JPH0363699A - speech synthesizer - Google Patents

Abstract

PURPOSE:To compose a voice signal of a vowel close to a natural voice by providing at least two pairs of resonators and antiresonators for altering a vocal chord sound source waveform into a waveform having a polar zero couple. CONSTITUTION:A polynomial waveform generator 101, a random number generator 102, and an impulse waveform generator 103 are actuated with an operation signal corresponding to a voice to be synthesized and the level of the generated sound source signal is adjusted by corresponding level modulation parts 5 - 9. Then a specific frequency band is emphasized or attenuated by resonators and antiresonators of branch parts 11 - 14 and an addition part 15 outputs a voice signal after articulation processing and radiation characteristic addition processing. A vowel has a pole frequency and a polar-zero frequency and >=2 pairs of resonators and antiresonators which generate waveform having the polar-zero couple are provided, so a vowel waveform which is closer to the waveform of a natural sound than a vowel having no polar-zero couple is obtained, so a voice which is reproduced and outputted is close to the natural voice.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a speech synthesis device for synthesizing speech, and in particular,
The present invention relates to a speech synthesis device that can be used as a text synthesis device or a speech rule synthesis device that inputs sentences and phonetic symbols and outputs them as speech.

[Prior Art] As an example of a conventional speech synthesis device, D.H.Klatt
is 5of+yare for a cascade/
parallel formatsynthesize
r, The Journal of the Ac
oustical Society of Ameri
ca 87(3), Mar, 1980 pp 971
-995, the circuit configuration of the proposed speech synthesis device is shown in FIG.

In FIG. 3, when synthesizing a voiced sound, the impulse waveform generating section 21 creates a sound source waveform for creating the voiced sound, in this example, an impulse waveform signal.

This impulse waveform signal is given a frequency characteristic depending on the type of sound generated by the resonators 22, 23, and 24. Further, the intensity of the signal is variably set by the intensity control units 25 and 26. The analog signal simulating the sound source waveform of the voiced sound (vocal cord sound source waveform) created in this way is sent to the resonator 32.
．． 34 to 38 and the anti-resonance Ja 33, a radiation characteristic is given by the -order micro-arithmetic unit 55, and the resulting sound is output as an audio signal.

On the other hand, when synthesizing unvoiced sounds, the noise waveform generated by the random number generator 27 is transmitted to the intensity modulator 28. It is outputted through a low-pass filter 29, and this outputted analog signal becomes a sound source signal of unvoiced sound, that is, noise due to turbulent flow caused by human exhalation passing through a narrow portion of the vocal tract.

Next, the types of sounds that occur, such as band silence.

A signal processing system is selected depending on the fricative, etc., and the intensity of the analog signal is adjusted by the intensity control section 3o or 31, and further selected from the selected system according to the phoneme to be generated. After being tuned by the anti-resonator of the resonator 5, a radiation characteristic is imparted by the −th order differential calculation unit 55 and output.

In addition, when generating a plosive sound in each of a voiced sound and a voiceless sound, a step waveform signal that decays exponentially as shown in Fig. 4 is generated by an fg generator (not shown). The signal is added to the sound source signal as a signal indicating the plosive part of the plosive.

In this way, the audio signal created by the above-mentioned electric circuit by simulating the human voice generation process is output as audio from a speaker or the like.

[Problem to be Solved by the Invention] However, the conventional speech synthesis device has a problem to be solved in that the vowel parts of the generated speech do not approximate natural speech.

This point will be explained in detail. For the phoneme "na" naturally spoken by a human, a voice signal waveform as shown in FIG. 5 is obtained, and the frequency characteristic of the constant portion 501 of the consonant "n" is as shown in FIG. 6.

Further, as the frequency characteristic of the steady portion 502 of the vowel "a", the characteristic shown in FIG. 7 (8) is obtained. Furthermore, Figure 7 CB
), P1=318Hz in the frequency characteristic of the vowel “a”
, Zl-450Hz, P2=2250Hz, Z2
A pole-zero pair of a pole (PI, P2) and a zero (Zl, Z2) with a center frequency of =2310 Hz is obtained as an analysis result. However, conventional devices do not have resonators and/or antiresonators to achieve the frequency characteristics produced by these pole-zero pairs, and as a result, the synthesized vowel sounds differ from human natural sounds. It wasn't a good approximation.

Therefore, the purpose of the present invention is to eliminate such problems,
#J is a speech synthesizer that can synthesize vowel audio signals similar to natural speech! ! It is about providing.

[Means for Solving the Problems] In order to remotely achieve such an object, the present invention provides a speech synthesis device that synthesizes speech by emphasizing and/or attenuating specific frequency components of a vocal cord sound source waveform. The present invention is characterized in that it includes at least two sets of a resonator and an anti-resonator for transforming a vocal cord sound source waveform into a waveform having pole-zero pairs when a vowel is generated.

[Function] The present invention focuses on the fact that the frequency characteristics of vowels have pole-zero pairs, and provides at least two sets of resonators and anti-resonators that create waveforms having pole-zero pairs. Compared to a vowel waveform generated by the device that does not include a waveform of pole-zero pairs, a vowel waveform that is more similar to the waveform of a natural sound can be obtained, so the reproduced sound is also closer to a natural sound.

〔Example〕

Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 shows the basic configuration of an embodiment of the present invention.

In Fig. S1, 1000 is a sound source section, which includes a vocal cord sound source waveform generation section 1. Friction sound source waveform generation part 2. Plosive sound source generation section 3
Consisted of.

When generating a voiced sound, the vocal cord sound source waveform generating section 1 is activated by an operation signal and generates a signal having a vocal cord sound source waveform. In this example, a waveform whose relationship between signal strength and time can be expressed by a polynomial is used as the vocal cord sound source waveform.

The frictional sound source waveform generation unit 2 performs an operation f when generating a fricative sound.
g is activated and generates a signal having a frictional sound source waveform. In this example, a typical random waveform is used.

The plosive sound source waveform generating section 3 is activated by an operation signal when generating a plosive sound, and generates a signal having a plosive sound source waveform. In this example, a waveform having a step change is used.

Reference numeral 2000 is an articulation unit that modulates the sound source signal generated in the sound source unit 1000 according to the type of sound dish to be synthesized. The articulation unit 2000 includes a buzz bar/nasal branch unit 11 and its intensity control unit 5. It is mainly composed of the branching section 12 for silent/voiced sounds, its intensity control section 6.7, its intensity control section 8, and the rupture section branching section 14 and its intensity control section 9.

The buzzbar/nasal sound branching unit 11 inputs the signal of the vocal cord sound source waveform via the intensity control unit 5, and emphasizes or attenuates a specific frequency component of f3 of the vocal cord sound source waveform, thereby articulating the buzz bar/nasal sound.

Depending on the type of phoneme to be synthesized, the cord silent/voiced sound branching unit 12 manually inputs the signal No. 3 of the vocal cord sound source waveform when a voiced sound is generated, and the signal of the fricative sound source waveform when a voiced sound is generated, Articulates unvoiced and voiced sounds.

The fricative sound branching section 13 inputs the signal of the fricative sound source waveform and modulates the fricative sound.

The plosive part branching section 14 receives a signal of the plosive sound source waveform and modulates the plosive sound.

The intensity control units 5, 6, 7, 8.9 receive parameters specific to the phoneme to be synthesized from an external device, such as a microcomputer, and variably set the intensity of the manually generated sound source signal.

The modulator 4 amplitude-modulates the signal of the frictional sound source waveform in synchronization with the fundamental period of the vocal cord sound source waveform. Note that when the vocal cord sound source waveform signal is not output, that is, when the vocal cords are not vibrating, the modulation section 4 does not perform amplitude modulation. The signals output from the above-mentioned branching units 11, 12, 13, and 14 are added to the adding unit 1.
5 to the radiation characteristic section 3000. The radiation characteristic section 3000 outputs an audio signal imparted with radiation characteristics by emphasizing the high frequency of the human input signal.

FIG. 2 shows an example of a specific circuit configuration of the circuit shown in FIG. 1. In the nine circuits, each sound source signal generator is given a radiation characteristic.

In FIG. 2, numeral 101 is a polynomial waveform generator, which in this example generates a vocal cord sound source waveform signal with high-frequency emphasis of a fourth-order polynomial waveform. Note that it is also possible to use waveforms obtained using other calculation formulas or waveforms extracted by analysis from natural speech.

A random number generator 102 generates a random waveform using random numbers and outputs a signal of a friction sound source waveform with high frequency emphasis.

Reference numeral 103 denotes an impulse waveform generator that generates an impulse waveform, that is, a plosive sound source waveform signal with high frequency emphasis applied to the step waveform.

Buzz bar/nasal branch part 11 shown in the first diagram. Branching section 12 for silent and voiced sounds. The fricative sound branch 13 and the rupture part branch 14 each have a resonator that emphasizes a predetermined frequency portion of the signal and an anti-resonator that attenuates the predetermined frequency portion, which are connected in cascade. In particular, the branch unit 12 for silent and voiced sounds includes a resonator 12 for creating a waveform representing a vowel pole-zero pair.
6j28 and anti-resonators 127 and 129 are provided.

Next, to explain the operation of the circuit shown in FIG. 2, each waveform generating section 10 is
1-103 is activated, and the intensity of the sound source signal is adjusted by the corresponding intensity modulation section.

Thereafter, a specific frequency band is emphasized and attenuated by the resonator and anti-resonator of each branch, and the adder 1
5 outputs an audio signal that has been subjected to articulation processing and radiation characteristic addition processing.

The number of resonators shown in Figure 2 is a number suitable for expressing frequency components up to about 6 kHz.Resonance depends on the degree of simulation of natural speech in the synthesized signal, for example, whether it is finely or coarsely simulated. It is also possible to increase or decrease the number of resonators and antiresonators.

Also, if the range of frequency components of the voice you want to express is, for example, 5
When the frequency is narrow, such as up to kHz, the number of resonators and antiresonators can be reduced.

In addition to the working examples, the following examples are given.

1) In this example, an example is shown in which resonators and anti-resonators are fixedly connected in cascade at various branch parts.
It is also possible to freely replace the connection positions of the resonator and anti-resonator. Furthermore, the connection order of the intensity control section, the resonator, and the antiresonator can also be changed.

2) In this example, in addition to being realized by an analog circuit,
Waveforms can be expressed as digital signals, and digital operations can be expressed using a dedicated digital circuit, or can be realized by software using a microprocessor, digital signal processing processor, or the like.

In this embodiment, the C signal does not always flow through all of the branching sections 11-14. For example, the plosive branching section 14 does not generate synthesized sounds except for the first plosive part of the plosive. Not contributing. Therefore, when performing speech synthesis by digital processing, the processing of each branch section jl to 14 is made into a subroutine, and only the subroutine that performs calculation processing for each type of phoneme to be generated is selected, thereby reducing the overall speech synthesis time. can be shortened.

[Effects of the Invention] The present invention focuses on the fact that the frequency characteristics of vowels have pole-zero pairs, and provides at least two sets of resonators and anti-resonators that create waveforms having pole-zero pairs. Compared to a vowel waveform that does not include a pole-zero pair waveform generated by a conventional device, a vowel waveform that is closer to a natural sound waveform is obtained, so the reproduced sound is also closer to a natural sound.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the basic configuration of an embodiment of the present invention, FIG. 2 is a block diagram showing a specific circuit configuration of the embodiment of the present invention, and FIG. 3 is a block diagram showing the circuit configuration of a conventional example. Figure 4 is a waveform diagram showing the shape of the attenuation waveform used in the conventional example. Figures 5, 6, 7 (^), and 7 (B) are human diagrams for explaining the synthesized sound of the conventional example. FIG. 2 is a characteristic diagram showing frequency components of natural speech. DESCRIPTION OF SYMBOLS 1... Vocal cord sound source waveform generation part, 2... Frictional sound source waveform generation part, 3... Plosive sound source waveform generation part, 4... Modulation part, 5-9... Intensity control part, II... - Buzz bar/branch part for nasal sounds, 12-... Branch part for band silent/voiced sounds, 13-... Branch part for fricative sounds, 14... Branch part for plosive parts, 15... Addition part, 101... - Polynomial waveform generator, 102-... Random number generator, 103... Impulse waveform generator, 111-115, 121-126, 128, 131, 1
33,135°141.143,145.148
...Resonator, 116.127.129,1
32,134,142,144...Anti-resonator.

Claims (1)

[Scope of Claims] A speech synthesis device that synthesizes speech by emphasizing and/or attenuating specific frequency components of a vocal cord sound source waveform, which transforms the vocal cord sound source waveform into a waveform having pole-zero pairs when a vowel is generated. A speech synthesis device comprising at least two sets of a resonator and an anti-resonator for deformation.