SU388369A1 - DEVICE FOR DISCRETE SYNTHESIS OF SPEECH - Google Patents

Description

The invention relates to a technique of synthetic telephony on distant lines of telephone communication. Discrete speech synthesis devices are known that contain a driver, a delay line (digital), multipliers, a low-pass filter, and an amplifier, in which a sequence of short pulses following the pitch frequency is used as a delay line (LZ) excitation signal. speech, and in the synthesis of deaf sounds - a sequence of impulses of crippled noise, previously limited by spectrum. However, in such devices, the synthesized speech signal has insufficient natural sound, especially with low male voices. At the same time, synthesized speech has a characteristic “crackling sound. In addition, excessive underlining of the deaf sounds in the synthesized signal also degrades the natural sound of the synthesized speech. The purpose of the invention is to increase the natural sound of synthesized speech. The goal is achieved by the fact that all taps of the delay line are connected to multipliers through pulse shapers, the second input of the driver through an additional delay unit is connected to the third input of the driver and to the second inputs of the driver, and to the control amplifier. FIG. 1 shows a block diagram of a device for discrete speech synthesis; in fig. 2 shows a single-pulse and four-pulse excitation signal and their phase spectra, as well as signals at the outputs of the pulse shapers; in fig. 3 shows timing diagrams explaining the operation of a device with a four-pulse excitation signal; in fig. 4 - timing diagrams that show the work of the device driver and amplifier. A device for discrete speech synthesis contains a driver with two inputs a and b, delay line 2, pulse shapers 3, multipliers 4, low pass filter 5, control amplifier 6, integrator 7, delay block 8, control signal sources 9. The device works in the following way. The pitch signal (FROM) and the tone-to-sound (T-III) signal control the operation of the exciter / tuned to a quadruple frequency of the speech fundamental tone, at the output of which an excitation signal is generated, representing at a "tone" a sequence of short pulses following 4 / from OT, with "syllable - impulses

clipping noise. A tone pulse or a noise impulse is fed to the input of the delay line 2.

When a device with analog LZ is excited, each excitation pulse with a frequency of 4 / c corresponds to the time response of the orthogonal filters, starting at the moment of the appearance of the excitation pulse. The resulting output signal at each time instant is represented as a sum of the time responses for each excitation pulse, acquiring a continuous character (see Fig. 3). Every fourth pulse and the corresponding time response has a polarity opposite to the previous three (such a response is marked in Hatch 3 in Fig. 3).

Digital LZs advance pulses of one polarity, which are fed to pulse shaper 5 to separate every fourth pulse from a sequence of pulses and invert it.

At one of the pulse shaping outputs, a three-pulse signal is formed (see Fig. 28), on the other, a single-pulse signal of opposite polarity. The outputs of each driver are combined so that a four-pulse signal is formed at the combined output.

Further, in the case of the cosine decomposition used in the device, the signals from the output of the pulse formers, symmetrically located with respect to the central branch of the LZ, are added (in the case of the sinus decomposition it is necessary to apply the operation of subtraction) the amplitudes of the harmonics of the expansion of the envelope of the original spectrum into the cosine series obtained in the analyzer. The sum signal from the multiplier outputs is passed through a low pass filter 5 with a cutoff frequency F equal to the upper limit frequency of the synthesized range to filter out synthesis products outside the subband.

The signal from the low-pass filter is fed to the input of the regulating amplifier 6, in which the level of muffled sounds (with "noise") is reduced to the level of sonorous sounds (with "tone") and the synthesized signal is amplified to the level necessary for the telephone to work.

The need to reduce the level of deaf sounds is explained by the fact that, when synthesizing these sounds, a pulsed random sequence is used as the excitation signal of the LZ, for example, pulses of clipped noise previously limited in frequency by the spectrum. At the same time, the synthesized noise signal has more power than the tone signal, especially at low OT frequencies, since the frequency of the appearance of noise pulses is much higher. This leads to the appearance of clicks when the tone-to-noise signal is switched, moreover, it is perceived by the ear as excessive underlining of the deaf sounds, which degrades the natural sound of the synthesized speech. Indeed, in LZ, the power of the signals taken from the taps is determined only by the pulse frequency, since their amplitude and duration are fixed. When the tone at the FRT frequency, for example, is 75 Hz, the frequency of the exciter pulses is 300 Hz, while the average frequency of the appearance of noise pulses is equal — and at F 2000 Hz

is 1000 Hz. Therefore, there is a need to reduce the path gain during the formation of deaf sounds. This adjustment is carried out as follows. In the transition from "tone to" noise, the gain of the amplifier (see Fig. 46) decreases

the tone-to-noise signal (see fig. 4a), which is fed to the control amplifier through the integrator 7. The integrator time constant is chosen so that the duration of the transient processes at its output corresponds approximately to the total signal delay time Gts. The inclusion of noise pulses at LZ occurs with the same delay, so that by the time of the appearance of noise pulses LZ is completely cleared of

OT pulses, and the path transfer coefficient already has a steady-state lower value, which eliminates the appearance of clicks in the synthesized signal. The required delay is provided by a delay unit 8, made, for example, in the form of a standby multivibrator, which generates a pulse with a duration slightly longer than Gls when switching from "tone to" noise (see Fig. 40). The signal from the output of the delayed unit is fed to the input of the exciter to delay the onset of noise pulses (see Fig. 4 g). The delay in switching on the noise pulses makes it possible to get rid of clicks in the transition from tonal to noise sounds.

In the proposed device LZ performed on waiting multivibrators. In addition, the device is built almost exclusively on the elements of discrete technology and can be fully implemented on integrated circuits

or integral nodes in the volume of several cubic decimeters.

Subject invention

A device for discrete speech synthesis, containing a two-input exciter, serially connected to a delay line, symmetrical about the center of the diversion of which is connected to the multipliers, and the multiplier outputs are connected to the control amplifier through a low-pass filter, which is designed to enhance the natural sound synthesized speech, all taps of the delay line are connected to multipliers through pulse shapers, the second input of the driver through an additional delay block is connected to the third mu pathogen entry and to the second inputs of the pulse shapers and via an integrator - to -regulates amplifier.

Jn

imp

Iot

..its

and MP

GP GP m I I 5 5 6 7 8 a U1P2 75 15KUiS 19 K

imp

Shield

c ICT and

I00№ 0 ,,.

III iI -IIb.tr, --- G -...

to

Phi2.1

L .- 1imp

K-number zapnoiiUKu often / pi / OT

/ 4-2 3 5 5 7 8 9 UP 121314 7515 J71819 K

 F-VJ W

 I SHMP

(4imp

 i

.LL

4. - yyyyy

g;

lz

/ gh

j

X

Pu. &. S

Ur-Stimulation Signal

g

 And sj q

I i

eleven (§

7

-6b /) (. One signal

4