DE2227476C2 - Method and arrangement for generating quasi-synthetic speech in a vocoder - Google Patents

Description

The invention relates to a method for generating quasi-synthetic speech and arrangements for performing the method in a vocoder with several parallel channels of parameter signals, Containing information about the envelope signals of the individual channels, the basic speech frequency and their amplitude, as well as the voiced or unvoiced sounds in which at voiced sounds are the pulses generated in a fundamental speech frequency generator and unvoiced sounds If the pulses generated in a noise generator are fed to channel synthesis units, in which these pulses are modulated with the envelope signal of the respective channel and finally added and in which either the modulated or the unmodulated pulses pass through bandpasses.

The electronic synthesis of human speech is beneficial for speech output from data processing systems and also for the purposes of transmission technology an ever greater importance. There are already a number of processes and also of corresponding arrangements for the generation of quasi-synthetic Language became known. The generation of quasi-synthetic language is used in this Context understood a structure of language from the basic language frequency.

From K. Steinbucb in »Taschenbuch der News processing ", Springer Verlag Berlin, 1962, pp. 851 ff., Along with a number of others under the Designation »Vocoder« of devices for the generation of quasi-synthetic speech that have become known also described the channel vocoder. At one of the

3 4

Like channel vocoder, the speech signal is resolved on the basis that the duration of the impulses is modulated. The invention is based on the knowledge, bandpass filters with consecutive passages that, by switching to pulse duration modulation, leave areas in a series of spectral channels, the application of digital technology, in particular the sets and the envelope curve of this Spectral channels established - 5 binary technology and thus the use of integrated. Furthermore, digital circuitry is possible through a frequency meter. As a result, the level of the basic speech frequency, also known under the term “pitch frequency when the method is carried out with regard to the sequence”, is determined. lent the flow sensitivity and the stability be-The determined data are with the help of one in the special advantages. Particularly advantageous in the case of multiplexing equipment known to transmission technology in the order to carry out the method according to the transmission channel or a storage device of the invention, the small space requirement and the device supplied. Significantly improved testability on the reception or output. In addition, on one side, i.e. in the synthesis part of the vocoder, the arrangements for so-called MSI separation of the individual parameter signals or LSI modules, which take up an even lower level of the fundamental frequency and the decision 15 space, can easily be integrated,
with regard to the voiced or unvoiced character A suitable arrangement for performing ters of the sound as well as with regard to the amplitudes of the method according to the invention contains a single channel, in the literature the tone-speech fundamental frequency generator, a noise level signal occasionally also be as a voiced tone signal - rator, a clock generator, bandpass filters, an adder is drawn. In a switching and control stage, 20 circuitry and channel synthesis units, each with one nadidem, whether it is a voiceless or buffer, and one code converter each, with a voiced sound, either a noise in each channel synthesis unit the code converter generator or a basic speech frequency gene - a counter with a preset follows and the raun serving pulse generator is switched on. The individual counter mutes are connected to the inputs of an emitted noise pulse or the Sprachgrund- 25 NAND gates, from whose output five-sequence pulses are then provided to the individual and a return to the counter is provided and new spectral channels are distributed through to a certain- also the duration-modulated Channel pulse taken from the frequency range, tuned bandpass passes, is bar.

niü the information of the respective envelope curve ampli- A significant simplification and reduction modulated by tudence and, after adding them, form the 30 of the effort results from the summary SMKhetic speech signal. It is also common to centralize from parts of channel synthesis units dit-es to modify the method so that the amplitude units. The application of a central code modulation takes place before the signals the bandpass converter has a special effect on vocoders run through. high number of channels and results in an arrangement

By dividing the voice band into a 35 voltage for carrying out the procedure, the one A series of around 10 to 16 spectral channels result in a fundamental speech frequency generator, a noise genes itself a number of functional groups for each rator, a clock generator, bandpass filters, an adder channel the synthesis side of the vocoder separate upstream and several channel synthesis units with are available and as “channel synthesis units” each contains a buffer in which the be drawn. 40 channel synthesis units on the buffer

A channel synthesis unit has a number of a simplified counter with a preset

Individual functions. First, the digital information follows, in which the individual counter stages with the

mation of the channel is stored for one sampling cycle. Inputs of a NAND gate are connected,

The digital signal is then converted, with its output being fed back to the

into an analog signal, for example a current. 45 counter is provided and from which the continuous modul

This analog current then modulates which can be taken from the lated channel pulse, in which, depending on

Noise-Cinerator originating Raasch impulses or from a voiced-voiceless signal.

the central code converter originating from the basic speech frequency generator on the one hand either with a

the basic speech frequency impulses. central pulse generator via a frequency divider

With the arrangements according to the state of 5 ° or directly and on the other hand, either with one

In all cases, technology is an arrangement or basic speech frequency generator

voltages in which the fundamental frequency pulses or noise generator is connected and in which the

Noise pulses experience an amplitude modulation. Output of the central code converter via a gate

This results in a number of disadvantages, since it is connected to the counting pulse input of the counter Use of amplitude-modulated signals is limited.

Special requirements for the accuracy and the A further simplification results in a

Constancy of the gain of the corresponding device arrangement according to the invention, which a speech

are to be made, the further requirements with regard to the fundamental frequency generator. a noise generator,

Temperature stability, precise comparison and testing a clock generator, bandpasses, an adder In the form of an exact measurement. 60 processing and several channel synthesis units, each with

The invention has for its object to contain a buffer memory, characterized in that in the Method and for performing the method channel synthesis units with the buffer to develop suitable arrangements in which a switching arrangement acting as a selector fails special requirements for accuracy is bound, which is also via separate lines and the constancy of the amplification of the devices to create the 65 from a central slop pulse generator which are also easy to test and are supplied to pulse trains generated that by which take up as little space as possible. the selector of one of the lines of the stop pulse

According to the invention, the object is thereby connected to the generator with a flip-flop stage,

5 6

that this flip-flop stage is also used as a start with a switching arrangement which, from central pulses, determines the pulses of different duration generated by the fundamental speech frequency generator or selects the pulses generated by the noise generator, and
F i g selected by the voiced-unvoiced signal. 6 is a diagram of these pulses.
Pulses are supplied, whereby the temporal 5 The channel synthesis interval shown in FIG. 1 between the start pulses and the stop consists of a buffer 1, which doubles the pulse series from row to row, and channel information C, B, A for at least one digitally modulated channel pulse can be taken from the output of the flip-flop stage of the 20 millisecond sampling cycle. In a further development of this arrangement that stores and when a transfer pulse occurs, the pulses io UP at input 11 are generated directly from stored information to a code converter 2 in a central generator, labeled C *, B *, A * with the various required pulse widths. The result is that the 3-bit code is passed on into a 7-bit code, an arrangement in which the channel synthesis converts to a buffer with an external compression of the 7- Bit code is connected in a 3-bit selector operating circuit arrangement, 15 code can be canceled again; a counter 4 with one in which a central generator is provided, the presetting 3 and a NAND gate S. Generated by several rows of pulses, the one occurring at 12 start pulse Sl is doubled be-duration of the pulses from row to row has the effect that by default 3 those and each of these pulse series are set to 0 via a separate counter level for which the code output <3 {'es generator is available for which 20 converters a 1 is offered. At the output 14 of these outputs connected to the selector, a NAND gate S then also appears and at the output of the selector the continuously modulo logical 1, which on the one hand represents a pulse PDM of the channel pulse directly or via an adapter stage for the basic speech frequency and can be taken accordingly. This arrangement results in the band filter, if necessary via an amplifier, in addition to the easy testing and the clear-cut stage, but on the other hand a particularly large saving in space. Gate input 13 for the counting pulses ZI back-

Is carried out on the basis of the execution shown in the drawing. The counting process is interrupted

The invention will be implemented in the following as soon as all counter stages are at logic 1: n.

explained in more detail. In the drawing shows the start-up supplied at 12 of the presetting

1 shows a channel s \ nthesceunit with a forward 3 ° impulse5 / are corresponding to the position of one of

set counter and a 3-bit 7-bit code switch controlled by a pitch signal

Ä ^ uiiHcr according to the invention, pulses either from a noise generator or

Fig. 2 shows an arrangement according to the invention, which is a known basic speech frequency gene-

a preset counter, a central encoder. The channel pulse fed to the filters is

contains transducers and pulse generators, 35 both with the previously used pulse amplitude n-

F i g. 3 a channel synthesis unit with a switching modulation as well as the modulation of the pulse arrangement, selects the stop pulses, according to the invention, a narrow rectangular

Fig. 4 is a diagram of in the arrangement after pulse, the maximum duration of which by a series of

Fig. 3 occurring pulses, considerations to be determined. The expression for

Fig. 5 shows a channel synthesis unit according to the invention- 4 ° the Fourier series of a rectangular pulse reads:

/ (ω _ο ί) = UK _V

sin n Ky · -ι

cos η

where K _{v = TJT 0} = Γ _ν · / _{0 is} the duty cycle and η = /// „(/ ₀ = basic language frequency , T _y - pulse I

broad). |

sin π -K _x .-. η sin T _x . · F -.- i
π · _Χ ν · .τ ⁼ T _x. 1 x '';

For frequencies at which this expression can be used to counterbalance the counting frequency f _z ~ / ₀ . With a relative

deviates sufficiently little from I, the harmonic 55 small range of the basic speech frequency is this

Just spectrum. The pulse duration can therefore lead to a readjustment of the power without major disadvantages

did not exceed the correct value. This value is negligible.

is made up of the highest amount of Fre- In contrast, it eats in most cases

quenzfo. If one tolerates a decrease in intensity necessary for the synthesis of the voiceless speech

the highest harmonic by the amount a _m = 1.75 db, 60 sections the power of the noise impulses to-

so the pulse width may match a maximum of T _v = 64 jis for.

f _k = 5 kHz. It is

The frequency of the counting pulse is at the

voiced speech sections and the lowest j ^ ._ l // o

Basic speech frequency 1 MHz. ^{The 65 r Ä} to be transmitted ^{: 11} * j / f _e
Power increases proportionally to the square root

the basic speech frequency and is by a _{Ver / f} the clock frequency of the noise pulse is the approx

ringerang of the pulse width 7 \. by increasing twice as high as the highest ear trac-

ι frequency must be. Are shown in Fig. 4 at a basic speech frequency. The stop pulses Pl cnt-

j Z ₀ = 100 Hz and a highest transmission frequency in the circuit according to FIG.

quenz / _f = 10 kHz results from the central code converter emitted pulses; you are

in time according to the selected compander The characteristic curve is staggered and is represented by a voice

γ _τ ~~ ^v fundamental frequency or a noise pulse started.

5 The switching arrangement 6 selects according to the

Information retrieved from the buffer memory one of the lines 61 to 68 and thus one of the stop pulse

In this case, it is also advantageous to line up an io and switch the stop pulse via the counting pulse generator with a pulse frequency from output 69 to the flip-flop stage 7, which can be used for the unvoiced sounds over about 5 MHz. Line 71 also carries the start impulses SI zugeder for the voiced sounds with input 13. The channel synthsceinhcit according to FIG. 3 of the counter for the counting pulses via an in the Vcr- thus works in the star-stop mode. The frequency divider, which divides in a ratio of 5: 1, is connected. 15 period modulated channel impulse PDM is the In FIG. 2, apart from the actual channel flip-flop stage are presented to a Filteranpaßstufe or directly unit further functional groups. routed to a bandpass filter.

A buffer 1 directly controls a binary die in FIG. 5 shown circuit of a channel

Counter 41 with a presetting 31. The counter synthesis unit consists of the intermediate memory 1. is only three-stage, since a central Codewandlcr 22 20 one controlled by this, is also provided as a selector, from which the counting pulses Zl via the operated switching arrangement 6 and an amplifier -Gate 13 get to counter 41. The code converter stage 92. In contrast to the circuit according to FIG. 3 generated from the periodic, from a 5 MHz, are generated in the circuit according to FIG. 5 does not generate the generator 20 via a 5: 1 divider 21 and the stop pulses for a flip-flop stage. Switch 51 to the input of the code converter gc-25 but the pulses shown in FIG

j reached 1 MHz pulses or directly from the 5 MHz Ti ... T 7 with different pulse widths, see above

! Impulses a series of z. B. seven impulses, whose that centrally all required impulse widths for

S The distance to the start impulse doubles in each case. Stand by luck. From the switching arrangement 6, for the

j start impulses SI act in the same way as the one in the circuit from the speech

pulse generator 24 or the noise generator 25 30 of FIG. 3 can be used, cntspre-

j via switch .V2 to the central code converter 22 in response to the information provided by the buffer

received impulses. The switches .S 1 and S1 are mations C *. B *. A * which modulated in its duration

j by the voiced / unvoiced signal TS via the channel pulse PDM directly via the output 69

Line 23 controlled. The continuously modulated channel is switched to the amplifier stage 92, which causes the pulse PDM to be matched to the bandpass filter via an amplifier stage 91,
routed to the bandpass filter in question. The method according to the invention is also

In FIG. 3, a channel synthesis unit is shown as a decoding method in PCM systems, in particular the invention, which arspcichcr from the intermediate in those with very low bit rates arspcichcr 1 for the information C, β, A, one as off - Can be used, for example, in the pulse delta modulation system selector operated switching arrangement 6 and a 40 men. In such a case, the flip-flop stage 7 occurs. The switching arrangement 6 digit of the pitch pulse of the sampling pulse is a commercially available data PCM or pulse delta module system.
dialing sound module, which is also known under the designations, the method according to the invention is

SN 74151 N and FLY 121 are available in an integrated form, which can be combined with a pulse ampli. Over 45 tuden modulation can be used on this module. It then becomes ζ. Β separated lines that applied from a central generator the supply voltage of the amplifier stage 92 in generated stop pulse series Pl to PS , the Fi g. 5 changed in amplitude.

1 sheet of drawings

Claims (5)

Patent claims: L method for generating quasi-synthetic Speech in a vocoder with several parallel Channels from parameter signals, containing information about the envelope signals of the individual channels, the basic speech frequency and its amplitude as well as the voiced or voiceless sounds, in which with voiced "> These are the impulses generated in a fundamental speech frequency generator and in the case of unvoiced sounds the pulses generated in a noise generator are fed to channel synthesis units, in which these pulses are modulated with the envelope signal of the respective channel and finally are added, and either the modulated or the unmodulated pulses bandpasses run through, characterized in that the duration of the pulses is modulated. 2. Arrangement for performing the method according to claim 1, containing a basic speech frequency generator, a noise generator, a clock generator, bandpass filters, an adding circuit and several channel synthesis units, each with a buffer and a code converter, characterized in that in the channel -Synthesis unit on the code converter (2) is followed by a counter (4) with a preset (3) and that the individual counter stages are connected to the inputs of a NAND gate (S), from whose output a return to the counter (4) is provided and from which the duration-modulated channel pulse (PDM) can be taken (Fig. 1). 3. Arrangement for carrying out the method according to claim 1, containing a Sprachgrundfiequenz generator, a noise generator, a clock generator, bandpass filters, an adding circuit and several channel synthesis units each with a buffer, characterized in that in the channel synthesis unit the buffer (1) is followed by a simplified counter (41) with a presetting (31) that the individual counter stages are connected to the inputs of a NAND gate (51), from the output of which a return to the counter (41) is provided and from which the continuously modulated channel pulse (PDM) can be taken that depending on a voiced-voiceless signal (TS) a central code converter (22) on the one hand with a central pulse generator (20) either via a frequency divider (21) or else directly and on the other hand either with a basic speech frequency generator (24) or a noise generator (25) and that the output of the central Codewan dlers (22) is connected to the counting pulse input of the counter (41) via a gate (13) (Fig. 2). 4. Arrangement for carrying out the method according to claim 1, containing a basic speech frequency generator, a noise generator, a clock generator, bandpass filters, an adder circuit with several channel synthesis units and a buffer, characterized in that the channel synthesis units Intermediate memory (1) is connected to a switching arrangement (6) acting as a selector, which is also supplied via separate lines (61 ... 68, the pulse series (P 1 ... P8) generated by a central stop pulse generator, that through the Select one of the lines of the stop pulse generator with a flip-flop stage (7) is connected, that this flip-flop stage also still as start pulses (Si generated by the basic speech frequency generator or the noise generator unc through the voiced / unvoiced signal selected pulses are supplied, with the time lent distance between the generated by the noise or the fundamental speech frequency generator th impulses acting as start impulses and the stop impulse series doubled from row to row and that at the output of the flip-flop stage the duration modulus channel pulse (PDM) can be seen (Fig. 3 and 4). 5. Arrangement according to claim 4, characterized in that a buffer (1) is connected to a switching arrangement (6) acting as a selector in the channel synthesis units. that a central generator is provided, which generates several rows of pulses (Tl ... TT) , the duration of these pulses doubling from row to row, so that one pulse train is available via a separate output of the generator, that these outputs are connected to the selector and the continuously modulated channel pulse (PDM) can be taken from the selector output directly or via an adapter (92) (FIGS. 5 and 6).