JPS5857759B2 - Drive sound source signal generator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION This invention relates to a PARCOR type speech synthesizer (Patent No. 876).
024 "Speech Synthesizer") or LSP type speech synthesizer (Japanese Patent Application No. 54-022318, "Speech Synthesizer"). This relates to a device that generates.

As shown in FIG. 1, the speech synthesis device is usually configured to input the drive sound source signal generated by the drive sound source generator 11 to a speech synthesis filter 12, and obtain a synthesized speech signal from the output thereof. .

In a conventional standard PARCOR type speech synthesizer, a configuration as shown in FIG. 2 was used for the drive sound source signal device 11.

That is, the impulse train generator 13 periodically generates impulses having a pitch period given from the terminal 14.

This impulse may be a constant value sample value which may also be an analog pulse.

Further, the white noise generator 15 is generally provided with an M-sequence generator or the like, and a signal for discriminating between voiced and unvoiced sounds is applied to the changeover switch 16 from the terminal 17. In the case of an unvoiced sound, the output of the white noise generator 15 is selected and supplied to the sound source amplitude multiplier 18 .

A sound source amplitude signal is given to the multiplier 18 from a terminal 19,
This value is multiplied by the input.

With this configuration, an impulse train (when voiced) or white noise (when unvoiced) having a desired amplitude is outputted to the terminal 21 as a driving sound source signal.

For example, as shown in FIG. 2, the synthesis filter 12 in FIG. and is output from the terminal 26.

However, the conventional driving sound source signal generator shown in FIG. 2 has the following drawbacks.

(1) In the case of voiced sounds, an impulse train whose period is the pitch period is generated, but in this case, the energy of the driving sound source signal is temporally concentrated at the time when the impulse is generated, and is zero at other points. .

For this reason, a wide dynamic range is required in the calculation of the speech synthesis filter, that is, a long word length in the case of digital, which results in an increase in hardware.

(2) As mentioned above, in the case of a voiced sound, the power of the driving sound source signal is proportional to the energy of the impulse and inversely proportional to the pitch period.

Therefore, in order to set the power of the driving sound source signal to a certain value, the magnitude of the impulse must be given by considering +3 pitch periods.

The magnitude of the impulse (pulse height) is proportional to the square root of the product of the driving sound source power and the pitch period.

Therefore, in order to obtain a driving sound source signal with a constant power, voiced sounds are different from unvoiced sounds, and the impulse size is not uniform, and in the case of speech synthesis with variable pitch period, correction calculations are required. I needed it.

This invention solves the above-mentioned drawbacks, handles both voiced and unvoiced sounds in a unified manner, disperses the energy of the driving sound source signal without being concentrated at a specific point in time, and has a simple configuration. The object of the present invention is to provide a driving sound source signal generating device in which the efficiency of use of a speech synthesis filter is increased and the equivalent dynamic range is expanded.

According to the invention, a pseudorandom sequence generator is initialized with a given pitch period in the case of voiced speech.

The output of the pseudo-random sequence generator is supplied as a polarity code to a sign inversion circuit, and the excitation amplitude given thereto is outputted to obtain a driving excitation signal.

For example, as shown in FIG.
is provided, and this pitch periodic pulse generator 28 is connected to terminal 1.
A pulse is generated with a pitch period given through 4.

The pitch period pulse generator 28 is composed of, for example, a counter, and counts, for example, a 125 microsecond (8 KHz) operating clock signal from a terminal 29 to generate a pulse for each pitch period applied to the terminal 14.

Further, a pseudo-random sequence generator 31 is provided which outputs a pseudo-random value, ie, 0 or 1, in synchronization with the operating clock of the terminal 29.

A gate 32 is controlled by a signal from the terminal 17 that distinguishes between voiced and unvoiced sounds, and in the case of voiced sounds, the pitch period pulse from the pitch period pulse generator 28 is passed through the gate 32 to the pseudorandom sequence generator 31 as an initialization pulse. given, and for each pulse the pseudo random sequence generator 3
1's internal state is initialized to generate the same pseudo-random sequence each time.

Value of the output of the pseudorandom sequence generator 31 (0 or 1)
Therefore, the positive and negative signs are transferred to the terminal 19 in the sign inverting circuit 33.
The driving sound source signal is added to the sound source amplitude and outputted from the terminal 21 as a driving sound source signal.

Regarding pseudo-random sequences, there are many documents such as "Correlation Function and Spectrum" (edited by Takashi Isobe) (University of Tokyo Press).

Although there are several types of pseudorandom sequences, an example of a circuit using an M sequence is shown in FIG. 5 in this work.

In FIG. 5, parts corresponding to those in FIG. 4 are given the same reference numerals.

Here, as the pitch period pulse generator 24,
Uses a presetable down counter.

An M-sequence generation circuit is used as the pseudo-random sequence generator 31, and an exclusive OR (exclusive OR) of each tap output of the first stage and the fifteenth stage of the 15-stage shift register 34 is used.
sive OR) in the circuit 35 and shift register 3
It is configured by returning to 40 inputs.

In this case, the sign inverting circuit 33 uses a two's complement circuit, and does not take the complement of the output from the pseudo-random sequence generator 31, that is, when the sign is positive, but converts the amplitude given from the terminal 19 into the output. If the sign is negative, the two's complement of the given amplitude is taken and output.

The case where an M-sequence generation circuit is used as the pseudo-random sequence generator 31 will be explained in a little more detail.

The M series is binary (O and 1 or ±1) in synchronization with the clock.
The generation method is easy with a pseudo-random sequence that takes (etc.).

The generation method is said to be able to generate a binary pseudo-random sequence with a period of 2n-1 clocks by exclusive ORing the appropriate tap outputs of the n-stage shift register and feeding it back to the input of the shift register. However, which tap should be used is described in the above-mentioned document.

To give an example, if the tap output of the first stage is written as Di, it is as shown in the following table.

(■ represents exclusive OR.

) For example, an example of circuit implementation in the case of n=5 is shown in FIG.

That is, five mask slave flip-flops FF1 to F
F5 constitutes a shift register 34, and its second
The exclusive OR of the outputs D2 and D5 of the 5th stage and the 5th stage is the circuit 3.
5 and returned to the input.

Initialization input is provided to each flip-flop FF through terminal 36.
It is applied to the set input or reset input of FF1 to FF5, and in the case of this figure, when initialized, the flip-flop F
The states of F and FF5 are each 11001.

Therefore, the M sequence output after initialization will always be 110011010010...
The period is 31 clocks.

If the M sequence is regarded as a kind of signal by associating output 1 with +1 and O with -1, adjacent sample values have almost no correlation with each other, and the signal spectrum is flat.

In particular, if the order n is set sufficiently large, periodicity can be practically ignored.

If initialization is performed periodically, the output signal will have periodicity while maintaining similar characteristics.

It should be noted that since the initialization is performed by setting all the contents of the flip-flops FF1 to FF5 to O, it is possible to initialize to any pattern as long as the contents are set to O.

In the embodiment shown in FIG. 5 as well, there is no major difference in the pattern when the 15-stage shift register is initialized, except that all the patterns are cello.

As is clear from the above explanation, the driving sound source signal obtained from the device according to the present invention always has a given amplitude value as its absolute value, and its sign is pseudo-random, and this point applies to both voiced and unvoiced sounds. The same is true.

Therefore, if the amplitude value is constant, the instantaneous power of the driving sound source signal is constant in any case.

Therefore, the following effects can be obtained by using the apparatus of the present invention.

(1) Both voiced and unvoiced sounds can be handled uniformly, and there is no need to correct the amplitude in consideration of the period even in the case of voiced sounds.

(2) Since the energy of the driving sound source signal is not temporally concentrated, the dynamic range of the speech synthesis filter can be narrower than before.

This means that when using a digital filter, the operation word length can be shortened.

(3) In the conventional method of using impulse trains for voiced sounds, the driving sound source signal includes a DC component, which narrows the dynamic range of the speech synthesis filter, and the DC component also changes with changes in the applied amplitude value. This caused low-frequency noise, which degraded the quality of the synthesized sound.

The present invention does not have this drawback because it generates almost no DC component.

Therefore, the present invention brings about great effects when applied to a speech synthesizer.

Although the driving sound source signal is generated by digital processing in the above description, it may be generated by analog signal processing.

In that case, the sign inversion circuit 33 may switch the sound source amplitude signal supplied to the operational amplifier between an inversion side and a non-inversion side using a pulse from a pseudo-random pulse generator 31, for example.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing a general speech synthesis device, Fig. 2 is a block diagram showing a conventional driving sound source signal generation device, and Fig. 3 is a block diagram showing a conventional sound synthesis device.
A diagram showing an example of a speech synthesis filter in an ARCOR type speech synthesis device, FIG. 4 is a configuration diagram showing an embodiment of the driving sound source signal generation device according to the present invention, and FIG. 5 is a configuration showing a more specific example thereof. FIG. 6 is a logic circuit diagram showing an example of a fifth-order M-sequence generator. 14: Pitch period input terminal, 24: Pinch period pulse generator, 29: Operation clock input, 17: Voiced/unvoiced sound switching input terminal, 19: Sound source amplitude input terminal, 31: Pseudo-random sequence generator, 32: Gate , 33: Sign inversion circuit, 34: Shift register, 21: Drive sound source signal output terminal.

Claims (1)

[Claims] 1: a pseudo-random sequence generator whose initial state can be set; a pitch-period pulse generator that generates a pulse for each given pitch period; and a sign inverting circuit that uses the output of the pseudo-random sequence generator as a polarity code and outputs a drive excitation signal having a given amplitude value.