parameter conversion section 7, respectively.

In the CSM synthesis unit 6 (Fig. 6), ω _i (ω ₁ to ω _o ) specifying the angular frequency of each of the n waves of the CSM supplied from the decoder unit 5 is converted to Variable frequency oscillator 601-1 to 601
−n frequency control input terminals to change the output angular frequency of these oscillators to a specified angular frequency ω ₁ ~
Set to ω _o . Furthermore, the parameters m ₁ to m _o specifying the intensity (power amplitude) of each of the n waves of the CSM are the same as those of the n variable gain amplifiers 602-1 to 602-n.
The oscillation power of each frequency is thereby controlled to a specified value.

The n outputs obtained in this way are added to the adder combiner 6
After addition and synthesis are performed in step 03, the signals are supplied to the next multiplier 604.

Now, the pitch period data and V/UV information (voiced sound/unvoiced sound information) separated by the decoder section 5 are supplied to a V/UV switch 606, respectively.

On the other hand, the random number generated by the random number generator 609 is
It is supplied to the period calculator 608, where it is converted so that the random number distribution width and its lower limit become specific values, and is supplied to the V/UV switch 206 as a data string that determines the phase reset time interval during unvoiced sound. be done.

Thus, when the above-mentioned V/UV information specifies a voiced sound (t), the switch 606 selects the above-mentioned pitch period data side and supplies it to the variable length window function generator 607. Furthermore, if the aforementioned V/UV information specifies unvoiced sound (UV), the switch 606 selects the data side representing a random time interval generated in the stochastic process of the output of the aforementioned period calculator 608. , instead of the digital data string representing the pitch period described above, the window function generator 6
Supply on 07.

Now, the window function generator 607 is for generating a window function that ensures the continuity of the audio waveform except for the discontinuity that occurs in the output waveform due to the above-mentioned phase reset. It also generates closely time-related phase reset pulses.

As mentioned above, a data string specifying the interval between the next phase reset pulse is input to the window function generator 607 via the switch 606. impulses having a time interval specified by , are supplied via line 6071 to the phase reset terminals of variable frequency oscillators with phase reset function 601-1 to 601-n, thereby Performs a phase reset of the oscillator.

Now, the window function generator 607 generates a variable length window function w(x) as shown below in synchronization with the generation of the above-mentioned phase reset pulse.

That is, if T is the value of the interval between phase reset pulses at that point specified by the input data, and x is the elapsed time since the previous phase reset pulse was generated, w(x) = 0.5 + 0. .5cos(πx/T) However, it generates a window function expressed as 0<xT. This window function w(x) is shown in FIG. 10A. The above-mentioned value of T represents the pitch period in the case of voiced sounds, and represents a variable generated in a stochastic process in the case of unvoiced sounds, so it changes over time in both cases. Therefore, this window function w(x) has a variable length and is synchronized with the generation of the phase reset pulse described above in the relative time relationship shown in FIG. 10B (the start and end points of the window function are (This almost coincides with the time point at which the phase reset pulse is generated.)

The window function thus generated is line 6072
is supplied to multiplier 604 via . As a result, the multiplier 604 generates n sine waves whose phase is reset for each phase reset pulse synthesized by the adder synthesizer 603, and the above-mentioned window function generated in synchronization with each phase reset pulse. The product with w(x) is obtained. The waveform obtained in this way is such that each sine wave is continuously converged to 0 by multiplying by the window function (x) just before the phase is reset, and each sine wave rises from 0 at the time of the phase reset. Therefore, continuity of the waveform is ensured, and discontinuity occurring in the phase reset waveform can be removed by multiplication by the window function w(x).

The output of the multiplier 604 from which discontinuities have been removed is supplied to the next multiplier 605, where it is loaded with the power information υ _p of each frame separated by the decoder section 5, and is subjected to CSM synthesis. Variable weight synthesis unit 9 via line 6000 as synthesized speech from unit 6
supplied to

Now, on the other hand, CSM/LPC parameter conversion section 7
(FIG. 7), ω _i (i=
1 to n) and parameters m _i (i=1 to n) specifying the intensity (power amplitude) of each of the n waves are supplied to the CSM parameter/autocorrelation coefficient converter 701,
Here, it is converted into an autocorrelation coefficient.

That is, n of the CSM specified as above
Let γ _l be the autocorrelation coefficient of tap l of the composite waveform of sine waves, then γ _l is calculated using the supplied parameter ω _i m _i as follows: γ _l = _o 〓 ⁱ⁼¹ m _i cos l ω _i It can be easily obtained by calculating l=1, 2,...,N.

The set of autocorrelation coefficients thus determined is supplied to the next autocorrelation coefficient/LPC parameter converter 702, where the LPC
It is converted into parameters and supplied to the LPC synthesis section 8.

For example, the LPC parameter α _i (i=1,...N) is
It can be easily determined by solving the following equation using Durbin's method.

In the LPC synthesis section 8 (Fig. 8), the supplied LPC parameters α _i (i=1, 2,...N) are
It is set as a constant of the LPC synthesis filter 801.

On the other hand, the pitch period data separated by the decoder section 5 is supplied to a pulse oscillator 802, which controls the period of this output pulse to be a specified pitch period.

Further, V/UV information (voiced sound/unvoiced sound information) is supplied to the V/UV switch 803 as a switching control signal, and in the case of voiced sound (t), the switch 803 selects the output side of the pulse oscillator 802. , unvoiced (UV)
In this case, the switch 803 is controlled to select the output side of the noise generator 805.

The output of the switch 803 selected in this way is supplied to the power controller 804, where it is subjected to power control based on the power information υ _p separated by the decoder section 5, and then output as a drive signal to the LPC synthesis filter 801. Supplied.

As described above, the LPC synthesis filter 801 synthesizes the audio signals, and the synthesized signal is sent to the LPC synthesis section 8.
It is supplied to the variable weight synthesis unit 9 via line 8000 as a synthesized speech signal from.

Now, from the CSM synthesis section 6 and the LPC synthesis section 8 via line 6000 and line 8000,
The variable weight synthesis section 9 (FIG. 9), which receives the respective synthetic audio signals, synthesizes the input audio signals from both in the following manner.

First, each input audio signal is supplied to power calculators 902 and 905. As a result, the power obtained by measuring the momentary power of each signal (power obtained by multiplying by an appropriate window function) is supplied to the weight calculator 907. The weight calculator 907 calculates the weight signal determined by the ratio of the two powers, that is, the weight W _CSM and the weight
W _LPC is generated and supplied to multiplier 903 and multiplier 906, respectively.

On the other hand, each of the input audio signals is supplied as the other input of the aforementioned multiplier 903 and multiplier 906, respectively, via temporary memory 901 and temporary memory 904. Temporary memory 90
1 and temporary memory 904 are used to adjust the delay time of each signal, eliminate the difference in delay time between the two, and eliminate unnaturalness during synthesis.

Now, each signal after delay correction is
After weight multiplication (weighting) is performed by the weights W _CSM and W _LPC described above in the multiplier 903 and the multiplier 906, the adder 90
8, the summation is performed and the output line 90
00 to be output as the final synthesized speech signal.

By using the above synthesis method, the weights for each input signal are controlled and synthesized according to the power ratio of each input audio signal.An example of this load control characteristic is shown in the 11th
As shown in the figure.

In other words, the ratio r between the CSM side power and the LPC side power
In the normal state where is almost equal to 1,
The weight W _LPC for the LPC side is almost 1,
On the other hand, the weight W _CSM on the CSM side is almost 0.

The error rate of the transmission path deteriorates, and as mentioned above,
When the power on the LPC side increases and the ratio between the two decreases,
W _LPC gradually decreases from 1, and W _CSM gradually increases from 0. Thus, the power on the LPC side is
In a state where the power on the CSM side is increased by about 4 times (r≈0.25), the weights W _LPC and W _CSM are almost equal to each other and are combined as 0.5.

Furthermore, when the error rate deteriorates and the power on the LPC side increases, W _CSM becomes almost 1 and W _LPC becomes almost 0.

As described above, when the transmission path is in good condition and there are no errors, the output from LPC speech synthesis with better sound quality is used, and when the transmission path is in poor condition and many errors occur, an error occurs. Output from CSM speech synthesis, which is highly resistant to noise, is used. Then, in the intermediate transition region, the weight for the synthesis of both changes continuously and shifts from one to the other, thereby causing a sudden switch in the output of the two synthesis methods, which have different sound qualities. I'm trying to eliminate the unnaturalness.

As described above, according to this embodiment, the CSM parameters (pseudo-formant information) are extracted by CSM voice analysis on the transmitting side, encoded together with other sound source information, and sent to the transmission path, and the receiving side is a sandwiched CSM speech synthesizer that directly performs CSM speech synthesis using the transmitted CSM parameters, and a
A quasi-formant vocoder that is equipped with an LPC speech synthesis section that performs LPC speech synthesis after converting into LPC parameters, and performs variable weight synthesis of the outputs of both as described above, thereby demonstrating the excellent characteristics of each speech synthesis method. can be provided.

Below, various circuits used in the CSM synthesis section 6 will be explained.

First, variable frequency oscillator 6 with phase reset function
An example of the circuit of No. 01 is shown in FIG. The constant current source 601 is controlled by the voltage applied to the frequency control terminal 6011.
2 and 6013 to charge and discharge the capacitor 6014, thereby making the oscillation frequency variable. The oscillation voltage waveform at the υ point is a triangular waveform that linearly rises and falls between the reference voltages +V _r and -V _r .

When an impulse is applied to the phase reset terminal 6015, the υ point is momentarily grounded and forcibly pulled back to 0 potential, and oscillation is restarted from there to perform a phase reset. The triangular wave oscillation output at the υ point is input to a sine wave converter 6016, converted to a sine wave, and outputted from a terminal 6017, which is used as the output of the oscillator 601. Sine wave converter 601
6 can be easily realized by, for example, reading out a sine function value stored in a ROM using an input waveform.

Further, such a variable frequency oscillator with a phase reset function can be easily realized using a computer program.

Next, the circuit side of variable gain amplifier 602 is shown in FIG. Adding the signal to be amplified to the terminal 6021,
By applying a control signal to the terminal 6022, the amount of negative feedback is controlled and an output having a controlled amplitude is obtained at the output terminal 6023.

In addition, it can also be realized by using an analog multiplier, or by using an analog waveform input as the reference voltage of the D/A converter and using control information expressed in digital quantities as the digital input. This can also be easily realized by

Next, an example of a circuit of the random number generator 609 is shown in FIG. It has a period of 2 ¹⁵ −1 with a 15-stage shift register 6091 and one half adder 6092.
Generates 15th order M-sequence pseudo-random numbers. The next random value is obtained by applying a shift pulse to clock terminal 6093 at the required time.

Next, a block diagram of the period calculator 608 is shown in FIG. 15A. This is suitable for using the random numbers uniformly distributed in the range of 0 to ²¹⁵ -1 output from the random number generator 609 mentioned above as random numbers for specifying the time interval of the phase reset pulse during unvoiced speech. It is composed of a constant multiplier 6081 and a constant adder 6082. By this, the first
As shown in Figure 5B, the random number distribution width D and the lower limit L
and can be set to appropriate values.

Next, an embodiment of the window function generator 607 is shown in FIG. This includes a register 6073, a presettable down counter 6074, and a counter 607.
5. Contains a read-only memory (ROM) 6076. Data T specifying the phase reset pulse interval supplied from the switch 606 is stored in the register 6073. Down counter 607
4 is a counter that counts high-speed clock CLK of a constant period.First, the contents T of register 6073 are counted.
is preset and counted down using the clock CLK. When the contents of the counter 6074 reach 0, a pulse is generated from the output terminal, thereby presetting the contents of the register 6073 again and starting counting down this value. Thus, a pulse train having a period proportional to T (for example, T/k) is generated at the output 6074-1 of the down counter 6074. This pulse train is processed by the counter 6075.
added as a clock. The count output 607 of the counter 6075 is incremented by this clock.
5-1 is the window function w(x) that is added to the ROM6076 as an address designation signal and written there.
data are read out in order and output to line 6072. When the contents of the counter 6075 reach k,
The last data in ROM 6076 is read and a reset pulse is output on line 6071. This reset pulse is generated by the oscillators 601-1 to 601-n.
It is used as the above-mentioned phase reset pulse supplied to the phase reset terminal of the register 6073, and is also used to set the next input data in the register 6073. In this way, phase reset pulses whose interpulse intervals have successively specified values and a variable length window function w(x) synchronized with these pulses as shown in FIG. 10B are generated.

Finally, I will explain CSM analysis.

As mentioned above, CSM analysis uses, for each analysis frame, N low-order tap values of sample autocorrelation coefficients calculated directly from the speech waveform to be represented;
Determine the angular frequency w _i of each sine wave to be synthesized and its intensity (power amplitude) m _i so that the N low-order tap values of the synthesized wave (sum of n sine waves) match. It is to be.

Now, if the autocorrelation coefficient of tap l of the composite wave is γ _l , then as mentioned above, γ _l = _o 〓 ⁱ⁼¹ m _i cos l ω _i .

On the other hand, the sample of the audio waveform to be represented X _t
Therefore, the sample autocorrelation coefficient υ _l of tap l in a certain frame is υ _l =1/M _M-1 〓 ^t=l X _t X _tl (1).

From this, γ _l =υ _l ...(2) l=0, 1, 2,...N However, if N=2n-1, the following matrix expression is obtained.

However, the above equation cannot be solved by simple matrix operations because ω _i and m _i are unknown. Therefore, we set ω _i = cos ^-1 x _i ...(4) and perform the substitution coslω _i = cosl cos ^-1 x _i ≡T _l (x _i ) ...(5). This T _l (x) is a Tchebycheff polynomial. If you do this substitution
Equation (3) is converted as follows.

However, in general, x ^l can be expressed as a linear combination of T _p (x), T _l (x)...T ₁ (x).

That is, x ^l = _l 〓 ^j=o S ^(l) _j T _j (x) …(7) where S ^(l) _j is the inverse Tchebycheff coefficient.

Using this S ^(l) _j , the sample autocorrelation coefficient υ _j
Define the linear combination A _l as shown below.

A _l = _l 〓 ^j=o S ^(l) _j υ _j …(8) However, l=0, 1, 2,…, 2n−1 Then, on the left and right sides of equation (6), respectively
By using the relationships of equations (7) and (8), the following relational expression is established.

Now, here, n with zero points at x ₁ , x ₂ , ..., x _o
Define the following polynomial Pn(x)≡ _o 〓 ^k=o P ⁽ⁿ⁾ _k x ^k = _o Π ⁱ⁼¹ (x−x _i ), and use this Pn(x) to solve equation (9). Create an equation similar to the left side, _o 〓 ⁱ⁼¹ m _i Pn(x _i ) ^l _i , and examine it. It is clear that the above equation is 0, but it can be further rewritten as follows.

0= _o 〓 ⁱ⁼¹ m _i Pn（x _i x ^l _i ＝ _o 〓 ⁱ⁼¹ m _io 〓 ^k=o P ⁽ⁿ⁾ _j ＝ _o 〓 ^k=o P ⁽ⁿ⁾ _ko 〓 ⁱ⁼¹ m _i x ^k+l _j = _o 〓 ^k=o P ⁽ⁿ⁾ _k A _k +l From the above, the following formula is obtained with l=0, 1, 2,...n.

However, since P ⁽ⁿ⁾ _o = 1 holds true. The matrix formed by A _i on the left side is generally called the Hankcl matrix. As described above, each A _i is given by equation (8) from the sample autocorrelation coefficient υ _j of the speech waveform to be expressed and is known.

Therefore, by solving equation (10), P ⁽ⁿ⁾ _p , P ⁽ⁿ⁾ ₁ ,...P ⁽ⁿ⁾
_o-1
The value of can be found.

Once each P ⁽ⁿ⁾ _i is found, as a solution to the n-dimensional equation Pn(x)=x ⁿ +P ⁽ⁿ⁾ _o-1 x ^n-1 +...P ⁽ⁿ⁾ ₀ =0, {x ₁ , x ₂ …, x _o } is found.

From this, each CSM frequency ω _i is obtained from ω _i =cos ⁻¹ x _i in equation (4), and the CSM intensity m _i is obtained using the following equation derived from equation (9).

Note that the matrix on the left side of the above equation is generally Vander
This is called the Monde (Juare del Monde) matrix.

To summarize the above, the analysis algorithm for CSM analysis is as follows.

(1) Calculate the sample autocorrelation coefficient.

υ _l =1/M _M-1 〓 ^t=l X _t X _tl (2) Define A _l using the inverse Tievisiev coefficient.

A _ll 〓 ^j=o S ^(l) _j υ _j (3) Find P ⁽ⁿ⁾ _i by solving the Hankel matrix equation by A _l .

(4) P ( _n ⁾ Find n x _i by solving an n-dimensional algebraic equation with i as a coefficient.

Pn(x)≡x ⁿ +P ⁽ⁿ⁾ _o-1 x ^n-1 +…+P ⁽ⁿ⁾ ₁ x+P ₀ =0 (5) Performing cos inverse transformation to find the CSM angular frequency {ω _i }, _i = cos ^-1 x _i (6) Find the CSM intensity {m _i } by solving the Van der Monde matrix equation.

By performing each of the above steps, CSM
Each angular frequency {ω ₁ , ω ₂ ...ω _o } and the intensity of each wave {m ₁ , m ₂ , ... m _o } can be obtained.

Note that, as an efficient method for solving the above-mentioned Hankel matrix equation, a method is known in which initial conditions are given and solutions are sequentially obtained.

Furthermore, since it has been proven that the above nth-order algebraic equation has only real roots, the roots can be found using the Newton-Raphson method or the like.

Further, as an efficient method for solving the above-mentioned Vander Monde matrix equation, it is possible to use a method of sequentially obtaining solutions by performing triangular matrix formation.

In the CSM analysis described above, this example uses a method of solving an equation in which the sample autocorrelation coefficient and the CSM autocorrelation coefficient are equal. A method based on line spectral frequency calculation and residue calculation can also be used.

Further, in this embodiment, a variable length window function having a specific function form is used, but this function form is merely an example, and it is clear that other function forms may be used.

Furthermore, the randomization generator, the period calculator, etc. are also shown as examples, and there is no need to be limited thereto.

In the above embodiments, m _i was used as the quantity specifying the intensity (power amplitude) of each sine wave of the CSM, but as a signal that actually controls the variable gain amplifier, √ _i which is proportional to the amplitude is used. It is clear that this can also be done using

In addition, in this example, as a method for evaluating the error rate of the transmission path, as the error increases, the probability that the LPC synthesis filter approaches the unstable region increases. Taking advantage of the fact that the weight increases, and using the ratio of the two, calculate the weight W _CSM and the weight that should be applied to the CSM side output.
The weight W _LPC to be applied to the LPC side output is calculated, but the error in the transmission path is evaluated using another appropriate method, and this is used to calculate the weight to be applied.
It is also possible to determine W _CSM and W _LPC .

Note that the variable load composite characteristic shown in FIG. 11 is an example, and it is clear that there is no need to be limited thereto.

(Effects of the Invention) As described above, when the present invention is used, an LPC speech synthesis method that can obtain better speech quality when the error rate of the transmission path is good, and a It is possible to provide a pseudo-formant type vocoder that combines the excellent features of both the CSM speech synthesis method and the durable CSM speech synthesis method.

This makes it possible to improve the performance of the vocoder.

[Brief explanation of drawings]

Figure 1 shows an example of a voice feature vector pattern based on CSM parameters, and Figure 2 shows a CSM line spectrum and a pattern obtained from the same voice sample.
Diagram showing an example of correspondence with LPC spectrum envelope, 3rd
Figure A is a diagram showing the spectral envelope of the diffused CSM and the pitch fine structure, Figure 3B is a diagram showing the CSM spectrum obtained by simple addition, and Figure 4 is a block diagram showing an embodiment of the present invention. , FIG. 5 is a block diagram showing details of the CSM analysis section of the above embodiment,
Figure 6 shows the CSM synthesis section of the above embodiment, and Figure 7 shows
A block diagram showing the details of the CSM/LPC parameter conversion section, FIG. 8 shows the LPC synthesis section, and FIG. 9 shows the details of the variable weight synthesis section. FIG. Figure B is a diagram showing the relative time relationship between the variable length window function and the phase reset pulse, Figure 11 is a diagram showing an example of variable load synthesis characteristics, and Figure 12 is a circuit of a variable frequency oscillator with phase reset function. 13 is a diagram showing an example of a variable gain amplifier circuit, FIG. 14 is a diagram showing an example of a random number generator circuit, FIG. 15A is a block diagram of a period calculator, and FIG. 15 B is a diagram showing the distribution of random numbers output from the period calculator, and FIG. 16 is a block diagram showing an example of a variable length window function generator. In the figure, 1... Sending side, 2... CSM analysis, 3
...coder section, 4...receiving side, 5...decoder section, 6...
CSM component section, 7... CSM/LPC parameter conversion section, 8... LPC synthesis section, 9... Variable load synthesis section, 2
01...A/D converter, 202...Hamming window processor, 203...autocorrelation coefficient measuring device, 204...
CSM analyzer, 205... Pituchi V/UV analyzer,
601-1 to 601-n...Variable frequency oscillator with phase reset function, 602-1 to 602-n...Variable gain amplifier, 603...Summing synthesizer, 604...Multiplier, 605...Multiplier, 606...V/UV switching vessel, 6
07...Variable length window function generator, 608...Period calculator, 609...Random number generator, 701...CSM parameter/autocorrelation coefficient converter, 702...Autocorrelation coefficient/LPC parameter converter, 801...LPC synthesis filter , 802...pulse oscillator, 803...V/
UV switcher, 804...power controller, 805...noise generator, 901, 904...temporary memory, 902,
905... Power calculator, 903, 906... Multiplier,
907...Weight calculator, 908...Addition synthesizer.

Claims (1)

[Claims] 1. A transmitting side that sends CSM parameter information extracted by CSM analysis of an input audio signal to a transmission path, and each frequency specified by the CSM parameter information transmitted via the transmission path. A plurality of variable frequency oscillators with a phase reset function are set to
It is equipped with a plurality of variable gain amplifiers set to each intensity specified by the CSM parameter information, a variable length window function generator and a random number generator for calculating the combined output of the plurality of amplifiers, and a pitch period when synthesizing voiced sounds. The phase of each of the oscillators is reset in accordance with the period calculated from the output of the random number generator during unvoiced sound synthesis, and the phase of the operation of the window function generator is CSM synthesis means that performs CSM speech synthesis in synchronization with the reset operation; LPC synthesis means that converts the transmitted CSM parameter information into LPC parameter information and uses this to perform LPC speech synthesis; a receiving side comprising variable weight combining means configured to perform variable weight combining of the output of the LPC combining means and the output of the LPC combining means in accordance with an evaluation of the transmission quality of the transmission line; type vocoder. 2. The pseudo-formant vocoder according to claim 1, wherein the ratio between the output of the CSM synthesis means and the output of the LSP synthesis means is used as an evaluation of the transmission quality of the transmission path.