DE1282098B - Method and arrangement for eliminating the fundamental frequency components in the spectrum channels of a channel vocoder - Google Patents

Description

FEDERAL REPUBLIC OF GERMANY

GERMAN

PATENT OFFICE

EDITORIAL

Int. CL:

GlOl

H04b

German class: 21 a2- 36/22

Number: 1 282 098

File number: P 12 82 098.3-31 (J 30613)

Filing date: April 14, 1966

Lay-out day: November 7, 1968

The invention relates to methods and arrangements for eliminating the influence of the fundamental frequency the results of the short-term spectral analysis of speech signals in a channel vocoder

In general, a vocoder has many, relatively narrow channels. The ratio of two neighboring ones Channel voltages to each other strongly depend on the level of the fundamental frequency, since a channel in whose pass band one or more harmonics of the fundamental oscillation fall, a larger voltage supplies as a channel in whose pass band no or fewer harmonics fall. Since this influence of the Basic speech frequency has an adverse effect on the short-term spectrum in the later speech synthesis it is advantageous to free the short-term spectrum from the basic speech frequency components.

In this context, it has already been proposed that the excitation channel of a channel vocoder obtained signals to describe the speech fundamental wave for a compensation of the fundamental wave influence to be used in the spectrum channels. Here, the known arrangement of the Channel vocoders so that in each spectrum channel adjustable attenuators are switched on, their degree of attenuation depending on one of the Basic speech frequency derived signal is controlled. This facility allows the ratio ever to make two channel voltages independent of the fundamental frequency without making certain requirements must be set to the fundamental frequency range (e.g. narrowing) of the analyzed language. There these facilities cannot be implemented without a certain additional expense, result from this certain disadvantages.

It is therefore the object of the present invention to provide a solution that, with minor restriction the area of application, which consists in the fact that the fundamental speech frequencies are not too high be approved, is advantageous in particular with regard to the effort.

For a method of freeing the spectrum channel voltages of a channel vocoder from the components of the Speech fundamental wave and its harmonics, the invention is that on the one hand the bandwidth of the Channel filter is chosen to be relatively large and, on the other hand, mean values of the channel voltages in the spectrum channels be formed over a variable time interval dependent on the speech stimulus.

In this context it is advantageous to make the time interval selected for the averaging of the channel voltages dependent on the analysis of the speech stimulus in such a way that it is suitable for voiced methods and arrangements for eliminating the
Fundamental frequency components in the spectrum channels
a channel vocoder

Applicant:

International Business Machines Corporation,

Armonk, N.Y. (V.St.A.),:

Representative:

Dipl.-Ing. A. Bittighofer, patent attorney,

7030 Boeblingen, Sindelfinger Str. 49

Named as inventor:
Ernst Rothauser,
Erwin Paulus,
Norbert Teufelhart, Vienna

Claimed priority:

Austria from April 26, 1965 (A 3825/65)

Sound is equal to or proportional to the period of the fundamental speech wave and is constant for unvoiced sounds.

One arrangement for carrying out the method is that for analog representation of the Mean values of the channel voltages in the spectrum analysis per channel the output of the channel filter with is connected to an integrator controllable in its integration time by triggering, the trigger input of which is connected to the excitation channel via a controllable switch that depends on a Voiced-voiceless discriminator the integrators for voiced sounds with the excitation analyzer connects, which determines the basic speech frequency and in each case at points of the same phase position, z. B. Zero crossings in one direction, emits an impulse, or the integrators with unvoiced sounds a pulse generator with constant pulse repetition frequency connects so that the integrators at each Triggering emit a pulse, the amplitude of which is proportional to the integral of the channel filter output voltage is about the time between two consecutive triggers.

The arrangement can be modified in a favorable manner so that for a pulse-excited Channel vocoder the output of an excitation analyzer known per se, which can be used for both voiced and unvoiced sounds at the time

809 630/581

a 4

can score points with the same phase position of the excitation function. So it is possible to adjust the integration intervals Emits pulses via a pulse amplifier with either a constant time or pro trigger inputs of the "integrators."

Another arrangement gives favorable select results.

in that for the digital representation of the Mittelwerfe 5 Another, not inconsiderable .Vorteil of the new of the channel voltages in the spectrum analysis for each method and system is a noticeable one Channel the output of the channel filter with an integra- Reduction of the circuit complexity, tor, which, controlled by a dis- An analyzer of the conventional type contains im criminator, in each case darm at its output a signal generally per channel except for the band filter, the emits, if the time integral formed in it io direct current circuit and the low-pass filter are still equal the envelope of the channel filter output voltage amplifier. For the digital representation of the voltage reaches a predefinable value, whereupon channel voltages are also added per channel the integration starts again at zero, that also includes a PAM modulator and an analog-to-digital converter Output of the integrator connected to a counter is required. The analog-to-digital converter is, however which, on the other hand, has a further input with 15, generally for all Kahale "only once the excitation channel provided via a controllable switch, since mostly a time division multiplex method for is coupled, which is applied depending on a voiced transmission of the channel values. Even Foreheadless discriminator the counter with voiced in the arrangements according to the invention it is in Lutes connects to an excitation analyzer, generally also an analog-to-digital converter which determines the basic speech frequency and gives each 20 to the amplitudes of the output pulses at points with the same phase position, e.g. B. to represent zero crossings digitally. In the new system, however, is one direction, emits an impulse, or which in each case the known group, consisting of low-pass, unvoiced sounds the counter with a pulse DC voltage amplifier, modulator and analog generator with constant pulse repetition frequency. Digital converter, by the integrator group, counter binds, the counter replaces with each pulse at its 25 and memory. The comparison of the two groups Input reset and the contents of a memory show that the circuitry is also difficult to achieve is transmitted, at the outputs of which the value of the realizing DC voltage amplifiers in the new Enveloping is ready. . ... . Systems are omitted, resulting in further advantages

This arrangement can also result in a favorable manner. .

be modified in that for a pulse 30 In the following the invention is based on by excited channel vocoders the output of an exemplary embodiment explained per se drawings known excitation analyzer, which is described for both · ■. It shows .

Voiced as well as unvoiced sounds at the time - Fig. 1 the block diagram of a known channel -

points equal phase position of the excitation function vocoder arrangement,. . . . .

Emits pulses, via a pulse amplifier with the 35 Fig. 2 the block diagram of a channel vocoder inputs the counts is connected. .'.... arrangement according to the invention with triggered

Another arrangement is characterized by 'integrators and a voiced-voiceless-Unterdaß for integration ^ t, variable integration time of the separation in the excitation channel,
triggered integrator per channel from an integration Fig. 3 the block diagram of a channel vocoder

capacitor consists, ~ its "eme electrode via a 40 arrangement with triggered integrators and a first controllable switch with the output of the excitation analyzer ^ without distinguishing between Channel filter and a second controllable voiced and unvoiced speech sections, Switch with one pole of a voltage source Fig. 4 is the block diagram of a channel vocoder

is connected, whereby the second switch via its' arrangement, similar to the arrangement of FIG. 2, control input through the signals of the excitation channel. 45 but with digital. Representation of the envelope that is controlled and its other electrode via a channel filter output voltage,
Resistance with that; the other pole - "of the voltage source_ - Fig. 5 the block diagram of a channel vocoder is connected and the output signal is connected. An arrangement similar to the arrangement according to FIG. 3,

The structure of a different arrangement is, however, with digital representation of the envelope of the characterized that 'for integration with variable 50 channel filter output voltage,

Integration time of the integrator from an integration F i g. 6 the block diagram of a triggered

capacitor consists of one electrode via an integrator for use in facilities according to first controllable switch with the output of FIGS. 2 and 3 and

Channel filter and a second controllable Fig. 7 shows the basic circuit diagram of a triggered

Switch with one pole of a voltage source and 55 integrator for use in facilities according to the other electrode of which with the other pole of FIGS. 4 and 5. '

Voltage source is connected. Furthermore, the analysis and synthesis of speech signals with '

Electrode of the integration capacitor via a channel vocoder to the bandwidth-limited voice connection line with a Schmitt trigger transmission or speech processing is known interconnected, its output with a mono- 60 Hch from an analysis and synthesis of the spectral Stable multivibrator is connected, its output envelope and the analysis and synthesis of the again with the control input of the second switch spectral fine structure. The starting point for the is connected and carries the information signal. Invention is the one shown in FIG. 1 shown known

The advantage of the new method and the associated 'channel vocoder. The speech signal s (O is composed of the spectrum channels with the help of arrangements compared to the conventional 65 of the spectrum analyzer SAn in that the integration intervals for the mean K ₁ to K _n and the fundamental wave analyzer AAn are broken down into value formation of the envelope of the filter output the excitation _{channel jr 0} . The spectrum analysis voltage of a channel vocoder can be freely selected tor SAn consists of a series of 'band filters BP ₁ '

to BP _n , which converts the speech signal into the desired speech signal a series of short pulses. Break down the frequency bands. _{The rectifiers G 1} to G _n and low-pass filters LP ₁ following each band filter after the interval between two successive pulses are controlled by the analyzer. For voiced up to LP _n , channel voltages are generated which depend on the section it corresponds to the period duration, for the fundamental frequency in such a way that in each channel 5 unvoiced sections it fluctuates statistically within a certain mean value of the rectifier limits due to a low-pass filter. The advantage of this type of output voltage generated over a time interval is that a synthesis of mixed forms is possible, the length of which is only possible through the impulse response of the voiced and unvoiced and that the low pass is fixed and therefore independent of the synthesis of voiced sections small. Fundamental frequency is constant. In the fundamental wave analyzer , changes in the period duration, which can occur from period to AAn from a period extending approximately from 0 to 300 Hz, are taken into account. As already mentioned, the output voltage of the excitation function is obtained and a measure for the short-term channel K ₀ is made available via the excitation low-pass filter of a channel. With the help of an energy content of the voice signal within the transmission link Ü, which is carried out in conventional technology 15 averaged over a range of the input filter and, for example, also systems at the time that is determined by the low-pass filter. It can include time-division multiplex transmission, it can be demonstrated that the amount of this short-term spectrum of the signals of the channels K ₀ to K _n from the transmitting end can only be changed slowly over time. When the vocoder is voiced, where the speech analysis takes place, it oscillates with the basic frequency, provided that the receiving side is transmitting on which the speech synthesis 20 is not greater than the cutoff frequency of the. Both low-pass filters are used on the receiving side. This results in a dependence of the short-term power spectrum on the respective synthesizers SSy and ASy assigned to the spectrum channels and also to the excitation channel. The fundamental frequency that the later synthesis of the spectrum channels K ₁ to K _n possess on the received speech signal is particularly disruptive. With regard to, side own modulators M, which in the respective 35 this synthesis is the dependence of the output channel transmitted portion of the spectrum signal with DC voltages of the analyzer on the basis of the fundamental wave signal α (t) of the fundamental wave synthesis frequency by the method explained below modulate tisators ASy. The output signals of all and arrangements eliminated. For this it is necessary that spectrum channels are then not for a set of the amounts of the short-term power spectra; Band filters AP ₁ to BP _n given, at the output 30 of which all fundamental frequencies are averaged over the same time, the synthetic speech signal g (t) arises. but in each case over a time that corresponds to the basic period

As already mentioned, the analysis and the speech signal is proportional, the simplest synthesis of speech signals with the channel vocoder, however, over the basic period itself,
from an analysis and synthesis of the spectral envelope F i g. 2 shows an arrangement for measuring ends and the spectral fine structure. For voiced mean values, for which the integration time T is either sections of the speech signal, the fine structure can be made constant proportional to the basic period of the speech signal or the quasi-periodic character of the speech signal, and the result of the signals is a line structure, for unvoiced sections measurement in form a number of amplitude-modulated, however, the power spectrum is continuous. The pulse trains with Tals distance between each other for the synthesis required information about the 40th following pulses is delivered. The duration of the spectral envelope is constant and much smaller than the SAn in the form of a number of variable equal to smallest possible time T. The impulses come to the. voltages at the spectrum channels K ₁ to K _n outputs of the in FIG. 2 provided triggered. The information about the spectral integrators I ₁ to I _n . At the input α of the Integra fine structure for the synthesis of the excitation function 45 gates are the filter output signals. At the times, K _{0 is} supplied by a special part of the analyzer of the occurrence of the impulses in the excitation channel. It consists of conventional vocoders - the integrators are triggered and emit: systems from a signal that only has two values at a time, an amplitude-modulated pulse.
can take and indicates whether the excitation in the eye- In the arrangement according to F i g. 2, T is voiced or unvoiced for vocal vision, and constant for 50 loose sounds, but voiced excitation additionally from a signal proportional to the fundamental period of the speech signal for voiced sounds; made information about the current base frequency. The time T is controlled by a there. During the synthesis, the information about the voiced-unvoiced distinction is used with the aid of the spectral fine structure in such a way that D% for the discriminator. Depending on the voiced-voiced sections the excitation function a 55 unvoiced differentiation of this discriminator is pulse train with the determined during the analysis of the trigger input b of the integrators is above the control fundamental frequency, while it consists of noise with unvoiced bar switch S for voiced sections of the section. The synthesis of a speech signal via contact c to the excitation speech signal with the channel vocoder can be switched on as an analyzer AAn and for voiceless speech multiplication of the short-term power spectrum of the 6o _: sections via contact d to a pulse generator. Excitation function with the spectral envelope approximately determined in the spectral analysis IG, which generates pulses with a constant pulse repetition frequency. In the spectrum channels Z ₁ to K _n arise are understood. in this way with amplitude-modulated pulses

The pulse spacing T also exists in the pulse-excited channel vocoder system.

the synthesis from the multiplication of the spectral 65; The arrangement according to FIG. 3 corresponds to im

Envelopes with the short-term power spectrum of the essentials discussed above. Only the one

Excitation function. The excitation function is both in the direction of voiced and unvoiced distinction

for both voiced and unvoiced sections of the by a more general means of analyzing the

ί 282

Suggestion function has been replaced. The output signals of the excitation analyzer AAn are transmitted via a pulse amplifier V via the excitation channel K ₀ to the trigger inputs b of the integrators. In the excitation analyzer of the type used here, the excitation function is a sequence of short pulses for both voiced and unvoiced sections of the speech signal. The interval between two consecutive pulses is controlled by the analyzer. For voiced sections it corresponds to the period, for unvoiced sections it fluctuates statistically within certain limits.

The arrangement according to FIG. 4 is used for the digital representation of the channel filter output voltages of the channel vocoder. In this arrangement, too, mean values of the channel filter output voltage are formed, the integration time of which can be controlled. In this arrangement, each individual channel consists of a band filter BP ₁ to BP _ni, a triggered integrator J ₁ to I _n , a threshold value discriminator D ₁ to D _n , a binary counter .SZ ₁ to BZ _n and a storage device S ₁ to S _n . The output signals supplied by the band filters are first added up in the integrators. If the sum reaches a certain aj threshold value which is determined by the discriminators connected to the integrators I ₁ to I _n via the connection g and which can be selected, then a pulse is emitted at the output of the relevant integrator. The value accumulated in the integrator is deleted and the integration is started again. : -Each pulse emitted at the output «of an integrator is transferred to a binary counter, which counts the pulses. At time intervals of T, the formation of which has already been carried out in connection with the F ί g. 2 and 3, the input / the counter reading is stored in the connected memories S ₁ to S _n and the counter is reset to zero. The stored counter readings are then "a linear measure for the mean values of the channel filter output voltages formed with a variable integration time.

The arrangement according to FIG. 5 differs from the arrangement according to FIG. 4 only in the type of derivation of the variable integration time T. It is the same as has already been explained in connection with FIG.

Assuming that eight levels are sufficient to characterize the mean values, the Binary counter with three digits each. The integrators are like that dimensioned so that they emit a maximum of seven pulses during the time Γ at full control.

One possibility for the circuitry implementation of integrators for use in arrangements according to FIGS. 2 and 3 is in Fi g. 6 shown. From the output of the respective channel filter BPi (i - 1, 2, 3 ... ri) via line a and the resistance R ₁ is transferred to the input of the integrator / {an alternating voltage with slowly varying envelope. The positive half-waves of this alternating voltage cause a current i _{c to flow in transistor 7! S 1} . No collector current flows during the negative half-worlds. The collector current behaves like a one-way rectified current,

To explain the mode of operation of the integrator circuit, it is assumed that the capacitor C is negatively charged as a result of a previous switching operation on the transistor TR _Z. The voltage across the capacitor is - Fs <0. As long as the transistor Ώ? 2 is blocked, the capacitor is discharged by the collector current i _{c of} the transistor TR ₁ . The resistance i? Ä is chosen so that the voltage drop across it is always small compared to the voltage across the capacitor · G. The discharge reduces the voltage on the capacitor continuously. If the transistor TR _{% is} triggered by a positive pulse at the trigger input b of the integrator via the resistor i? _{a is} opened briefly, then the capacitor is simultaneously charged again to the voltage - Vs. The pulse-like charging current causes a pulse on resistor R ₃ , the maximum amplitude of which is proportional to the voltage change on the capacitor. The signal of the short-term mean value of the channel filter output voltage is picked up at the '' electrode E ' and transmitted via the respective channel Ki.

In order to also use the integrator in arrangements according to FIGS. 4 and 5 require additional equipment. These can, as F ί g. 7 shows, consist of the Schmitt trigger ST and the monostable multivibrator SS . The Schmitt trigger ST assumes the role of the discriminator D ₁ , whose breakover threshold can be set so that the trigger tips over when the voltage on the capacitor has reached a certain threshold value. Tilting this Schmitt trigger causes the monostable multivibrator SS to emit a short positive pulse and thereby opens the transistor TR _2. As a result, the capacitor is charged again to the negative voltage - Vs. In addition, the Schmitt trigger tilts back into its starting position. Every time the integration has reached a certain size, the integrator emits an impulse and starts the integration again. The output pulses of the integrator reach the counters via output e, where they are counted in a certain time unit.

Claims (8)

Patent claims: 1. A method for ^{releasing the spectrum channel j} voltages of a channel vocoder from the components of the fundamental speech wave and its harmonics, characterized in that, on the one hand, the bandwidth of the channel filters (BP ₁ to BP _n ) is selected to be relatively large and, on the other hand, in the spectrum _{channels (Jt 1} to K _n ) mean values of the channel voltages are formed over a variable time interval dependent on the speech excitation (FIGS. 2 to 5). 2. Procedure for releasing the spectrum channel tensions a channel vocoder from the components of the speech fundamental wave and its harmonics Claim 1, characterized in that the selected for the averaging of the channel voltages Time interval depends on the analysis of the speech excitation in such a way that it is for voiced Sounds equal or proportional to the period of the fundamental speech wave and constant for unvoiced sounds is. 3. Arrangement for performing the method according to claims 1 and 2, characterized in that that for the analog representation of the mean values of the channel voltages in the spectrum - analyzing each channel (K ₁ to K _n) of the output (α) of the channel filter (BP ₁ to BP _n) nnt a durchTrigge- tion in its integration time controllable integrator (I ₁ to I _n ) is connected, whose trigger input (b) is connected to the excitation channel (K ₀ ) via a controllable switch (S) , which depends on a voiced-unvoiced discriminator S (Di ) connects the integrators for voiced sounds with the excitation analyzer (AAn) (c), which determines the fundamental speech frequency and in each case at points with the same phase position, ζ. Β. Zero crossings in one direction, emits a pulse, or which, in the case of unvoiced sounds, connects the integrators to a pulse generator (IG) with a constant pulse repetition frequency (d), so that the integrators emit a pulse with each triggering, the amplitude of which is proportional to the integral of the envelope of the channel filter output voltage is the time between two consecutive triggers (Fig. 2). 4. Modification of the arrangement according to claim 3, characterized in that for a pulsed ao excited channel vocoder the output of a known excitation analyzer (AAn), which emits pulses for both voiced and unvoiced sounds at the times of the same phase position of the excitation function a pulse amplifier (V) is connected to the trigger inputs (b) of the integrators (I ₁ to I _n ) (FIG. 3). 5. Arrangement for performing the method according to claims 1 and 2, characterized in that for the digital representation of the mean values of the channel voltages in the spectrum analysis per channel (K ₁ to K _n ) of the output (a) of the channel filter (BP ₁ to BP _n ) is connected to an integrator (I ₁ to I _n ) which, controlled by a discriminator (D ₁ to D _n ), emits a signal at its output (e) when the time integral formed in it is above the envelope the channel filter output voltage reaches a predetermined value, after which the integration starts again from zero, in that furthermore the output (e) of the integrator with a counter (BZ ₁ to BZn) connected, on the other hand via a further input (/) with the excitation channel ( K ₀ ) is coupled via a controllable switch (S) which, depending on emem voiced-unvoiced discriminator (Di ), connects the counter for voiced sounds with an excitation analyzer (AAn) (c), which the Sprachgrundf frequency determined and in each case at points with the same phase position, ζ. Β. Zero crossings in one direction, emits a pulse, or which connects the counter with a pulse generator (IG) with a constant pulse repetition frequency for unvoiced sounds (d), whereby the counter is reset with each pulse at its input (f) and its content is transferred to a memory (p ₁ to S _n ) is transmitted, at the outputs of which the value of the envelope is available (FIG. 4). 6. Modification of the arrangement according to claim 5, characterized in that for a pulse-excited channel vocoder the output of a known excitation analyzer (AAn), which emits pulses for both voiced and unvoiced sounds at the times of the same phase position of the excitation function, via a pulse amplifier ( V) is connected to the inputs (/) of the counters (5Z ₁ to BZ _n ) (Fig. 5). 7. Arrangement according to claims 3 and 4, characterized in that the triggered integrator (Ii) per channel (Ki) consists of an integration capacitor (C), one electrode (E) via a first controllable switch (TR ₁ , R ₁ ) is connected to the output (ä) of the channel filter and via a second controllable switch (TR _% , R ^ to one pole (-) of a voltage source (Vs) , the second switch via its control input (b) through the signals of the Excitation channel is controlled, and the other electrode (E ') is connected to the other pole (+) of the voltage source via a resistor (R ₃ ) and carries the output signal (FIG. 6). 8. Arrangement according to claims 5 and 6, characterized in that the triggered integrator (Ii) consists of an integration capacitor (C), one electrode (E) via a first controllable switch (7! R ₁ , R ₁ ) with the output (a) of the channel filter and to one pole via a second controllable switch ₍₂ 77?) (-) of a voltage source (Vs) and the other electrode (e ') to the other pole (+) of the voltage source is connected, and that one electrode (E) is interconnected via a connecting line (g) with a Schmitt trigger (ST) , the output (b) of which is connected to a monostable multivibrator (SS) , the output (e) of which is connected to the control input of the second switch (HR ₂ ) is connected and carries the information signal (Fig. 7). For this purpose 2 sheets of drawings 809 630/581 10.68 © Bundesdruckerei Berlin