DE2214521A1 - Speech synthesis circuit - Google Patents

Abstract

In a synthesiser which, for each sampling period, reconstitutes a language element by means of three sinusoidal components obtained with the help of variable-frequency generators and variable-attenuators, those components are simultaneously subject to predetermined rephasing operations carried out at an auxiliary frequency identified with the pitch frequency (or frequency of vibration of the voice) at the time of emission of vowels or voiced consonants. This auxiliary frequency is delivered by a further variable-frequency generator. In addition, the signal representing the sum of these components is amplitude-modulated by a modulating signal at the auxiliary frequency.

Description

THOMSON - CSP
173 “Vol. Haussmann
Paris 8e /! France

Our reference: T 1160

Speech synthesis circuit

The invention relates to a speech synthesis circuit that receives digital information which is with a following period T corresponding to a sampling period T expresses information that is the following main information and makes it possible to restore a speech element approximately by that for each sampling period a certain number ρ of sinusoidal signals are added, the frequency and amplitude of which is the main information and which are called "main components" here, where ρ is a fixed number η or less than this is. The main type of information given above is given in PR-PS 2 044 290. In short is a main component a temporal segment of a formant, where a formant is a temporal sequence of Spectral components of the same or neighboring frequencies

Lei / Gl

209 8 41/081 2

_ ₂ _ '22H521

is defined, which corresponds to an (absolute or relative) energy maximum in the speech spectrum. the Formant sections for each sample of speech transfer v / earth, v / earth determined on the basis of criteria, as set out in the aforementioned patent. The sound is analyzed with Using a filter battery, and the frequencies of the main components are only approximate transfer.

The aim of the invention is to provide a speech synthesis circuit of this kind, in which it is possible by using auxiliary information, the speaking Identify person to some extent.

This auxiliary information is the information called "pitch information" in Anglo-Saxon literature, which is called pitch information here. This information is during the delivery of vowels and voiced consonants ("voiced" in Anglo-Saxon literature) formed by a frequency which is the frequency of vibration of the speaking person's vocal cords and for short is called "pitch frequency" target; it is generally between 80 and 350 Hz, relating to the vowels and voiced consonants Spectral components are harmonics of this pitch frequency, as is the case with mute consonants ("unvoiced" in Anglo-Saxon literature) is not the case.

Various arrangements for the measurement of the pitch frequency on the transmitter side have already been described. A corresponding bibliography can be found in the book "Speech Analysis, Synthesis and Perception" by

209841/0812

J.L. Flanagan, Springer-Verlag, Berlin-Heidelberg-New York, 1965.

The simplest arrangements of this kind work with peak value detectors: the speech signal becomes one Band filter or, for the sake of greater security, two band filters, which normally produce a signal that is delivered when vowels or voiced consonants are emitted with the pitch frequency is amplitude modulated. Each of these band filters is followed by an amplitude detector, which in turn is followed by a peak value detector is connected downstream, where the frequency of the peak values corresponds to the pitch frequency, when this appears in the signal; otherwise the measuring arrangement delivers a very strongly fluctuating one Output frequency.

In numerous systems for digital voice transmission, the frequency supplied by the measuring arrangement, the is called the auxiliary frequency here, continuously on an auxiliary channel transmitted regardless of whether they are with the Stiramhöhe frequency is identical or not; However, one also transmits a special signal, the formation of which is one requires complicated material and that indicates whether the auxiliary frequency is the pitch frequency. if If this is the case, the speech synthesis takes place with the help of a generator that generates harmonics of this frequency, which is obviously only restored down to the exact quantizations; otherwise it becomes the auxiliary frequency not used, and the synthesis is done with the help of a noise generator.

In the speech synthesis circuit described in the aforementioned French patent, in which

2098A1 / 0812

22H521

the synthesis according to a completely different principle with the help of a small number of "main components" the pitch information is not used.

The invention enables the pitch information to be used by means of simultaneous resetting on predetermined phases for all main components, these provisions on predetermined phases be done with the pitch frequency; this makes it possible to use the sum signal of the main components to make this frequency periodic.

According to the invention is a speech synthesis circuit for synthetic speech generation based on a periodic Information that contains main information relating to a language element and through which Frequencies and amplitudes of ρ is formed by sinusoidal principal components, where ρ is a variable Is a number that is at most equal to a fixed number η> 1, with η frequency-variable generators and η amplitude adjustment arrangements, which are each assigned to one of the η generators, the η generators and Amplitude adjusting arrangements controlled by the main information for the restoration of the main components and with an adder circuit which is the sum of the restored principal components thereby characterized in that for the use of auxiliary information, which is formed by an auxiliary frequency that occurs during the transmission of vowels and voiced consonants the vibration frequency of the vocal cords of the speaking person ("Btimmhöhenfrequenz") is an arrangement for Resetting of each of the restored main components to a predetermined phase is provided, wherein

2098A 1/0812

the provisions on the predetermined phases for the Main components take place simultaneously and from the Auxiliary information can be controlled so that it is compatible with the Auxiliary frequency take place at least when the auxiliary frequency is the pitch frequency.

An embodiment of the invention is shown in the drawing, the only figure of which is the block diagram shows a preferred embodiment of a speech synthesis circuit according to the invention.

The drawing shows the information source 1, which is used for Feeding the speech synthesis circuit is intended.

This information source can be, for example, the last stage of a receiving arrangement, which is for the Speech synthesis attitude necessary information in In the form of parallel binary signals. Each of the successive pieces of information is at the outputs of the information source 1 is maintained for a period T which is equal to the analysis period.

It is assumed here that the maximum number η of main components is three.

Information source 1 has seven multiple outputs (each wire corresponds to a binary digit), namely the outputs 11, 12 and 13, which supply the frequencies F .., Pp and F-Z of the main components, the outputs 21, 22, and 23, which have the corresponding aplitudes A ^, Ap or A ^ deliver, and finally the output 10, the supplies the auxiliary frequency f. "As will be seen, it is superfluous to supply the device with a special signal, that indicates whether the auxiliary frequency is the vocal pitch frequency during reception or not.

209841/0812

22U521

The outputs 11, 12, 13 are each connected to one of three code converter circuits 31 »32 and 33, respectively, which _{add the following numbers to the digital information representing the frequencies F 1} , Fp, F (which is generally given by the identification number of an analysis filter) assign:

F Fr> J?

^N 1 ⁼ ^ ^{N = N =}

where F is a frequency much higher than that Audio frequency band analyzed by the transmitter (for example to 3500 Hz), while q is a fixed integer, for example on the order of 10.

Three counting variable frequency dividers 41, 42, 43 receive the pulses from a clock 2 with the frequency F. These frequency dividers are each provided with multiple control inputs that are connected to the outputs of the transcoder circuits 31, 32 and 33 are connected, and their outputs are connected to the Frequency control inputs 61, 62 and 63 of three signal generators 51, 52 and 53 respectively, to which they receive pulses of the frequency 2qF., 2qFp or 2qF.

Each of these three generators is designed in a manner known per se so that it is essentially a shift register with q contains stages, the output of which is connected to the input via an inverter, which has a digit "1" converted into a digit "0" and vice versa. The q stages are each provided with auxiliary outputs that are marked with a Circuit are connected which contains a resistor network fed by a voltage source. The shift register is initially filled in a suitable manner, for example with q digits "O". The content

209 8 Λ 1/081 2

22U521

of the shift register changes with each increment pulse, and the position of the digits "1" in the shift register determines the resistances which contribute to the formation of the output signal of the generator. This output signal is thus formed by a staircase voltage. The resistor network is calculated in such a way that for every 2q incremental pulses that are supplied with a fixed subsequent period, the envelope curve of the output signal (except for a direct current component) represents one cycle of a sinusoidal oscillation.

For a periodic sequence of incremental pulses the generator delivers a staircase signal which contains a certain number of cycles, and this staircase signal only needs to be passed through a low-pass filter so that it is converted into a sinusoidal signal while the DC component is normally lost, for example due to the Reinforcements.

Such generators are described in the article "Digital Generation of Low-Frequency Sine Waves" by Anthony C. Davies in IEEE Transactions of Instrumentation and Measurement, Volume IM 18, No. 2, June 1969, Pages 97 to 105 described. Such generators and, in general, all sine wave generators where a Resistance network is used, which is switched by a shift register, are referred to here as "shift register-resistor network - generators" called.

Under these conditions, the circuit diagram shows the Inputs 61, 62 and 63 of the generators 51, 52 and 53 the inputs of the incremental arrangement of the contained therein Shift register.

2098 ^ 1/0812

It is easy to see that under these conditions the three generators each deliver output voltages, their envelopes (except for a direct current component) Sections of sine waves with the frequency P., Fp or F, are.

It is also easy to see that the contents of the shift register at a given point in time are the Phase of the sinusoidal signal is determined at this point in time.

The three generators also have inputs 71, 72 and 73, respectively, which reset all stages of their shift registers enable, this state of the corresponding shift register of phase + 270 ° des corresponds to a sinusoidal signal.

The three generators each feed one of three variable ones Attenuators 81, 82 and 83, the control inputs with the outputs 21, 22 and 23 of the Information source 1 via one of three digital-to-analog converters 91j 92 or 93 are connected.

The following should be noted here: If for a given sampling period ρ is less than η or even zero (periods of silence) this information can easily be expressed by information of amplitude 0 on the channels which corresponds to the absence of principal components, so that it is meaningless how the corresponding generators behave in the period concerned, as their possible output signals then be blocked by the adjustable attenuators.

The output signals of the attenuators 81, 82 and 83

2 0 9 8 Λ 1/0812

are added in an adding circuit 55. The on Output of the adder 55 appearing signals have due to a coupling capacitance that is at a inserted at any suitable point, its DC components are lost. The output of the adder circuit 55 feeds the carrier frequency input of an amplitude modulator 65 »its output with the input one is the output signal of the speech synthesis circuit representative electroacoustic transducer 85 is connected via a low-pass filter 75.

The auxiliary frequency is used in this preferred embodiment of the speech synthesis circuit not only for the above-mentioned provisions to vorbestimrate phases, but also a (in the modulator 65 successes _m constricting) Araplitudenmodulation of the output signal of the adder circuit 55 'and in such a manner that the instantaneous value of the modulation signal with the frequency f goes through a minimum in the resets to vorbestimrate phases. The overall effect obtained in this way is very satisfactory.

In the specified embodiment, the phase of each main component is reset to the same fixed value, which corresponds to the content 0 in all stages of each shift register.

Finally, in this exemplary embodiment, nothing is changed in the circuits when the auxiliary frequency is not the treble frequency, which means that the aforementioned special signal is not required. The erratic frequencies that successively form the auxiliary frequency for the delivery of silent consonants, lead to phase reset and to an amplitude modulation that varies erratically from

2098A1 / 0812

22U521 - ίο -

change from one synthesis period to the next, and it has been found that this allows for synthesis The hearing effect obtained from silent consonants was of better quality than that obtained when the phase reset and the amplitude modulation for the synthesis of silent consonants can be suppressed. This also simplifies the circuits.

Thus, in the circuit shown, the output 10 of the information source 1 feeds the generation of a Signal of frequency f a circuit of the same type as in terms of frequency for the staircase Signals with the frequencies F .., Fp, F., used will.

This circuit includes a code converter circuit 30, This is followed by an adjustable frequency divider 40 and a generator 50.

However, since the frequency f is below the frequencies of the majority of the main components, the 'supply takes place of pulses to the adjustable frequency divider 40 through a fixed frequency divider 90, which in turn is fed by the clock 2.

The output signal with the frequency f supplied by the generator 50 becomes the modulation input of the modulator 65 supplied. In this example the modulation signal is the one provided by the generator 50 in the form of a staircase Signal that has not been stripped of its direct current component, so that it passes through a minimum of the Value 0 goes when the shift register of generator 50 is not filled with zeros; further is the Modulation depth set so that the modulated

209841/0812

Signal at the same time as the modulation signal to zero will. The output of the modulator 65 then goes through the low pass filter 75, which the output of the modulator 65 smooths and thereby makes the discontinuities disappear, both of the stairs the modulation signal as well as the modulated signal originate.

The zero crossings of the output signal of the generator are determined with the aid of a decoding ^{circuit, which here consists only of a simple II} neither - nor "gate which has two inputs which are connected to the two outermost stages of the shift register in generator 50. It can be namely easily verified that this V _e rschieberegister can simultaneously have a zero in each of these two extreme steps only if all the stages in the state 0 is. the output of gate 35 will be the inputs 71, 72 and 73 of the generators 51, 52 or 53 and thereby causes the resets to the predetermined phase at the desired moments.

A training that improves the restoration of vowels and voiced consonants is that a periodic signal of period 0 = 1 / f is used as the modulation signal, of which one cycle is divided by two sinusoidal half-cycles (except for the direct current components that cause them to Give a minimum value of zero) with a duration of 0/4 for the Increase from zero to maximum and duration 3 0/4 is formed for the relapse from the maximum to zero.

Because of this asymmetry, it is no longer possible to use a generator whose shift register q Includes stages to help a signal of frequency f

209841/0812

22H521

of port switching pulses of the repetition frequency 2 q f.

However, one can use a similar generator, for example, whose shift register 2 q Contains stages and is initially filled with a digit "1" and (2q-1) digits »0», the loop does not contain an inverter between the input and the output of the shift register.

A network of 2q resistors is then used, each of which can optionally be used for one digit of the number "1" is connected to the voltage source, which makes it possible to cycle the staircase signal to give any desired shape.

A given phase of the envelope of the output signal is indicated by being in a given stage the only number "1" circulating in the shift register is determined, and the output of this stage can directly control the provisions to the predetermined phase.

Another solution is to use the Pig. 1 is retained, and that it port switching pulses are supplied, the frequency of which for the rising sections of the signal (transition of the shift register from the state "louder 0" to the state "louder 1") three times greater than for the falling ones Signal sections (return to the "louder 0" state).

2 0 9 8 Λ 1/0812

Claims (3)

Claims 1J Speech synthesis circuit for synthetic speech generation on the basis of periodic information that contains main information relating to a speech element and is formed by the frequencies and amplitudes of ρ sinusoidal main components, where ρ is a variable number that is at most equal to a fixed one Number n> 1, with η variable-frequency generators and η amplitude adjustment devices, which are each assigned to one of the η generators, the η generators and amplitude adjustment devices being controlled by the main information for the restoration of the main components, and with an adder circuit, which components Sum of the restored main components, characterized in that for the use of auxiliary information which is formed by an auxiliary frequency which, during the transmission of vowels and voiced consonants, the oscillation frequency of the vocal cords of the corresponding person ("vocal pitch frequencies z ¹¹ ), an arrangement is provided for resetting each of the restored main components to a predetermined phase, the resets to the predetermined phases for the main components taking place simultaneously and being controlled by the auxiliary information so that they take place at the auxiliary frequency at least when the auxiliary frequency is the pitch frequency. 2. Speech synthesis circuit according to claim 1, characterized in that that the resets to predetermined phases with the auxiliary frequency carried out independently whether this is the pitch frequency or not, that an amplitude modulation arrangement for the 209841/0812 Amplitude modulation of the sum signal of the main components is provided by a modulation signal at the auxiliary frequency, and that the resets then take place on predetermined phases when the instantaneous value of the modulation signal has passed a minimum goes. 3. Speech synthesis circuit according to claim 2, characterized by a variable frequency auxiliary generator that is controlled by the auxiliary information so that it is a Signal with the auxiliary frequency supplies, and that each of the η generators and the auxiliary generator is a shift register resistor network generator is, wherein the shift register of each of the η generators has a phase control input that the resetting of the Shift register in a predetermined state enables the phase reset arrangement by the Auxiliary generator and by a decoding arrangement, which then supplies a control signal when the shift register of the auxiliary generator is in a predetermined state, is formed that the output of the Decoding arrangement is connected to the phase control inputs, and that the amplitude modulation arrangement by the auxiliary generator and by an amplitude modulator, whose carrier input with the output the adding circuit and its modulation input are connected to the output of the auxiliary generator is. 209841/0812