FR2466826A1 - SYNTHESIZER OF SOUNDS - Google Patents

Abstract

In this sound synthesizer, the signals from a source 15 of sound signals are sent to a unit 16 forming a filter synthesizer and which receives control parameters from a parameter transformation unit 13 and from a unit 14 interpolation of these parameters. The filter synthesizer is composed of second order circuits serving as a second order filter with zero on a unit circle in a complex plane, means for cascading second order filter circuits of different coefficients and feedback circuits with two types of cascade operating circuits while the coefficients of the second order filters are controlled by control parameters. (CF DRAWING IN BOPI)

Description

The present invention relates to a sound synthesizer, with the aid of which it

  It is possible to reconstruct a sound having substantially the same qualities as an original sound, from its characteristics transmitted or memorized in a memory in the form of a small amount of information.

  For example, in the case of the reconstitution of the

  from the characteristic parameters of the original language, according to the prior art, the output signal of a pulse generator simulating the vibration of the vocal cords and the output signal of a generator of

  noise simulating turbulence is permuted or mixed

  together, depending on whether the language is spoken or not, and

  the resulting output signal is modulated in amplitude

  to the amplitude of the speech, so as to produce

  re an excitation source signal that is applied to a

  filter simulating the resonance characteristics of the

  voice reel in order to obtain the synthesized language. A synthesizer system using autocorrelation coefficients

  (Parcor) and a synthesizer system of

  mants or characteristic sounds are examples of such a language synthesizer system using the characteristic parameters. The synthesizer system indicated in

  first place is described for example in the article by J.D.

  Markel et al., "Linear Prediction of Speech", pages 92-

  128, Springer-Verlag, 1976, in which the partial autocorrelation coefficients or coefficients denoted Parcor

  of a language waveform are used as para-

  characteristic meters. If the absolute values of the coefficients

  Parcor cients are all less than unity, the speech synthesis filter is stable. The Parcor coefficients

  may include a relatively small amount of information

  for a synthesis of language and automatic

  of these coefficients is relatively easy, but the

  Individual meters differ widely in terms of spectral sensitivity. Therefore, if all parameters are quantized using the same number of bits, the

  spectral distortions caused by quantum errors

  for the respective parameters differ widely from each other. In addition, the Parcor coefficients are poorer in terms of their interpolation characteristics and the interpolation of the parameters has the effect of provoking

  spurious noises that provide an indistinct language.

  In particular for a low bit rate, the quality of the language is altered by the spectral distortion and it is not possible to obtain a satisfactory quality of the synthesized language. Moreover, since the Parcor coefficients do not correspond directly to the spectral properties, for example to the main frequencies of a sound, the Parcor coefficients are therefore not appropriate in

  general for a synthesis of language.

  The synthesizer system of the formants or sounds

  This is described, for example, in J.L.

  Flanagan, "Speech Analysis, Synthesis and Perception", pages 339-347, Springer-Verlag, 1972. This system synthesizes the language using the main frequencies of the sounds and their intensity as parameters and whose advantage lies in the fact that the amount of information for them

  parameters can be low and can easily be

  the correspondence of the parameters to the spectral quantities

  the. However to get the main frequency of the sounds

  and their intensity, it is necessary to use

  general dynamic parameters and statistical properties of the parameters, and a complete automatic

  the main frequency of the sounds and the intensity of these

  is difficult. Therefore it is difficult to automatically obtain a high quality synthesized language and there can be a noticeable reduction in quality.

  synthesized language, due to an error in the ob-

the parameters.

  An object of the present invention is to provide a sound synthesizer, which is able to synthesize a sound of

  high quality using a small amount of information.

  Another object of the present invention is to provide

  to create a sound synthesizer

  very easy of the characteristic parameters and function-

  stably and in which the differences in the spectral sensitivity between the parameters are low and in which the precision of quantification of the

  parameters is the same in the case of the same quantity bits

  fication. Another object of the present invention is to provide a sound synthesizer which has characteristics of

  excellent interpolation techniques for the parameters used

  which is therefore able to provide a synonymous sound

  high quality thesis using a small amount of information. Another object of the invention is to provide a synthesizer of sounds that can be constructed with a

relatively simple structure.

  In a linear forecast analysis, the enve-

  The spectrum of speech or language spectrum is approximated by a transfer function of a multi-pole filter, which is given by the following relation (1): p 1 + a] Z + 2Z + + apZP ( 1)

  where Z = e i @, X is a normal angular frequency

  2nfAT, LT is a sampling period, f is a sampling frequency, p is the degree of analysis, ai (i = 1, 2, ... p) are forecast coefficients, which are parameters for controlling or adjusting the resonance characteristic of the filter, and a is the gain of the filter. Therefore, Ap (Z) is represented by the sum of two polynomials and can be expressed as Ap (Z) = 1/2 {P (Z) + Q (Z)} (2)

P (Z) = A (Z) - Z.ZPA (Z) (3)

  pp Q (Z) = A p (Z) + Z.ZPAp (Z) (4) (a) When the degree of analysis p is even, expressions (3) and (4) are put in the form of factors, as indicated below: p / 2 2 (Z) = (1 - Z) I (1 - 2cos.z + Z i-1 i (5) p / 2 (Z) = (1 + Z) i (1) - 2cos 0 z + Z2 i = l 1 (b) When the degree of the analysis p is odd, the expressions (3) and (4) are put in the form of factors, as indicated below: (pl) / 2 P (Z) = (1 - Z2) (1 - 2cosw.z + Z2) i = 1 (6) (p + 1) / 2 Q (Z) = 1 (1 - 2cosO.z + Z2) - i - i11

  wi and ai present in expressions (5) and (6) are

  under the name of spectrum pair of lines (hereinafter

  after the abbreviation LSP) and are used, in the pre-

  invention, as parameters for the representation of

  information of the spectral envelope.

  By expressing Ap (Z) as indicated by the relation (2), the transfer function H (Z) takes the following form:

H (Z) = A (Z) 1 + (A (Z) - 1)

a (7)

1 + {P (Z) - 1 + Q (Z) - 1}

  The function H (Z) is also associated with a filter having two feedback loops whose transfer functions would be P (Z) - 1 and Q (Z) - 1, respectively. The transfer functions P (Z) and Q (Z) correspond to these anti-resonant circuits, whose outputs cancel each other for wi and si. The frequency characteristic or frequency response of Ap (Z) then becomes: 2A 2Z) p22 2 | AP (Z) - 2P [cos 2 T [(Cosw - Cosw.) P / 2 + Sin 2 il (Cosw - Cosw .)] (8) jw l 1

  in which Z e By the expression (8) indicated above

  above it appears that in a region where the adjacent line spectral frequencies are close to one another, [A (Z) 12 is small and the transfer function H (Z)

  provides an intense resonance characteristic. In modification

  relying on LSP parameter values w. and G. describing the

1 1

  resonance characteristic of the transfer functions, it is possible to obtain an arbitrary spectral envelope of the language. The procedure for obtaining the LSP parameters is as follows: during the first phase of operation, the autocorrelation coefficients of the language are calculated at

  intervals of, for example, 10 to 20 ms; during the second

  of the operating phase, the coefficients of prediction a, of the transfer function H (Z) are calculated from the coefficients

  of autocorrelation, and during the third phase

  the solutions of the two polynomials P (Z) and Q (Z) from the prediction coefficients, on the basis of the relationship provided by the expression (2), which provides the

  LSP parameters i and Oi. By adjusting the coefficients of the

  synthesizer using the parameters representing the

  formation of the spectral envelope of the language, one can obtain

  a filter whose transfer function H (Z) is equal to

  slow to the spectral envelope of language. The function of

  transfer of the reaction loop located in the syn-

  thesizer is provided as a connection in case of

  second order filters, whose zeros are located on a circle of unit radius in a Z plane, as indicated by the expressions (5) and (6). Since these second-order filters are of identical constitution, it is possible to simplify this constitution by means of multiple use of a second-order filter with timeshare operation or so-called pipeline operation. It is also possible to perform the operation of the filter by means of the processing provided by an electronic computer, without realizing

  second-order filters in the form of circuits.

  As described above, in the present invention the characteristics of the filter. synthesizer

  are set or controlled by the parameters xi and ei men-

  above, but in addition to these parameters LSP w i and if one uses a fundamental frequency parameter and an amplitude parameter as is the case with this type of language synthesizers used before. By means of the fundamental frequency parameter a source of voiced sounds is controlled so as to produce a pulse or group of pulses having the frequency indicated by the parameter; the output signal from the source of voiced sounds or the output signal from a noise source is selected according to whether the sound to be reconstructed is voiced or unvoiced; the output signal

  chosen is applied to the sound synthesizer filter, and the

  of a signal present at the entrance or exit of the

  synthesizer filter is controlled by the amp parameter

  study. The parameters LSP i and O1 are subjected to a trans-

  training providing their cosine functions with the aid of a parameter transformation device

  to obtain the magnitudes -2coswi and -2cosai, which are used

  as control parameters for the control of

  coefficients of the second-order filters of the synthetic filter.

  sound and corresponding respectively to the parameters.

  The control parameters are interpolated by a

  interpolation, which puts them in the form of

  cosine parameters of LSP parameters: -2coswe and -2cos6i. Similarly the interpolation device can be used to interpolate the amplitude parameter. LSP parameters ì and Oi provide excellent flexibility

  interpolation, and the interpolation is performed at

  periods of time whose duration is equal to the period

  sampling period or twice that sampling period

  the original sound provided for the purpose of obtaining the parameters

  very; for example the LSP parameters wi and ei are updated at each frame of 20 ms and the parameters in each

  frame are further interpolated every 125 ls. It is also

  it is possible to perform the interpolation in state o

  are present the LSP parameters wi and ei and the conver-

  shot to get the command parameters.

  LSP parameters wi and e. represent a small amount of information per frame, compared with the control parameters provided for the synthesizer filter used in the past to perform the synthesis of the language and

  have a great interpolation characteristic. In con-

  sequence it is appropriate to transmit or memorize LSP parameters wx and 6. as they are, and it is possible i

  to convert LSP parameters wi and ei received or reconstructed

  killed in order to obtain command parameters for the

  synthesizer filter, used in other synthesis systems

  language, ie Parcor coefficients or linear prediction coefficients. In this way, one can also use the parameters LSP wi and if in the existing language synthesizers. The sound synthesizer according to the present invention can be used to synthesize not only ordinary language,

  but also sounds such as a timed signal tone

  re, an alarm tone, a musical instrument sound, etc. FIG. 1 represents a schematic block showing the

  fundamental constitution of an embodiment of the syn-

  sound synthesizer according to the present invention.

  FIG. 2 is a block diagram showing a specific operational example of the tone synthesizer according to FIG.

present invention.

  FIG. 3, formed of FIGS. 3A, 3b, 3C, is a circuit diagram showing an example of a filter of the

  first or second order constituting a

the synthesizer filter.

  FIG. 4A is a diagram illustrating an example of the synthesizer filter unit, in the case where the degree of

the analysis is even.

  FIG. 4b is a diagram illustrating an example of the synthesizer filter unit in the case where the degree of

the analysis is odd.

  Fig. 5 is a diagram showing the relationship between LSP parameters xi and si and the spectral envelope

of language.

  Figure 6 is a circuit diagram illustrating a

  specific operational example of the unit of the synthetic filter

  in the case where the degree of the analysis is equal to 4.

  Figure 7 is a circuit diagram illustrating a

  specific functional example of the unit of the synthetic filter

  grader, obtained by equivalent transformation of the circuit

shown in Figure 6.

  Fig. 8 is a circuit diagram showing a specific example of the unit of the synthesizer filter in the

  case where the degree of the analysis is equal to 5.

  Fig. 9 is a circuit diagram showing a

  specific operational example of the unit of the synthetic filter

  grader, obtained by means of an equivalent transformation of the

  circuit shown in FIG. 8.

  Figure 10 is a block diagram illustrating an example

  of the synthesizer filter unit, using the system

  time-share computing system (known as the "pipeline" system).

  Figures liA to 11I, inclusive, represent

  chronograms showing the variations of the signals

  in the respective elements during the course of

  filter unit shown in Figure 10.

  FIG. 12 is a circuit diagram illustrating the case where the implementation of the filter, carried out by means of the

  operation described with reference to FIG. 11, is effected

  killed by means of a serial filter connection.

  Fig. 13 is a block diagram illustrating an example

  ple of the synthesizer filter, using a microcomputer.

  Figure 14A is a diagram showing the values

  power as a function of time, in the case of

  the synthesis of words "ba ku o N ga".

  Figure 14B is a diagram showing the variations

  LSP parameters wi and 6i as a function of time in

  the case of the synthesis of words "ba ku o N ga".

  Figure 15 is a diagram showing the distribution

  the relative frequency of LSP parameters wi and 8i in

frequency function.

  Fig. 16 is a diagram showing the relationship between the number of quantization bits per frame and the

  spectral distortion due to quantification.

  Fig. 17 is a diagram showing the relationship between the spectral distortion obtained by interpolation and the frame length in the case of parameters which have been

interpolated.

  FIG. 18 is a diagram showing an example of language synthesis by conversion of LSP parameters wi and

e. in parameters a.

  Referring first to Figure 1, it can be seen that the characteristic parameters of a language to be synthesized are applied from an input terminal 1i to

  an interface unit 12, during each corresponding period

  at a constant time interval (hereinafter referred to as the frame period), for example every 20 ms, and are maintained in this interface unit 12-. Among the parameters thus introduced, the parameters LSP wi and ei

  representing the information of the spectral envelope are

  sent to the parameter transformation unit 13; and,

  among the parameters representing the information of the

  In addition, the amplitude information is sent to a parameter interpolation unit, and the other parameters, namely the information representing the fundamental period (base interval) of the language and the information indicating whether the language is a parameter. voiced or unvoiced, are sent to

  a unit 15 for producing sound signals.

  In the parameter transformation unit 13, the input parameters LSP w. summer. are transformed into the control parameters -2cosw. and -2cose; for a unit 16 forming the synthesizer filter and which are sent to the parameter interpolation unit 14. In this unit

  14 of interpolation of parameters, the interpolation values

  control parameters and the amplitude parameter

  sound source are calculated respectively at intervals of

  time valles regular so that the spectral envelope

2466826-

  can undergo gradual variation. The parameters of com

  interpolated command are sent to the synthesizer filter unit 16 and the sound source amplitude parameter

  is sent to the production unit 15 of sound signals.

  In the sound signal production unit 15, a sound signal depending on the characteristics of the language is produced from the information of the basic interval and the information concerning the loud or muffled sound, and the sound signal thus obtained shall be sent to Unit 16 of

  synthesizer filter as well as the interpolated parameter of am-

  sound source. In unit 16 of the synthetic filter

  a synthesized language is produced from the sound signal and control parameters. The output signal of the unit 16 of the synthesizer filter is sent to the unit

  17 of digital-to-analog conversion, whose output terminal

  tie 18 delivers a corresponding analog signal. A uni-

  The control unit 19 produces different clock rates to properly operate the speech synthesizer and sends said clock rates to the units.

respectively.

  Figure 2 shows each unit of the figure

  1 in a slightly more detailed form. In each period

  frame, the information concerning the voiced or voiceless sound of the language is sent from the interface unit 12 to a register 23 of voiced sounds and to a register 24 of unvoiced sounds and a voice frequency parameter indicating a basic interval of the voice is stored in a register

  25. The content of the interval register

  the base 25 is preset in a regressive counter 27.

  The down counter 27 counts down, according to a regressive count.

  sif, the pulses of a sampling frequency

  from a terminal 26 and whenever its contents are canceled,

  the counter 27 performs during this time interval the pre-

  positioning the contents of the basic interval register

  25 and simultaneously sends a pulse to a gate 31.

  At gate 31 are applied the output signal

  from the voiced sound register 23 and one or more output pulses from a pulse generator 28, and when these input signals coincide, the content of a sound source amplitude register 34 is sent by intermediate of the gate 31 to an adder 32. In other words, when the language to be synthesized is an intense sound, the amplitude information is sent to the adder 32 from the source amplitude register 34 during each period of the fundamental speech frequency of the base interval register 25, while the amplitude information from the sound source amplitude register 34 is preset during this period.

  time interval from the interpolation unit 14.

  In the case where the language to be synthesized is an unvoiced sound, the output signal from the unvoiced sound register 24 and a pseudo-random serial pulse from a pseudo-random signal generator 36 are sent to a gate 37 and, at each coincidence of the two input signals, the amplitude information located

  in the amplitude register of sound source 34 is

  traveling through gate 37 to adder 32.

  A sound signal thus obtained from the adder 32 is

  it is amplified, if necessary, by an amplifier 39, and is then sent to the unit 16 of the language synthesizer filter. In the parameter transformation unit 13, the parameters LSP wi and Oi and the amplitude parameter are positioned in a register 21, during each frame period, from the interface unit 12. The parameters LSP wi and 6i are sent to a parameter converter 22, in which they are transformed to provide

  the command parameters -2coswi and -2cos6i. The converts

  parameter 22 is constituted for example by a

  conversion table of a fixed memory (ROM), which is

  in such a way that when access to that memory has been

  place with addresses corresponding to wi and Oi, the parameters

  very -2coswi and -2cos6i are extracted by reading. A re-

  The shift detector 20 alternately receives the output signal from the parameter converter 22 and the amplitude parameter stored in the register 21 and converts them to provide a serial signal which is applied to

  the interpolation unit 14 of the parameters.

  In the illustrated example, the interpolation unit 14

  parameters is represented as having to achieve a

  linear interpolation When a switch 29 is connected, the parameters of a frame are sent to a subtractor 30 in which a difference between the parameter is detected.

  and the parameter of the previous frame from an addi-

  33. The difference is stored in a register

  of differences 38, via a switch 91.

  Then the switch 91 is switched to the output side of the difference register 38 and the contents thereof is put into circulation. At this time, the contents of the difference register 38 are taken from higher bit positions

  a predetermined bit position and is sent to the ad-

  editor 33, in which it is added to the contents of an interpolation result register 92. For example, in the case of a period of 16 ms of update of the parameters,

  if it is necessary to provide interpolation

  128 times during a frame update period, then the interpolation step width is a value obtained by dividing the difference by 128, which is obtained by means of a seven-bit shift , of the difference

  in the difference register 38, towards the end

  mite bits of lower weight. The result of the addition made by the adder 33 is sent to the interpolation result register 92 and, simultaneously, is used as the output signal of the interpolation unit 14.

  settings. In this way are derived from the addition

  neur 33, d values which are obtained by sequential addition

  of values equal to once, twice, three times,

  30 ... the offset value of the difference register 38 in para-

  meter from the previous frame located in the result register.

  interpolation 92, at each circulation of the regis-

Differences 38.

  In this example, the interpolation unit 14 of the para-

* meters is used for the control parameter and the parameter

  amplitude meter on a time-shared basis, so that although this is not shown, the parameter of

  command and the amplitude parameter are interpolated alternately

  and the interpolation result register 92 is used in common for both parameters. The interpolated amplitude parameter in the interpolation unit 14 of the parameters is sent to the amplitude information register 34 located in the unit 15 for producing the sound signals, whereas the interpolated control parameter as indicated below above is sent to the language synthesizer filter unit 16 as information for setting the coefficient

  from this unit. We choose the period of update of the pa-

  range, that is the frame period, in the range of

  10 to 20 ms and choose the interpolation period of

  that it is between one and two intervals of

  pling. The interpolation method is not specifically limited to linear interpolation, but can be

  another type of interpolation. The aim is to guarantee

  firing of progressive variations of the interpolated parameters.

  The synthesizer filter unit 16 includes a

  reaction key feeding the output back through the

  of filter circuits 41 and 42 connected reciprocally

  in parallel. The filter circuits 41 and 42 are powered by the interpolated control parameter from an input terminal 44 and the output signals from

  filter circuits 41 and 42 are added one to the other

  by means of an adder 43, whose output signal * is, in turn, added to the input signal of the filter unit 16, in an adder 45. The sum output signal from this adder is sent back to the filter circuits 41 and 42 and, simultaneously, spells on an output terminal 55.

  In each of the filter circuits 41 and 42, a circuit having several zeros on a unit circle in a complex plane is used. The filter circuits 41 and 42 may both be constituted by a multi-stage cascade connection of first-order and / or second-order filters. In the case where the filter circuits are made in the form of digital filters, a first-order filter, such as for example that represented in FIG. 3A, which consists of a filter, may be used.

  delay line 51, the delay time of which is equal to one

  sampling period, and by an adder allowing

  to add the delayed output signal and a signal of

  non-delayed, and a second order filter such as that shown in Figure 3B, which consists of two

  delay line stages 51 and the adder 52 enable

  both to add the delayed output signal and the non-delayed input signal and a second order filter such as that shown in FIG. 3C, wherein the

  output from a multiplier 53 making it possible to multiply

  bending the delayed output signal from the first stage of the delay line 51 by the factor -2coswi, the delayed output signal from the two delay line stages 51 and the undelayed output signal are added together by the adder 52. The transfer functions of the filters represented in FIGS. 3A, 3B and 3C are respectively 1 Z, 1-z2 and 1-2cosw iZ + Z It is also

  possible to use higher order filters.

  The combination and number of such filters depends on

  the degree of the analysis and are selected as shown in Figure 4A or 4B depending on whether the degree of the analysis is even or odd. In FIG. 4A, the degree of the analysis is 10, namely an even number, and the circuit of

  filtering 41 is constituted by the series connection of a

  first order 56, having the transfer function 1-Z, and second order filters 57 to 61 each having the transfer function 1-2cosiZ + Z while the output signal present on the output terminal 55 is multiplied by

  +1/2 in a multiplier 63 and is sent to the serial circuit.

  The output signal from the second-order filter 61

  the last stage and the output signal from the multi-

  plier 63 are added to one another by an adder 62 and the sum output signal is sent to the adder

  43. In the filter circuit 42, the output signal pro-

  from the multiplier 63 is sent to a series circuit formed of a first-order filter 64, having a transfer function 1 + Z, and second-order filters 65 to 69,

  each having the transfer function 1-2cosO.Z + Z 2, tan

  say that the output signal from the serial circuit and

  the output signal from the multiplier 63 are added.

  to one another in an adder 71, whose output signal is sent to the adder 43. The multipliers

  53 of the second order filters 57 to 61 receive respectively

  the control parameters a, = -2cosw1 to a5 = -2cosw5 and the multipliers 53 of the second order filters 65 to 69 respectively receive the control parameters b. = -2cosO at b = -2cult i i 5 5

  Figure 4B represents the case where the degree of

  lysis is 11, which is an odd number. In the filter circuit 41, the first-order filter 56 used in the

  The case of Figure 4A is deleted and is replaced by a

  second order 72 having a transfer function 1-z2. In the filter circuit 42, the filter of the first

  order 64 is deleted and replaced by a se-

  order 73, whose parameter is b6 = 2cose6.

  In the filter circuits 41 and 42, the parameters

  wi and O. are the frequencies of anti-

  resonance for which the output signals of the circuits

  filtering 41 and 42 are equal to 0.5. Correspondingly

  te, in the case where the antiresonance frequencies applied to the filter circuits 41 and 42 are close to each other, the output signal from the adder 43 approaches the unit and the gain of the feedback loop

  is close to unity. Therefore a characteristic of reso-

  In this case, w1 to w5 and 1 to G5 are the antiresonance frequencies, which are characteristic of the information of the spectral envelope of the language. These parameters and the characteristic curve of the spectral envelope are interconnected by a relation as shown in FIG. 5, from which it appears that the characteristic resonance characteristic of the spectrum can be

  expressed by the spacing of neighboring parameters. These parameters

  are very close to each other in the following order-relation 0 <6 11 <02 <w2 ... <iw <iw (8 ')

  The characteristic of the synthesizer filter is

  to be stable when the above condition is met.

  Here is a description of an example

  Specific functional unit of the unit 16 of the synthesizer filter.

  According to the term between braces at the denominator of relation (7), we obtain, from relations (5), the following equations: p / 2 2 P (Z) - 1 = 1 - Z) il (1 - 2 cosi.Z + Z) - 1 i = lp / 2-1 i 2 = Z {(a1 + Z) + it (ai + + Z) 1 (1 + aZ + Z) i = 1 i + 1 j = 1 J p / 2 2 - 1 (1 + aZ + Z)} (9) j = 1 Q (Z) p / 2-1 i 2 1 Z {Q (+ Z) - 1 = Z {(b + Z) ( 1+ bz + zi = l i + 1 j = lp / 2 2 + R (1 + bZ + Z)} (10) j = l 1

at. = -2cosw.

1 1

b. = -2cosO.

1 1

0 <i '<fi (11)

  A digital filter is created which has the function

  multi-pole transfer scheme, providing

  the spectral envelope of the language provided by the former

  pressure (1), using the relationships provided by expressions (7), (9) and (10). FIG. 6 represents the case where P = 4. In FIG. 6, the parts corresponding to those existing in FIGS. 3B to 4 bear the same reference numerals. The input signal from terminal 54

  is added by the adder 45 to the output signal

  from the adder 43, and the sum output signal is sent to the output terminal 55 and simultaneously is multiplied by +1/2 by the multiplier 63. This multiplication by 1/2 corresponds to that indicated in the denominator of

  the expression (7). The output signal from the multiplier

  63 is sent to the delay line 74, whose time

  delay is equal to a sampling period, ie

  tell the unit of time. The delayed output signal is

  as an input signal to each of the second order filters 57 and 65, wherein said signal is sent

  delay lines 51, multipliers 53 and addi-

  52. In both multipliers 53,. the signals

  These multipliers are multiplied

  by a1 and b1, and the output signals provided by

  this multiplication are sent respectively to an addi-

  94, which adds said signals with the output signal from the delay line 51 present in each of the filters 57 and 65. The output signals from the two adders 94 are sent to a common adder 81 and, simultaneously, are sent to the adder 52 via a delay line having a delay time equal to one sampling period in each of the filters 57 and 65. The output signals from the two adders 52 are respectively sent as as output signals from the filters 57 and 65 to the second order filters 58 and 66 of the next stage. The filters 58 and 66 are identical in constitution to that of the filters 57 and 65 and the coefficients for the multipliers 53 are respectively a2 and b2. The output signal coming from

  the adder 94 of each filter is sent to an addition

  82 to be added to the output signal of the addi-

  81. The output signals from the additions

  52 of the two filters 58 and 66 are sent to the addition

  43 in order to be subtracted from each other, said addi-

  43 receiving further the output signal from

of the adder 82.

  The delay line 74 corresponds to the outside Z

  braces in expressions (9) and (10), and

  filters 57 and 58 each constitute a filter of the second

  dre having a transfer function 1 + Z (aj + Z) and similarly the filters 65 and 66 each constitute a second order filter having a transfer function 1 + Z (bj + Z). Therefore the series connection of the second order filters 57 and 58 provides the third term between the brackets in the expression (9) and the line

  delay 51, the multiplier 53 and the adder 94

  killed in the filter 58 provide (ai + 1 + Z); consequently

  this circuit and the second order filter 57 provide the second term between the brackets of the expression (9) and the output signal is sent by the adder 82 to the adder 43. The delay line 51, the multiplier 53 and the adder 94 present in the filter of the second

  order 57 realize (a1 + Z) and the output signal is

  sent to adder 43 through the adders

  81 and 82. In this way, the terms existing between

  colads in the expression (9) are produced by the second-order filters 57 and 58 and by the adders 43, 81 and 82. In the same way, the terms present in the braces located in the expression (10) are realized by the second order filters 65 and 66 and by the adders 43, 81 and

  82. The forms of expressions (9) and (10) differ only in

  the fact that the signs of the third terms situated between the braces are opposite and, in view of this difference, the sign of the entry to the adder 43 is different. Therefore, the adder 43, the second order filters 57, 58, 65 and 66, the multiplier 63 and the delay line 74 provide the expression (2) and the arrangement shown in Fig. 6 embodies the expression (1). ) as a whole. In this arrangement, the expressions (9) and (10) are materialized by the fact that the filtering circuit 41 is produced with a series assembly of (P / 2) second-order filters 57 and 58 and the filtering circuit 42. by a montage

  series of (P / 2) second order filters 65 and 66 in the

  reaction, taking the knots out of the filters of the second

  order of the filter circuit 41, that is to say in

  both the jacks 96 and 97, on the output sides of the adders 94 so as to obtain the total sums with the adders 81, 82 and 83. The arrangement for taking the output signals on the jacks of the circuits of filtering will be referred to below as

  of "tap output link type arrangement".

  In FIG. 6 the second order filters are arranged with respect to the adder 43 in such a way

  that the values of j are in ascending order in the direction of

  said adder 43, but they can be arranged according to a

  descending order of the values of j. In such a case, for example

  pl, as shown in FIG. 7, the output signal from the delay line 74 is sent to the filters of FIG.

  second order 58 and 66, the output signals of which are

  sent through the second-order filters 57 and the adder 43. In FIG. 7, the preceding stage of each second-order filter of FIG. 6 is permuted with respect to the immediately following stage; indeed the

  circuit 94 making it possible to add to one another the

  output signals from the delay line 51 and the multiple

  The driver 53 is exchanged with the delay line 95. The output signal from the delay line 74 is sent via the connection points 96 and 97.

  at the nodes of the second order filters 57 and 58. In addition,

  In other words, the arrangement of the circuit of FIG. 6 is of the tap output connection type, while the arrangement of the circuit of FIG. 7 is of the tap input connection type. The circuit starting with the link socket 96 and ending with the adder 43 forms the first term located between the expression hooks (9) and the circuit from the link socket 97 to the adder 43 form

  the second term located between expression braces (9).

  The second order filters 65 and 66 of the filter circuit

  41 are also formed in a similar way. With regard to

  surround the filter circuit 41, the output signal of the

  delay line 74 is multiplied by -1 in a multiplication

  98, in order to materialize the minus sign for the third

  sth term located between the braces of the expression (9).

  When b is odd, we derive the following identical equation:

  te from the expression (8) corresponding to the terms located between the braces to the denominator of the expression (7). (p-3) / 2 i 2 P (Z) -1 = Z {(a1 + Z) + E (ai + 1 + Z) I (1 + aZ + Z) i = lj = 1] (p-1) ) / 2) 2 - ZE (1 + aZ + Z)} (12) j = l J (p-1) / 2 Q (Z) - 1 = Z {(b1 + Z) + Z (bi + l +) Z) i = 1l ix (1 + bZ = z2)} (13) j = l 1 a1 = -2cos bi = -2cos O <wi, ai <'(14)

  As in the case where p is even, two types of fil-

  digital terminals, designated respectively by the circuit terms "of the tap output link type" and "of the type

  with a "plug-in" connection are embodied in

  my like those shown in Figures 8 and 9,

  according to the relationships of the expressions (7), (12) and (13).

  In FIGS. 8 and 9, it is assumed that p is equal to 5. In FIGS. 8 and 9, the filter of the first order 72 corresponds to Z in the third term between braces of the expression

  (13) and the second-order filter 73 must provide a

  such that the products of the transfer functions (1 + blZ + Z2) and (1 + bZ + Z2) of the filters 65 and 66 are

multiplied by (b3 + Z).

  As will be understood from FIGS. 6 to 9, the + 1/2 multiplier 63 and the delay line 74 can also be arranged anywhere in the

  reaction loop. Since the filters of the second or-

  are of the same type, it is possible to simplify

  positive material by realizing the layout of the circuit of

  that the so-called "pipeline" operation is obtained using, on a multiplexing basis,

  time distribution or time sharing, a multiplication

  53, several adders 52 and 94 and several

  delay gages 51 and 95 constituting a second order filter.

  FIG. 10 illustrates the case where the exemplary embodiment of the filter represented in FIG. 12 is arranged in such a way as to

  provide the so-called "pipeline operation" operation.

  In this example, p = 10, and an operation of a set of parameters, sent by the interpolation unit is performed with a period of 176 clock rate pulses. In Figure 10, the elements corresponding to those of Figure 12 are designated by the same reference numerals. The input side of a 16-bit static shift register 74, which assumes the function of the delay line 74, is switched by

  a switch S1 between the output side of the decoder register

  ge itself and the output side of the adder 45. One side

  input of the multiplier 53 and the input side of the addition-

  52 are switched via a switch S2 on the output side of the shift register 74, on the

  output side of a (27-d) th shift stage counted at

  firing from the input of the shift register 74 and on the output side of a 31-bit shift register 101, d being a

  delay of the multiplier 53. The multiplication

  The converter 53 is connected at one of its ends to the output terminal 55 and to the input side of the adder 94 and supplies, at its other output end, the input signal.

  multiplier, delayed by 22 clock pulses.

  and is sent to the shift register 51 to (154 + d) bits. The output signal from an adder 81 is returned by reaction to the input side of this adder via a gate 102 and a shift register

  at 16 bits 103, which achieves a cumulative addition by the

  intermediate of the adders 81 and 82 of Figure 12. The gate 102 is open only during the interval of

  time between d + 2 and 145 + d. An entry of the addition-

  neur 43 is switched via a switch S3 between the output sides of the adders 52 and 81, and

  the other input of adder 43 is switched through

  a switch S4 between the outputs of a 16-th and a (d + 1 th shift stage of the shift register 101. The input side of the shift register 101 is switched via a switch S5 between exits

adders 43 and 52.

  The switches S1- to S5 are each connected to the fixed contact side during the operating period, i.e. during 176 clock pulse,

  during a clock period indicated by the mar-

  on the fixed contact. The shift registers 51, 95, 101 and 103 are dynamic type shift registers at respectively (154 + d) bits, (175-d) bits, 31 bits and

  16 bits and are all powered by shift clocks.

  The dashed line leading to the input of each of the adders 43, 45, 52, 81, 94 indicates the timing of the operating limit of each parameter; for example +0 indicates a repetition every 16 pulses

  clock rate and a delay in operation of each

  that adder is chosen equal to a clock rate pulse. In Figure 11 there is shown a timing diagram of the operation of each part of the circuit of the figure; FIG. 1A represents the timing of the clock, FIG. 11B represents the inputs of the coefficients a 1,

  bi and A in the multiplier 53 from the terminal of

  44; figure liC represents the multiplicand of the multiplier

  tiplicator 53; Fig. 11D shows an input signal

  multiplier 53 and resulting in the addition of

  neur 94; Figure liE represents the other input signal of the adder 94; FIG. 11F shows the signal of

  output from adder 94; Figure 11G shows

  the output signal of the adder 81 and consequently

  the contents of register 103; Fig. 11H shows the signal from the shift register 95 and entering the adder 52; FIG. 11I represents the output signal of the adder 52. FIG.

  input and output signals as signals appearing

  at the level of the respective circuit elements in the case where the second-order filters are connected in a

cascade mounting.

  As shown in FIG. 11, during the period between the clock pulse 0 and

  16, a coefficient a1 (t) and a multiplicand x1 (t) are multi-

  multiplied by the multiplier 53 in such a way that

  re to perform the multiplication in the second order filter 57 of Figure 12 and the result of the multiplication

  is obtained from a d-th pulse of clock of hor-

  box. During the period between clock pulses 16 and 32, as shown in FIGS. 11B and 11C, a coefficient b1 (t) and a multiplicand y1 (t)

  are multiplied one by the other in order to achieve the multi-

  in the second-order filter * 65. The multiplication

  of x1 (t) is delayed, as well as 22 bits of the multiplier 53, of (176 + d) clock rate pulses by the shift register 51, so that, as shown in FIG.

  11E, a multiplicand x1 (t-1) is sent to the addi-

  94 to the output signal a 1x1 output from the multiplier 53 at this time, and the output signal sum x1 '(t) is sent via of the adder 81

  to the shift register 103 for cumulation. That is,

  say that the output signal of the adder 81 is sent

  to the signal transmission system formed by the addition of

81, 82, ... of Figure 12.

  The output signal from the adder 94 is also sent to the (175-d) bit shift register

  as shown in Figure 11H. As a result

  during the period between clock pulses 0 and 16, the output signal from the shift register is x1 '(t-1), as shown in FIG. 11H, and this signal of output is added to the multiplicand x1 (t) in the adder 52, whose output signal x2 (t) is applied as an input signal to the

  second-order filter 58 of FIG.

  tie x2 (t) from the adder 52 is sent by the

  intermediate of the shift register 101 to the multiplier 53.

  As shown in Fig. 11C, the output signal x2 (t) is multiplied by the coefficient a2 (t) in the multiplier 53 during the period of clock pulse 32 to 48. Prior to this multiplication,

  bl (t) and y1 (t) are multiplied, as has been described previously.

  previously, and the output signal provided by this multiplication

  cation is treated in a similar way, which makes it possible

  denoting the output signal y2 (t) in the output of the second order filter 65 during the period between the clock pulse 48 and 64. In this way the multiplication of the coefficient a by the multiplicand x and the multiplication of the coefficient b by the multiplicand are carried out there alternately all the 16 pulses of clock rate, and the results of these multiplications

  are sent to the shift register 51, as indicated by

  by alx1, blY1, a2x2 b2Y2, ... in FIG. 11D. In

  in addition to the second-order filters 57, 58, 59, 60 and 61

  respectively, from these signals, the signals x1 '(t), x2' (t), x3 '(t), x4' (t), x5 '(t) and x2 (t), x3 (t) ,

  x4 (t), x5 (t), x6 (t), which are sent to the decay registers

  95 and 101. Similarly y1 '(t) to y5' (t) and y2 (t) to y6 (t) are respectively provided by the second order filters 65 to 69 and these output signals are sent to the registers. with shifts 95 and 101 alternating with x '(t) and x (t) respectively. During the period between clock pulses 145 and 161, the output signal Y6 delivered by adder 52 at this time and x6 in the shift register, supplied previously, are

  subtracted from each other in the adder 43 and the

  gnal (x6-y6) is sent via switch S5 to shift register 101, where it is delayed by (d + 1) clock rate pulses. The output signal

  delay is taken from switch S4 so as to be

  at the entrance of the adder 43 during the com

  taken between clock rate pulses 147 + d and

  163 + d. The output signal delivered by the decoder register

  age 103 at this moment is sent to the adder 43 by

  via the adder 81 and the switch S3.

  The output signal from the adder 43 at this instant becomes the output signal of the adder 43 on

  Figure 12 and this output signal is sent to the addition-

  neur 45, in which it is added to the input signal

  present on the terminal 54 so as to provide the signal Z (t).

  The sum output signal Z (t) is sent to the register 74

  in which it is delayed by the delay line 74 represented

  shown in Figure 12. The delayed output signal is

  multiplier 53 and at this time the coefficient A is output as an interpolated output signal.

  amplitude on terminal 44 and the multiplication signal

  cation A.Z (t) is delivered by the multiplier 53 to the output terminal 55. This multiplication is performed in the case where the output signal from the unit 16 of the

  synthesizer filter is multiplied by the amplifier information

  Study A in a multiplier 104 shown in Figure 12. From the shift register 74 is taken an output signal Z (t) / 2 which has been shifted by a bit, and this signal is taken through the switch S2 to be sent to multiplier 53 in the form of the signal

  Z (t-1) / 2, i.e., x (t) and y (t), during the immediate period

  following operation for a new set of parameters. The output signal present on the terminal of

  output 55 can also be obtained in the form of

  parallel output signals by means of a signal buffer

  output 105, a static shift register.

  The so-called "pipeline" operation described above can also be applied to other types of synthesizer filter units 16. Moreover, as is apparent from the arrangement of FIG. 10, the operation of the filter can be carried out by addition, multiplication and delay.

  so that this filter treatment can be implemented

  using a microcomputer. For example, on the

* Figure 13, by successively reading,

  interpretation and execution of programs in a memory

  of programs 107, a central processing unit 106

  troduced into this memory, from an input gate 111, a sound signal and control parameters sent

  respectively by the sound signal production unit 15

  res and by the interpolation unit 14 at terminals 108 and 109,

  and the central processing unit 106 selectively performs

  quential operations described above with reference to Figure 11. A RAM 112 is used instead

  registers 51, 74, 95, 101, 103 and 105 of FIG.

  The results of operations are recorded in the memo

  re live 112 and are extracted by reading the latter

  at a rate appropriate for the implementation of

  tions. The output signal thus obtained is sent from an output port 113 to the output terminal 55. The central processing unit 106, the memories 107 and 112 and the gates 111 and 113 are connected to a transmission bus 114. .

  The output signal of the unit 16 of the synthesizer filter is obtained by any of the indi-

  above. The output signal is converted by the uni-

  digital-to-analog conversion circuit 17 shown in FIG. 2 as an analogue signal for providing a

  speech output signal.-In the digital conversion unit

  analog signal 17, if the input signal in this unit is a serial signal, it is then applied to a shift register 115 and the contents of the latter are converted by a

  digital-analog converter 116 in analog form.

  As previously described, the parameters

  LSP wi and Oi, which are part of the characteristic parameters

  of the language used in the present invention can be obtained by calculating solutions of expressions (5) and (6). Figures 14A and 14B show the results of the analysis of a "bakuoNga" speech expression using LSP parameters wi and 6i. In FIGS. 14A and 14B, the abscissas represent the time t, while the ordinates of FIG. 14A represent the power and the ordinates.

  of Figure 14B represent the normal angular frequency

  Lisée. By considering the instants in FIG. 14B, it can be seen that the frequency increases according to the order of the parameters G1, w1, 02 'w2,'... 05, W5, that this order does not change and that the parameters 6. and w. do not coincide with each other in a frame. Therefore, it is certain that the unit 16 constituting the synthesizer filter is always stable. The frequency distributions of the parameters LSP 0i and w are shown in FIG. 15, in which the abscissas represent the normalized angular frequency f and the ordinates the relative frequency D. As shown in FIG. 15, each parameter is not distributed on a

  extended range of frequencies, but is limited to one

  latively narrow frequencies, so that the parameters

  very LSP wi and ei can be quantified in connection with the

  frequency range in which they are distributed.

  The LSP wi and ei parameters have a low

  quantization distortion. Figure 16 shows a distortion

  DS spectral resolution of a synthesized language, when different parameters have been quantified in a different way, the abscissas representing the number of quantization bits B per frame and the ordinates the spectral distortion DS. The line 117 represents the case where, given the only

  distribution of parameters, the Parcor coefficient is

  linearly adjusted only if the coefficient has been

  tributed; the line 118 represents the case where the number of quantization bits of the Parcor coefficient has been increased by taking into account the spectral sensitivity in addition to the distribution of the parameter, as in the case of line 117, in particular in the case of a significant disturbance

  spectrum; the line 119 represents the case where the parameters

  very LSP w. and have been quantified taking into account

  the distribution of the parameter; and the right 121

  presents the case where the parameters LSP wi and Si have been quan-

  tified taking into account the parameter distribution and

spectral sensitivity.

  It can be seen from FIG. 16 that in the case of using the same number of quantization bits, the

  spectral distortion becomes weaker in accordance with

  lines 117, 118, 119 and 121. Since the

  lines 119 and 121 are close to one another, the

  meters LSP wi and groin are not affected so much for distortion

  spectrum, even if one does not take into account the sensitivity

  spectral unity. Correspondingly, since it

  suffice to perform the quantification taking into account only

  the range of the parameter, the quan-

  cation is easy.The division ratio of the number of quantization bits per frame, for which the spectral distortion is 1 dB, in the case of the line 119 to the number of quantization bits in the case of the line 117, is equal at 0.7. Similarly, the ratio of the number of quantization bits per frame, for which the spectral distortion is 1 dB, for the line 118 to the same number for the line 121 is equal to 0.8. From these data,

  it will be understood that the parameters LSP wi and ei are excel-

  slow.'A dB is a difference threshold of distortion

  spectral of a synthesized language.

  Figure 17 shows characteristics of

  terpolation, for which the abscissae represent a

  frame length Tf and ordinates the spectral distortion

  the Ds. FIG. 17 shows the spectral distortion of a synthesized language in the case where a frame, in which an initial language was analyzed in 10 ms, was used as a reference, the length of the frame was increased from to 70 ms and the parameters were interpolated every 10 ms. The line 122 represents the case where the Parcor coefficients have been used, whereas the line 123 represents the case where the LSP parameters wi and si have been used. As

  this is apparent from Figure 17, in the case of the same distortion

  sion, the frame length Tf can be made longer

  using the LSP parameters that the frame length Tf obtains

  This is because the parameter update period can be increased, so that the total amount of information can be reduced by the same amount. Furthermore, since the LSP parameters have a smaller number of bits per frame than the Parcor coefficients, as can be seen from Fig. 16, the amount of information for the same distortion may be small because of producesreductionreports

  in Figures 16 and 17-; indeed, in the case of

  very LSP, the amount of information can be approximately 60% of the amount of information present in the case of

Parcor coefficients.

  In the case of the use of the LSP parameters, it is less judicious, than in the case of other parameters,

  to interpolate them with a shorter period than the period

  sampling of the original language used in the ob-

  the parameters. Experiments have shown that the interpolation period could be about twice or less than twice the sampling period of the original language, but that, when the interpolation period is also about four times the original sampling of the original language, noises or parasites appeared and made indistinct the synthesized language. Therefore, it is preferable that the interpolation period is equal to the sampling period of the original language or

double that period.

  As described above, it is relatively

  It is easy to obtain LSP parameters automatically and therefore to obtain them in real time. In addition the LSP parameters exhibit an excellent interpolation characteristic and a quantization characteristic

  with a slight variation, and allow the transmission

  sion and memorization of the language in a small quantity

  of information. In the synthesis of language, we can reconstruct

  storing and synthesizing a high quality language with a small amount of information and provided that the relation of the expression (8) is true, the stability of the filter

synthesizer is guaranteed.

  In Figure 2, it is also possible to expand

  the spectrum by producing, from the production unit 28

  pulses, a train of pulse groups, such as

  than the Barker series, instead of the pulse train.

  The interpolation unit 14 can also be provided for

  the previous stage of the parameter transformation unit13

  very. Indeed the LSP parameters coming from the unit of in-

  terface 12 may also be subjected to the cosine transformation in the parameter transformation unit 13, after having been interpolated. In this case, the use of a fixed memory is not very profitable, since its memory capacity must necessarily be enormous; therefore, it is preferable to perform the conversion of the parameters using a cosine approximation operation rather than using the fixed memory described in the example of FIG. 2. In FIG. 2, the information indicating whether the language is a voiced sound or not, is introduced and

  loaded into the register of voiced sounds 23 and in the

  24, but this information does not always need to be provided, that is to say that there is provided a detector circuit for detecting whether the parameter

  period of the fundamental frequency applied to the

  interval register 25 is or is not zero; in the.

  case of the detection of a zero, it is decided that the sound is an unvoiced sound and the door 37 is open, whereas in the case of values other than zero, it is decided that the

  sound is a voiced sound and door 31 is open. The command

  performed using the amplitude parameter can also be

  made in conjunction with the output signal of the filter unit 16, as previously described.

  in connection with the embodiment of Figure 12.

  In the foregoing, a filter which comprises, in the feedback loop, the means making it possible to connect several first-order and second-order filters having different coefficients in series and for each of which zero is on a unit circle, thanks to

  the use of LSP settings. However, the syn-

  The thesaurus does not always need to be specifically limited to such a filter and the language synthesis can also be performed by transforming the LSP parameters into some other types of parameters and using other filters. For example, as shown in Figure 18, on which the elements corresponding to those

  of Figure 1 are designated by the same reference numerals.

  frequency, the period parameter of the fundamental frequency among the characteristic parameters sent to the unit

  interface 12 is transmitted to the production unit 15 of

  sound level and the amplitude parameter is sent to the

  The amplitude parameter thus inter-

  poled is sent to the sound signal production unit 15.

  res, in which it is treated as has been described

  previously with reference to FIG. 2 and which sends a

  sound signal to the unit 16 forming the synthesizer filter. The LSP parameters are sent to a LSP parameter transformation unit 124, in which they are transformed into

  other types of parameters, such as for example a para-

  meter, a Park parameter, or the like. For example, from the LSP parameters we obtain the polynomials P (Z) and Q (Z) using the expression (5) or (6) and from the polynomials we obtain the prediction coefficients ai of the transfer function. H (Z) using expressions

  (1 and 2). Interpolating the forecast coefficients a.

  thus obtained in the interpolation unit 14, as required, the characteristics of the unit 16 of the sound synthesizer filter are controlled. The filter unit 16

  is for example in the form of a cyclic filter

  wherein, as shown in Fig. 18, the sound signal from the sound signal producing unit 15 is multiplied by a by a multiplier

  and is sent to an adder 126 in order to be sub-

  the output signal of an adder 127, and the. if-

  the output signal of the adder 126 is sent to the input terminal 55. The output signal thus obtained on the terminal

  output 55 is sent to a series of lines

  late D1 to Dp, each of which provides an equal delay time

  to a sampling period. The output signals pro-

  from the delay lines Di to Dp are multiplied respectively

  by coefficients a to a, from the interpolation unit 14, in multipliers M1 to Mp. The

  The output signals provided by the multiplications are

  sequentially, then are added between

them in the adder 127.

  It will be noted that many modifications and variations

  can be made without departing from the concepts

  provided by the present invention.

Claims (18)

  A sound synthesizer, wherein a sound source signal and control parameters for controlling the characteristics of a filter are applied to a synthesizer filter unit and wherein the coefficients of the synthesizer filter unit are controlled by control parameters so as to obtain a synthesized sound signal, characterized in that the unit (16) of the synthesizer filter is constituted by second-order filter circuits (57-61, 65-69) serving as second-order filters, with respective zeros located on   a unit circle in a complex plane, by means of   putting cascade of such circuits of   second-order filtering with different coefficients   and feedback circuits (41, 42) for returning the output signal from the synthesizer filter unit to the input of the synthesizer   two types of such cascade operation circuits.   Sound synthesizer according to claim 1,   characterized in that the source (15) of sound signals   res is constituted by a source (25) of period signals   fundamentally controlled by a fundamental period parameter   to produce an impulse or group of   pulses having the period indicated by the parameter, a noise source (36) for generating pulses   and selection devices (31, 37) make it possible   to select the output signal   from the source (35) of fundamental period sounds or the output signal from the noise source (36) depending on whether the language to be synthesized is voiced o no.   3. Sound synthesizer according to one of the claims   1 or 2, characterized by the fact that it includes   an amplitude control device (19) for controlling the amplitude of a signal present at the input or output of the unit (16) forming the synthesizer filter,   by means of an amplitude parameter.   4. Sound synthesizer according to any one of   Claims 1 to 3, characterized in that the circuit * 2466826   filtering device (57-61, 65-69) is constituted by a first delay line (51) for delaying the input signal by a unit time interval, a first adder (94), to which sent the delayed output signal and the output signal from the unit (16)   constituting the synthesizer filter, a second line   delay (95) for delaying the output signal from the first adder by a unit time interval, a multiplier (53) for multiplying the output signal of the first adder by the coefficient, a second   adder (52) for adding together the   output signal provided by the multiplication, the output signal provided by the second delay line and the signal   input of the second-order filter circuit (57-61, 65-   69) to provide the output signal of the filter circuit. second order.   5. Sound synthesizer according to any one of   Claims 1 to 3, characterized in that the circuit   second-order filtering (57-61, 65-69) consists of   a first delay line (95) for delaying   unit time the input signal of the circuit   second-order filtering film, a multiplier (53) for multiplying the input signal of the second-order filtering circuit by a coefficient of the latter, a first adder for adding the signal   output provided by the multiplication and the output signal.   from the first delay line (95), a second   delay line signal (51) for delaying the sum output signal during a unit time interval, and a second adder (94) for adding the output signal from the second delay line and the input signal second order filter circuit so as to provide the output signal of the second order filter. 6. Sound synthesizer according to any one of   claims 1 to 5, characterized in that the cir-   second-order filtering (57-61, 65-69) is   second-order filter, that several second-order filters of this type, having different coefficients, are cascaded to form the cascade device, and that two filters of the second-order filter are   second order connected in cascade and presenting coefficients   These two different circuits constitute the two reaction circuits.   7. Sound synthesizer according to one of claims 4 or 5, characterized in that the circuit of   Second-order filtering consists of a digital filter   second-order, that the second-order digital filter is used in a multiplex operation by a time-sharing system, which operates the filter several times over a unit time interval   and that modifies the filter coefficient during each function   tioning. 8. Sound synthesizer according to any one of   Claims 1 to 5, characterized in that the circuit   the filtering circuit and the cascade operation circuit are constituted by operational devices making it possible to   perform filter processing by interpretation and execution program.   9. Sound synthesizer according to any one of   Claims 1 to 8, characterized in that it comprises   in addition a device (13) for transforming the parameters,   to obtain the control parameters by transmitting them   forming their cosines for the purpose of controlling the   of the unit constituting the synthesizer filter.   10. Sound synthesizer according to any one of   Claims 1 to 9, characterized in that it comprises   furthermore, an interpolation device (14)   terpolate the command parameters and send them to   the unit (16) forming the synthesizer filter.   Sound synthesizer according to claim 9,   characterized in that it further comprises a device   interpolator (14) for interpolating the parameters   very representative of the characteristics of the unit (16)   the synthesizer filter, and that the interpolated parameters are submitted by the device (13) for transforming the   parameters, to a transformation providing theircosinus.   12. Sound synthesizer according to one of the claims   10 or 11, characterized in that the interpolation period in the interpolation device is equal to the sampling period of an original sound signal or twice that period.   13. Sound synthesizer according to any one of   claims 11 to 12, characterized by the fact that   interpolation system is used on a functioning basis.   multiplex for the interpolation of a parameter of am-   plitude. 14. Sound synthesizer, characterized in that it comprises a source (15) of sound signals intended to produce a sound signal, a source (12) of LSP parameters intended to produce the LSP parameters, a device (13) for parameter transformation for transforming the LSP parameters into control parameters of a type different from the LSP parameters, and a filter unit (16)   synthesizer fed by the sound signal and whose   The characteristics are controlled by the control parameters of processed.   15. Sound synthesizer according to claim 14, characterized in that the source of sound signals is   constituted by a source (25) of period-fundamental signals   mental, controlled by a fundamental period parameter   to produce a pulse or group of pulses possessed   in the period indicated by the parameter, a noise source (36) for generating random pulses, and selection devices (31, 37) for selectively picking up the output signal from the sound signal source of the fundamental period or the output signal from the noise source, depending on whether the sound   to be synthesized is a voiced sound or not.   16. Sound synthesizer according to one of the claims   14 or 15, characterized in that it further comprises an amplitude control device (19) for controlling the amplitude of a signal present on the input or output side of the unit (16). ) of the synthesizer filter of   sounds, using an amplitude parameter.   17. Sound synthesizer according to any one of   claims 14 to 16, characterized in that the   positive (13) parameter transformation is a   used to transform the LSP parameters to provide the forecast coefficients and the filter unit (16)   Sound synthesizer is a cyclic digital filter.   18. Sound synthesis method, wherein sound signal parameters representing a sound signal   and control parameters provided for the ordering of   characteristics of the filter device are sent to an information carrier, and according to which the sound signal is   produced in accordance with the parameters of the sound   from the information medium and according to which the sound signal is sent to the filtering device, while the characteristics of the latter are regulated by the control parameters from the information medium, which results in the production of a synthesized sound signal,   characterized by the fact that a spectral envelope of the sound   origin is approximated by means of the function   transfer ratio H (Z) of said filtering device, expressed by: H (Z) = a (Z- = p = p1 + a1Z + 2Z + ... pZ in which Z ej LA in which Z = e, o is a constant, w is a normalized angular frequency equal to 2wfAT, AT is a sampling frequency, f is a frequency, p is a   degree of analysis and ai (i = 1, 2, ... p) represents coef-   ficients of prediction, the term Ap (Z) being further expressed in the form of a sum of two polynomials P (Z) and Q (Z) by: A (Z) = {P (Z) + Q (Z)}   P: o: P (Z) = Ap (Z) - Z.ZPAp (Z-1) Q (Z) = Ap (Z) + Z.ZPAp (Z1) pp these polynomials being each put in the form of factors ,   while the angular frequencies, which cancel the poly-   nomes, are used to form said control parameters.