JPS61112199A - Voice synthesization by carrier modulation - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] This invention was made possible by the applicant's discovery that just as light has a frequency-specific sound, it also has a frequency-specific sound, which is the original sound of a vowel. This method is essentially different from conventional methods that aim at waveform or spectrum reconstruction such as segment synthesis, linear prediction, and sine wave superposition, and can synthesize vowels with extremely natural sound quality. This is an innovative method.

The present invention can be implemented using either hardware or software.

The principle and examples will be explained. Listen to the output of a low-frequency oscillator (at about 300 Hz it sounds like a boom, at about 1 kHz it sounds like a buzz, and above 2 kHz it sounds like a beep. These are plosive sounds because they are modulated by a step function that rises when they reach the ear. So, if you make it have a sufficiently sharp rise, it becomes a vowel.
! It can be seen that when 100% modulation is applied to one wave, it sounds like a vowel. The modulated waveform at this time is as shown in FIG.

v (t) = (1+cos (2sft)) sin
(2πnft), but if n is chosen as a prime number of 3 or more, one corresponding to English vowels can be obtained. Considering resonance, it can be predicted that the sound will consist of frequencies other than the fundamental wave that have no harmonic relationship. Of course,
The frequency bands before and after these also exhibit the same sound, with no clear boundaries and transitions through intermediate vowels. Also, the fundamental wave is 20
At 0 Hz, the 2nd, 3rd, and 4th harmonics are u, o, and α, which are not necessarily prime multiples. In other words, the vowel is not a sound wave waveform but a frequency-specific sound. Just as we have the sense or concept of color for the frequency of light, we have the sense or concept of vowel for the frequency of sound waves. Then, the vowels can be made to correspond to the value of n as shown below, and it is also possible, although not easy, to make a compound that corresponds to the seven colors of light.

n=3 5 7 11 13 17 19 23uoa
Δaa! ei Red-Orange-Yellow-Green Blue-Indigo-Purple The present invention is based on the discovery of these new facts by the applicant, and is completely new and different from conventional methods.

The frequency of vowels, which are sounds specific to these frequencies, is too high when n is 11 or more, and the pitch frequency of vocal fold vibration indicates the loudness of the voice because the energy increases in proportion to the square of the vibration frequency and the square of the amplitude. Since the components are relatively weak, the sound becomes small, making it impractical.The present invention is a method for improving these vowels into sounds that are closer to human sounds, and will be explained below using examples.

The former can lower the apparent frequency heard by the ear by modulating it with a pitch-based sawtooth wave.

For the latter, it is necessary to reduce the amplitude, and for the ninth, the amplitude of the sawtooth wave must also be reduced, but since this is a decrease in the pitch component, 1 + a - s, which is the addition of the steady value to the vowel component.
in (2gnft) and Ova≦1 to prevent a decrease in the pitch component. Also, in order to avoid the generation of a second harmonic component, the sawtooth wave also has a steady value and has only positive amplitude. In the above, the vowel frequency was used as the carrier wave, but in this case, it is the same no matter which one is considered as the carrier wave.If you follow the idea that the carrier wave is the one that carries information, it is reasonable to consider the sawtooth wave as the carrier wave. The relationship at a certain 0 frequency is the opposite of that of radio waves, but whereas radio waves have extremely large energy in the carrier wave and the energy of the information itself is small, when the frequency of the carrier wave is lowered, the energy is small and the energy of the information itself is small. The larger the size, the better the efficiency. Living organisms are thought to use efficient methods, so sawtooth waves are considered carrier waves. The obtained vowel waveform is shown in Figure 2. One pitch Looms is composed of 400 bytes of data, and this is repeatedly output for the required time. This is significantly different from superposition, in that the amplitude of the modulated wave component also decreases as the amplitude of the carrier wave decreases, but this is because if the vowel is a harmonic of the fundamental frequency of vocal fold vibration, its amplitude is proportional to the displacement of the fundamental wave vibration. This is consistent with the assumption that he is deaf. These vowels are like the sound of a whistle, and cannot be said to be of good quality, especially when n=11 or more.Before discussing the improvement, we will explain in detail the reason for using a sawtooth wave as the carrier wave.

If the fundamental frequency of vocal fold vibration is 100 Hz, it is thought that harmonics up to about the 30th harmonic are included, but the sum total will be considered. Assuming that all of these harmonics have equal amplitudes and are sine waves with a phase of 0, their sum approaches the differentiation of impulses that repeat in the fundamental period. Figure 3 (a) shows the sum up to the 10th harmonic as This is the vowel a, which shows what is divided by. In this case, in order for the sum of 0 equal amplitudes, which is a mixture of vowels with the highest frequency and half the frequency, to become white noise, the phase must be random.If the 0 phase is π/2, it is an impulse train, and in the intermediate phase It becomes an intermediate waveform. It is thought that when light contains continuous frequencies, comrades with matching phases form an impulse train, giving it a particle nature.

Although it is possible to create such a vowel experimentally, it is impossible in terms of energy for all harmonics of vocal cord vibration to have the same amplitude. Therefore, if the energy is constant, that is, the amplitude of the nth harmonic is 1/n of the fundamental wave, and the sum of sine waves with phase 0 is determined, a sawtooth wave is obtained. This is a high-quality vowel, but when modulated with another frequency, it becomes a vowel of that frequency and simply works to lower the apparent frequency of the vowel. This is the reason why a sawtooth wave is used as a carrier wave. In fact, adding up to about the fifth harmonic is sufficient as a substitute for the sawtooth wave. This waveform is shown in Figure 3(b).If we consider that the vibration of the vocal cords generates a sawtooth wave and the resonance of the vocal tract enhances a specific frequency component, this is the vocal tract theory, but #! ! The tooth wave contains its components with uniform intensity, so they are superimposed, and the amplitude strength as shown in Figure 2 does not appear.Modulation must be included.The harmonic vibration of the vocal cords itself is the fundamental vibration. It is physically reasonable to think that it is proportional to the momentary displacement, and this is the modulation effect of the vocal cords. In this case, the sum of harmonics is a sawtooth wave with a rounded shape. For these reasons, one of the features of this method is that a sawtooth wave or a similar wave with a pitch period is used as a carrier wave and is synthesized by a resonant modulation effect with a vowel frequency. Since vocalization is caused only by exhalation, a steady value is added to the carrier wave to form a positive waveform. Further, depending on the phase characteristics of resonance, it is preferable that the phase of the resonance frequency component is delayed by π/2.

Next, a method for improving the sound quality of the synthesized vowels will be explained. The method is extremely simple: mix nasal sounds in the combinations and proportions shown in the table below. The resulting vowel pattern is shown in Figure 4. The zero percentage is determined as follows. If the energy of α is 1, the amplitude of the original vowel is determined by fa/f so that it is equal to the desired value. If α, o, and u are mixed in this ratio into the high-frequency vowels above , the energy will be doubled, so adjust by multiplying both by 0.7.

uo (5GNe in 1/ i Protovowel 2 1.4 1 .5..4..3 .2
．． 2 Mixing Δ=, 3 a+=, l −a=, 7
o=1 u=1.6u, what is mixed into o is the third harmonic, which is converted into energy of q, and approximately ill is added in opposite phase.

This applies distortion that crushes peaks to improve sound quality and raise frequencies. The mixing ratio does not have to be exact and can vary considerably depending on your preference, but the phase is important.

1-acos(2+cnft) +bcos(2gm
ft), the carrier wave is modulated by a mixed wave with n<m, a, b>0.

Also, if 0 is mixed into a vowel of n; 17 or higher, it becomes a vowel e, but if U is mixed in, it becomes a nine-ba vowel i, which also works to lower the frequency of these high-frequency vowels. As a result, the vowel i becomes U and e.
It is possible to display all the vowels arranged in a ring. This is the same as creating a circular color combination diagram by mixing blue with red, which has a long wavelength, and creates purple, which should have a shorter wavelength than blue, but considers it to be an intermediate color.

By repeatedly outputting each pattern shown in FIG. 4 for the necessary time, it is possible to produce extremely high-quality vowels of the necessary length. When pronouncing a word by combining these words, it is necessary to make the pause time between vowels sufficiently long to improve intelligibility; if the pause time is too short, it will sound like a sutra. arise. Even if you speak quickly, if you keep your speaking time short and your pauses long enough, your intelligibility will not be lost.

Vowels uttered by humans are equivalent to handwritten letters, and variations in pitch period and formant frequency are due to human nature. Although the vowels synthesized by N do not have these, the waveform and sound quality are extremely close to human speech.

As described above, the present invention is based on a completely original idea, and is significantly superior to conventional methods in terms of simplicity, practicality, and characteristics of synthesized speech.

[Brief explanation of drawings]

Figure 1 shows a waveform in which a vowel frequency sine wave is 100% modulated with a 100 Hz cosine wave. Figure 2 shows the waveform of a sawtooth wave modulated by a vowel frequency. Figure 3 shows the waveform of the sum of harmonics. Figure 4 shows the waveform of a vowel synthesized by this method. Original patentee Akira Wakabayashi Figure 1 (Δ) Figure 2 (8) Figure 3 (b) 4rII Procedural amendment October 31, 1985 2 Name of the invention Speech synthesis method using carrier wave modulation 3 Relationship with the case of the person making the correction Address of patent applicant Kanaka "Waken Hiratsukashita" Ikanchiro
21-17 Daikancho, Hiratsuka City, Kanagawa Prefecture 〒254 Location 0
463-23-5422 Waka Bayashi Aki 6 years old List of attached documents (1) Drawings
Contents of one amendment 1 The scope of claims is amended as follows. Just as light has a frequency-specific color, sound waves also have a frequency-specific sound, and this is a speech synthesis method that improves the sound quality by focusing on the fact that it is the original sound of a vowel. Or a sine wave or cosine wave with a 9M vowel frequency using a similar wave as a carrier wave, A2. The detailed description of the invention is amended below. b) Delete the following 44 characters from the 10th character on page 2, line 19 to the 9th character on page 3, line 1. Furthermore, if the fundamental wave is 200 Hz, the 2nd, 3rd, and 4th harmonics are U, O, and a, which are not necessarily prime multiples. ) Replace page 9, lines 1 to 17 with the following. By repeatedly outputting each pattern shown in FIG. 4 for the necessary time, it is possible to produce extremely high-quality vowels of the necessary length. When pronouncing a word by combining these words, it is necessary to make the pause between vowels sufficiently long to improve the clarity; if it is too short, it will sound like a sutra. arise. In order to prevent this and shorten the pause period, there must be a start mark and a stop mark for each vowel, which reduces the amplitude of the sawtooth carrier wave to 1. It should be increased or decreased so as to form an envelope of the first-order lag of a rectangular wave with the pause section between vowels as O.The start pitch at the beginning of the start mark must be distinguishable from the pause section and have a sufficiently small amplitude. The amplitude of each vowel in the above formula table is simply for determining the proportion.
The amplitude after blending is made constant and is used as the reference amplitude for each vowel. Next, we will discuss the synthesis of consonants. Conventionally, consonants are said to be composed of noise, but although noise is an element that causes individual differences, it is unavoidable due to the structure of the human vocal organ and has no relation to the essence of the consonant, so we have removed it. The essence of a consonant lies in the latter periodic component, and can be said to be a combination of vowels. Just as the visual concept of the frequency of light is color, and the mixing of various frequencies produces various colors, the auditory concept of the frequency of sound waves is the concept of vowels and consonants. Tazo. Speech is never a pure spectral sound, but is a mixture of at least pitch periodic waves, which corresponds to a mixed color, and the difference between vowels and consonants is that they are often a somewhat complex mixture of consonants. There are two ways to mix the six colors that exist: one is to mix light or paint directly, and the other is to mix the details of each color visually.
There are two ways. Mixing by amplitude addition or amplitude modulation corresponds to the former, and the eight vowels are synthesized by this. Although many consonants are synthesized by the latter. At this time, each frequency is arranged in time division on the time axis and mixed by hearing. Each consonant will be described in detail below. 1) Synthesis of row W In the combination of u and o, adding the third harmonic at a phase that crushes the peak of the main frequency means giving saturation distortion. Therefore, the saturation distortion is lower in O than in U, and α requires almost no distortion, which indicates that the opening of the mouth becomes larger in this order, and furthermore, this means that the wider the mouth is opened, the higher the resonance frequency becomes. Therefore, considering that Wa's vocalization falls from a closed mouth state to a wide open state, the main frequency must change continuously from the state of U to the state of α, in fact. This is observed, and this frequency transition can be realized by fairly rough frequency discretization, and 3 of uoα
By simply arranging the vowels in this order two or three pitches at a time, you can synthesize a fairly good wa. Actually, the number of pitches α is increased according to the length of the utterance, and the start and stop marks are formed as a whole. Furthermore, the amplitude of the front part where the two frequencies change should be smaller than the highest amplitude of the final vowel, and the number of pitches should also be smaller, which corresponds to the fact that the accent is on the vowel in the syllable. In order to distinguish these from the 0 phonetic symbol shown in Figure 5 (1) in the composite symbol formula, the 1-pitch waveform of the reference amplitude constituting the vowel is expressed in capital letters, with the number of repeated pitches on the right shoulder and the amplitude ratio on the lower right. This shows the composition method by arranging the symbols with , into connected classes. In this way, Japanese consonants undergo a frequency transition and fall to the final vowel at the beginning of utterance, forming what is called a syllable. The other sounds in the wa line are uoi, uou, uoe. The zero compound symbol expression uo is obtained by changing the final vowel of wa. It is not necessary to include the vowel O except for wa and wo, but the distinction from the 7th line will not be clear. 2) I has the same main frequency as the composite U of the Y row, and e has the same main frequency as O.
Since there is, a transition such as ieα can be considered, but this is negative. To clearly distinguish between tazo and wa, the vowel field is added to make it 1aaaα. Also, yu = iaaaou, yo = i e a! It is o. These composite symbol formulas are shown in Figure 5 (2). In the above-mentioned sequence of vowels compared to Munsell's color wheel, Wa is the one that moves from the low frequency side toward α, and vice versa. What goes towards α is yah, so a! By adding , this difference becomes clear, and the distinction between wa and ya becomes clear. 3) Synthesis of m row The synthesis of m is 200 Hz and 600 Hz in reverse phase (cos
In the case of a wave), a third-order saturation distortion of approximately 100% is added to the waveform, which is then modulated with a sawtooth wave. The sound of 200 Hz itself is U, and 300 Hz, which has the same U sound, is also possible, and 1 m is considered to be a U with increased third-order distortion by closing the mouth, so the vowel 1 Japanese M line is U in the wa line is m
It was replaced with . The number of pitches must be 5 to 6 pitches, and the 0 compound symbol formula, which reduces the number of 0s, is shown in Figure 5 (3
). 4) The composite n of the second row is the vowel a, which is obtained by adding 400 Hz to the same extent to 200 Hz to cause second-order distortion, and then modulating it with a sawtooth wave. The field is considered to be α plus the action of the tongue, and the action of the tongue gives second-order distortion, as estimated from the fact that it is near the second harmonic of α. However, as is the case with actual pronunciation, the distinction from m is not clear from this alone; when a vowel, etc. is added before or after n, the sound of the tongue landing on the upper jaw and the sound of leaving it are n.
By superimposing the pitches at the beginning and end of , it becomes clear to distinguish it from m. This is a mixed wave with a strong frequency component around 3 kHz, and it is superimposed on the first half of the pitch waveform of n.
The details of this mixed wave having a waveform as shown in the figure will be described later, but the composite symbol expression of n including these is shown in FIG. 5 (4). 5) The composite r of the A row is composed of one pitch in which the frequency component representing the sound of the tongue leaving forms a clean envelope as shown in FIG. Therefore, for the U line, each vowel in the A line can be directly connected to this, and the composite symbol formula for the U line is shown in Figure 5 (5). 6) If you remove O from the synthesized sound uoa of the synthesized w in the line 6 and add a shortening to the extent of adding one pause pitch, it becomes ha. The human vocal organ, which can only change its resonance frequency continuously, is able to make such a jump in frequency through rupture.If you close your mouth, increase the internal pressure, and exhale the accumulated air all at once, you will exhale in a lump. The mechanism of rupture is that after expiration occurs, there is a temporary cessation of exhalation, during which time the shape of the mouth passes through 0, and when exhalation returns to normal, it is in state A. The result is a syllable with two parts: a consonant and a vowel. The compound symbol for ``ha'' is shown in Figure 5 (6). This ``c'' line is uttered with the mouth closed, which was written in the old days as ``fa'', ``hui'', etc., and there is a pause between the previous sound. If the section is short, Ha → Wa, Hi →
A, F, etc. transitions occur. The consonant part of the modern C line is different, but I will discuss this later. By changing the consonant part of the C line, the following various consonants are produced. 7) Synthesis of the P line The B line is considered to be a stronger rupture of the C line, but a strong rupture means that the amount of jump in frequency with respect to time is large. Therefore, the P line can be synthesized by reducing the number of pitches of the prefix U of the C line and entering the rest pitch immediately after the start pitch. Figure 5 (
As shown in 7), it is best to use m, which indicates a state with the mouth closed, as the starting pitch; in actual speech, there is no pause between the previous sound and the above m is packed, and the pitch changes depending on the speed of speech. The number of pitches changes. 8) Synthesis of k, h, t, ah, ts, s A characteristic of these consonants is that they are composed of white noise. White noise is a type of noise, not noise.Spectrally, it can be seen as a mixture of sine waves with uniform amplitude, but there is no precedent for it being synthesized using this method, and it is simulated using random numbers, hence the name. . Degento was the first in the world to succeed in synthesizing this based on spectra using the method described below. y=ΣA 5in(2nπft + 2yc R)n

nR is the nth value of a uniform random number sequence between O and 1. That is, y is the sum of all harmonics whose phases are randomly defined. When A is constant, it becomes so-called white noise. This method is an innovative method that can synthesize noise variations with arbitrary spectral distributions. However, since 29 is a periodic function with the period of the fundamental wave, if this is not desirable, it is necessary to set the total length as the fundamental period. For each consonant in this item, the maximum value of n for a fundamental wave of 100 Hz is approximately the value shown in the table below. k h r/n t ch ts sn
= 17 23 31 47 61 71 83 The sound of tongue release and release in r and n also belongs to this category.It is better to emphasize the high frequency range of about 3 db10ctav@ rather than keeping the amplitude An constant.For example, the synthesis of k and S One pitch of the waveform is shown in Figure 8.Actual speech is produced by narrowing the exhalation passageway and emphasizing the high frequencies. The narrow interspecies wave number becomes high, and due to the pressure regulating effect due to the high internal pressure, the pitch period becomes less clear and the number of pitches increases. h, ch,,
ts and s correspond to this, and in order to blur the pitch period, it is best to synthesize by taking the consonant section head as the basic period.If you use the pitch period as the basis, the pitch period will clearly appear, so it will produce a sound like scraping metal with a grinder. occurs. These consonants gradually increase and after reaching the highest value 2-3
It is modulated with a waveform that attenuates with pitch, and at the same time the highest frequency also shifts toward the lower side. Therefore, t consists only of the above-mentioned attenuation part and consists of about 2 pitches, but ch includes t,
ts includes ch, S includes ts, is composed of 2 to 3 pitches, and the pitch period is relatively clear. By connecting each vowel in the A line to these consonants, you can synthesize the power, su, ri, and ha lines. 9) Synthesis of the B row It is best to use the pitch period sine wave and its second harmonic with the same phase and amplitude as the carrier wave, as shown in Fig. 9 (a), for the synthesis of the B row. si
The 2-pitch waveform obtained by modulating the n-wave is shown in the same figure (b).
It is extremely similar to the waveform of an actual speech bar, and with this alone, it is possible to synthesize bass using the above-mentioned method of synthesizing bass. Tazo, if you add one pitch of bu made in the same way at the beginning, the sound of ba will become stronger. If 700 Hz is modulated by the absolute value of the carrier wave used in N plus a steady value of about 25, then B is obtained. If this is used to synthesize the above-mentioned PA lines, a waveform that is closer to the actual waveform will be obtained. Also, if the increase in pitch intensity is made more gradual, this becomes a vowel A. Unlike the vowels described above, this is more similar to an actual speech waveform that vibrates above and below the zero level. This carrier wave can also be seen as a sawtooth wave with high frequency components removed. 10) Synthesis of gt dt zyng g=kBu, d
=tBu, z=sBu, ng=nkBu. However, the number of pitches of S in the synthesis of 2 must be reduced, and ts may be considered to be used. As mentioned above, Bu cannot be divided into a consonant part and a vowel part, and when the number of pitches increases, it can be heard as having a vowel U. 1. Japanese voiced sounds are added to these English voiced sounds or clear sounds with each part of the B line. Just keep the sound going. 11) Synthesizing the Yo sound To synthesize the Yo sound, just follow the Y line to each consonant part of the I stage according to the Roman alphabet, and the iea: is usually composed of one pitch each. 12) During the utterance section of n, the pitch amplitude suddenly drops to below 25, and this state is maintained thereafter. Therefore, when the next vowel comes, a phenomenon occurs in which the vowel changes to a sub line. Each of the conditions described above in the nine Japanese vowel consonant synthesis examples is standard, and the permissible range is extremely wide. This makes it difficult to identify these standard conditions through the mathematical analyses. Therefore, while conventional speech synthesis only reproduces the voice of one person as a model, the one-line synthesis method can synthesize voices that are unrelated to the individual person, and the former reproduces handwritten characters. For example, this method is equivalent to the generation of printed characters. Of course, the tone can be changed in various ways by shifting two frequencies or adding noise, so it is also possible to simulate a specific human voice. For this purpose, it is also possible to perform highly compressed segment synthesis, which divides speech into mother and child parts in the Roman alphabet based on this method, and then combines and reproduces the subdivided parts as much as possible. Furthermore, the two-line synthesis method is also effective as a speech recognition method, as is clear from the examples. Note that the two-line synthesis method can synthesize English speech based on its spelling. As described above, the present invention is based on the completely original idea that speech is a sound pressure and time division combination of sounds with individual frequencies, and is superior to conventional methods in its simplicity and practicality, as well as in the characteristics of synthesized speech. It is significantly better than that. 3 Add the 9th series of explanations from Figure 5 below to the brief explanation of the drawings. Figure 5 is a symbolic formula showing how to synthesize each consonant. Figure 6 shows a waveform in which the sound of the tongue touching and leaving the upper jaw is superimposed on the n pitch waveform. Figure 7 shows the waveform of r. Figure 8 shows part of the waveforms of k and S. FIG. 9(a) shows the waveform of the carrier wave in the PA row. (b) is a waveform modulated at 700 Hz.

Claims (1)

[Claims] This is a speech synthesis method that focuses on the fact that just as light has a frequency-specific color, so sound waves also have a frequency-specific sound, and that this is the original sound of a vowel, and improves the sound quality of the sound. A speech synthesis method in which a sawtooth wave or similar wave with a pitch period is used as a carrier wave and modulated with a sine wave or cosine wave of the original vowel frequency, or a mixture or modulation wave of two or three.